This Week in Tech Episode 1102 Transcript

Please be advised that this transcript is AI-generated and may not be word-for-word. Time codes refer to the approximate times in the ad-free version of the show.
 

Leo Laporte [00:00:00]:
It's time for TWiT This Week in Tech. Jason Hyner is here, Dan Patterson, Owen Thomas. There is a lot to talk about. Google with another, I'd say, win in court, although it might be the best we could expect. World War III almost breaks out due to an AI hallucination. Oh my goodness. And stand back, here comes the AI disinformation and a whole lot more this week. On TWiT, stay tuned.

Owen Thomas [00:00:31]:
Podcasts you love.

Leo Laporte [00:00:32]:
From people you trust. This is TWiT. This is TWiT, This Week in Tech, episode 1102, recorded Sunday, September 20th, 2026. Exclave subpoena. It's time for TWiT This Week in Tech, the show where we cover the week's tech news. And well, this was kind of a slow week, so it's a good thing we have interesting people like Jason Heiner, who is the editor-in-chief of The Deep View, all about AI. Jason, you get to go to all the conferences. I'm so jealous.

Jason Hiner [00:01:11]:
Yeah, uh, I mean, it's, it's fun, it's exhausting, it's, you know, all the things.

Leo Laporte [00:01:17]:
Yeah, but, but you're covering, you're covering the number one topic of the day, AI.

Jason Hiner [00:01:23]:
Yeah, I mean, I knew last year I wasn't keeping up, and so like, I was like, you know what, I really need a job where I can just think about this, write about it, talk about it, learn about it, read about it all day long. And you know, be careful what you wish for.

Leo Laporte [00:01:35]:
Yeah, you got it.

Jason Hiner [00:01:36]:
Because that's where I am, and that's, that's my life now.

Leo Laporte [00:01:39]:
I love it. Well, it's, it's, you know what, I'm glad to have somebody with your, uh, skill and, uh, and frankly context, you know, your, the depth of knowledge to cover it, because there's a lot of people out there Making a lot of silly noises because they don't have any context or understanding. And then we go to the heart of the AI revolution where the doomers live. Owen Thomas is here from the San Francisco Business Times.

Owen Thomas [00:02:06]:
We got the doomers, the boomers, the zoomers, the loomers, the 3 engineers, and the rumors. We got them all.

Leo Laporte [00:02:12]:
You got them all. The home of effective altruism, San Francisco, California. Owen's the managing editor of the San Francisco Business Times, so you're really at ground zero for this.

Owen Thomas [00:02:21]:
Hey, if there's going to be a polyamorous sex cult running the future of, uh, of the world, it's going to be in San Francisco.

Leo Laporte [00:02:28]:
If you're just tuning in, kids, that is the latest on— well, we'll get to that because I want you to explain this whole thing to me. But first, let's say hi to Dan Patterson. He's also in the house. He is senior director of content at Blackbird AI, using AI To prevent, block, and inform when disinformation strikes. And that's a very important thing.

Dan Patterson [00:02:50]:
Yeah, all day, every day. And we are kind of inside the beast, although totally different. You know, we hear about the frontier models all the time. We don't hear so much about every other company that is either integrating or using their own AI systems. And that's exactly what we do at Blackbird.

Leo Laporte [00:03:09]:
Yeah. Yeah, unfortunately, and I'm going to call them by their real name, the AI frontier, AI cartel dominates everything. But really, there's a lot of interesting stuff. I saw on Monday Night Football Thomson Reuters advertising a custom model they've built. They started with Quen, the Chinese Open Weight Model, that had all of the, you know, Westlaw and all the Reuters stuff all the data that that company has in a model to be used by lawyers and financial people and stuff. That's, I think, in many ways really the future of AI. That is the future.

Dan Patterson [00:03:47]:
Yeah.

Leo Laporte [00:03:47]:
Yeah.

Jason Hiner [00:03:48]:
Absolutely.

Leo Laporte [00:03:48]:
You've been saying that for a while, Jason, that it's the small models, SLMs, that are really—

Jason Hiner [00:03:54]:
I don't want to cut in on Dan because Dan has a great perspective on this too.

Dan Patterson [00:03:58]:
Well, I think it's exactly that. It's the small models and it's the Uh, it's just the stuff that is innovative but we don't hear about. It's less sexy, but it is probably more interesting, especially to an enterprise audience and to a business audience, uh, that, that is, you know, maybe not AI hacking into everything, but, uh, and, and not hyped, but adding efficiencies, uh, incrementally day by day.

Leo Laporte [00:04:28]:
Well, uh, yeah, it's not sexy. It's not going to go— Anderson Cooper is not going to interview the guy behind JEV and say, well, let's talk about classifiers, because it's just not sexy. Much more sexy, let's talk about how the world is going to end in the next few years.

Dan Patterson [00:04:43]:
I have so many thoughts about those stories.

Leo Laporte [00:04:45]:
I know, I'm sure you all do. You know, I don't— I know people are tuning in going, not again. So we're gonna, we're gonna put a pin in that. We will get to that conversation, but I think probably we should start with a couple of big breaking stories. One, The judge has finally unsealed, uh, on Wednesday, Judge Leonie M. Brinkema, the U.S. District Court for the Eastern District of Virginia, who last year found that Google was a monopoly and broke the law in its advertising business to control the technology. She said it was a monopoly last year, but last week said, but I'm not going to break them up.

Leo Laporte [00:05:25]:
even though the Department of Justice is begging me to break them up. She said, I'm not gonna break them up, but I am gonna seal the order for a week until everybody, both parties, have a chance to look at it. Well, it was unsealed, and we now know— I think that— I don't know, I'm curious what you think. My guess is the judge did the right thing. Breaking up Google seems like a— first of all, not necessarily even a solution. You're just gonna create another ad monster with some other company.

Owen Thomas [00:05:52]:
But also, yeah, there's history here, right? Microsoft was going to get broken up in its antitrust trial. And, and that, you know, that ended up being a, you know, a settlement, basically, you know, kind of similar to this monitoring with conditions. I think, you know, I think the judge in this case probably looked at that history and said, you know, by the time If I were to impose a breakup order, Google was going to fight it tooth and nail. And by the time anything happens, the industry is going to look completely different anyway.

Leo Laporte [00:06:29]:
Good point.

Dan Patterson [00:06:30]:
Yeah.

Leo Laporte [00:06:31]:
That was kind of the conclusion of the judge last year who ruled Google a search monopoly, but said, but I can't force them to sell their search business because it's already— it's too late. It's already, you know, AI is coming for it anyway.

Dan Patterson [00:06:46]:
This really just highlights the narrative, the theme that we've talked about here on this show, and I'm sure all of us have in our own lives and work environments where not just AI, but tech broadly is moving so much faster than government can regulate, keep up with, and really understand. That gap has gone from being significant to now almost insurmountably huge. The gap and the speed.

Leo Laporte [00:07:14]:
So the judge said you don't have to sell your ad exchange, but you have to open it up, which from Google's point of view might be even worse. So there's going to be a court-appointed technical monitor with full access. This is exactly what the Microsoft consent decree allowed too, by the way. Full access to Google staff, systems, and source code. Both sides had proposed this monitor and technical committee, And websites using Google's ad server will no longer have to use its ad exchange. This was such a blatant, obviously monopolistic move to say, I'm gonna control both the sales and the buying.

Owen Thomas [00:07:51]:
Yeah.

Leo Laporte [00:07:52]:
I'm gonna control the whole thing, and good luck to all of you. So the AdX must bid into rival publisher ad servers on the same terms it offers to Google's DFP ad server. Uh, and also Google has to connect ADX. So ADX is the buyer, DFP— it's so confusing— but they have to connect it to PreBid, which is an open-source bidding framework publishers adopted. So it's really— this is actually, I think it strikes me, not knowing a huge amount about this, as a, as a very good decision. Like this this is better than forcing Google to sell the business, then somebody else buys it and they become dominant. This way others can participate. It's an open market, which I think seems like a—

Jason Hiner [00:08:43]:
It's more pragmatic for sure.

Leo Laporte [00:08:45]:
Pragmatic is a good word. Yeah.

Jason Hiner [00:08:46]:
At the level of like, you know, trust and principle, the thing is that, you know, the US is so business-friendly, so, you know, corporate-friendly, and then you have the EU at the other end of the spectrum, which is so consumer, you know, friendly, right? So if you're in the US and something rises to the level of the Justice Department actually taking notice—

Leo Laporte [00:09:11]:
By then—

Jason Hiner [00:09:11]:
Actually trying to intervene, like, it has to be pretty egregious, right? Because most of the things, they're gonna wave them off and say, right, like, if it— if it's— if it's on the margins at all of something that has— is going to cause any harms in the US, it'll get waved off. In Europe, it'll get— it'll attention. It might even get, you know, a fine or some kind of action to intervene. But in the US, if something rises to this level, it just has to be so egregious. And so when I think of this, that's what I think of. I think of like this, this just got to the point where it was so egregious that they're like, oh, this is actually causing damage to consumers, which causes damage to sort of larger— has the possibility of causing damage to the larger economy and jobs and businesses and that. And so otherwise it just would never even have gotten this far. And then to like your— to your point, Leo, like for the fact that they're trying to be a little bit pragmatic, it is in one sense a good sign, right? Knowing that, you know, to Dan's point, if we move— if we move on this at this pace we usually move at, you know, at sort of government and regulatory pace, we're probably going to have very little impact because government— because government just can't move as fast as business and even less so in sort of the very AI-empowered businesses that we have, you know, now that can just move at a different Well, Judge Brinkema also said she complained about—

Leo Laporte [00:10:38]:
she says she has a lack of trust, this is a quote, lack of trust that Google will comply with an order from this court. So this is another thing, we're in a situation where the courts sometimes, they don't have really much of an enforcement arm. They're just— the courts are going, well, we could say break up, but we might not be able to get it. The EU wants them to break up. The Department of Justice wanted a breakup.

Owen Thomas [00:10:59]:
Right.

Leo Laporte [00:10:59]:
But the judge said, yeah. Um, The Next Web quotes Barry Lynn of the Open Markets Institute, who said the decision was, quote, more proof that the US judiciary is abdicating its congressionally mandated duty to apply antitrust laws. I'm not sure I'd go that far.

Jason Hiner [00:11:16]:
I'm not sure what that meant, but I think I agree. No, I, I think that they're just saying like they're gonna, they're gonna plow, they're gonna do whatever they want anyway, and they have powerful lawyers and And so we're going to try to, you know, do something that'll be more effective.

Leo Laporte [00:11:30]:
Yeah, maybe. Let's hope.

Dan Patterson [00:11:32]:
Yeah.

Owen Thomas [00:11:32]:
I mean, is it, is it so pragmatic that it is kind of losing the plot? I think that's, that's the issue that it's raised. Like, can we, are these companies too big to regulate? That is, that is a harrowing thought.

Leo Laporte [00:11:47]:
Well, especially because what it then does is it gives incentive to companies to grow as fast as they can, as big as they can, as dominant as they can. Look what's happening in the AI market so that they can't be broken up so that they have a Yeah, they have a barrier, a moat against, uh, anything the government might try to do against them.

Owen Thomas [00:12:05]:
Yeah, it's like you've got to regulate them in the narrow window between like when they're, when they're too small to, you know, uh, to be regulated under antitrust to, you know, overnight they become so big that, you know, that they're dominant.

Leo Laporte [00:12:19]:
It's like that window is Game of Thrones. You want to get the dragon as it comes out of the egg because once it gets too big, forget it, you're out of it. Yeah, yeah. I'm sorry, go ahead, Dan.

Dan Patterson [00:12:28]:
No, that's a perfect metaphor. That dragon egg is, when it's emerging, it's happening faster and faster.

Leo Laporte [00:12:38]:
Get it before it can burn you to death. Google's ad tech business, worth $30 billion, 8% of Alphabet's revenue. Although, ad tech revenue has declined. I did not realize this for the last 4 years.

Owen Thomas [00:12:52]:
I mean, that's stunning that it's so, it's so big and yet so small. It's like not even a significant part of, of the business at this point. And, you know, was once seen as kind of, you know, the, you know, the golden egg of, you know, golden goose of, uh, of Google's business.

Leo Laporte [00:13:11]:
Just as with the search, uh, uh, order by the court a couple of months ago, I feel like this is a victory for Google. Google basically got what it proposed, which is, we'll open it up, we'll have a monitor, you can keep an eye on us. That's exactly what Microsoft proposed in its consent decree. And I think you could argue that the Microsoft consent decree worked better than breaking up Microsoft would have. I would argue that. I've heard people argue that, that in fact Google exists because of Microsoft's consent decree. Microsoft saying, well, all right, all right. We're not gonna force everybody to use Internet Explorer.

Owen Thomas [00:13:51]:
And to that point, OpenAI and Anthropic don't seem to have been slowed down by, you know, meaningfully by Google's dominance in, say, ad tech.

Leo Laporte [00:14:00]:
This is the history of technology, isn't it? It moves so fast.

Owen Thomas [00:14:05]:
We'll have to see how OpenAI does growing its ad revenue, because if Google sees that as a threat and, you know, and cracks down, I think that, that will be very telling. But ChatGPT ads are pretty nascent.

Leo Laporte [00:14:21]:
Yeah, there's other ways to make money. We'll talk about Meta's Muse agent in, in a little bit. I think— I think this is why it's so interesting. They're zigging, they're zagging. It's hard to pin them down. They're finding new forms of revenue. They're finding new ways to destroy the incumbents. I wouldn't want to be Google in this situation.

Leo Laporte [00:14:43]:
Both Google and Apple have a little bit of a disadvantage because they're old-line incumbents. They, they not only can they not move as fast, they don't want to cannibalize their business. It's the innovator's dilemma, isn't it?

Dan Patterson [00:14:53]:
Yeah.

Leo Laporte [00:14:54]:
Google should be winning in AI. I mean, they invented TPUs, they invented the whole thing.

Jason Hiner [00:15:01]:
And transformers and all.

Leo Laporte [00:15:03]:
Yeah.

Jason Hiner [00:15:03]:
And I wouldn't—

Leo Laporte [00:15:04]:
all the scientists.

Jason Hiner [00:15:05]:
I wouldn't count them out for the— for, for 2 reasons. They are way behind. And this has not been— '26 has not been— 2026 has not been a good year for them.

Owen Thomas [00:15:15]:
Right.

Jason Hiner [00:15:16]:
Like they have just fallen further and further behind, which is why they had the big shakeup, you know, a couple of months ago.

Leo Laporte [00:15:23]:
Right.

Jason Hiner [00:15:23]:
Well, you know, but I wouldn't count them out yet. Just sort of the last thing was just that they have 2 things that are to keep in mind. They have a brand that consumers know. Right. And still and trust largely. the ones who don't know about ad tech. Um, then they have also this— because they are a hyperscaler, they have a massive amount of compute, um, that Anthropic and OpenAI are still struggling to line up. So I'd say don't count them out for those reasons.

Leo Laporte [00:15:54]:
And then in other court news, this one, uh, just broke actually, uh, yesterday. A U.S. judge has signaled that he's going to reject the TikTok privacy settlement. Um, TikTok and ByteDance proposed a $400 million settlement with the Justice Department. On Friday, the federal judge said, eh, not so fast. Judge Wu in Los Angeles said he was inclined to reject the request to end the consent decree and scheduled a hearing for Monday. I think TikTok thought it had skated by pretending to sell part of it to Americans. companies and storing the data in Oracle servers.

Leo Laporte [00:16:36]:
The issue goes back to, believe it or not, 2019, when the FTC sued saying that Musical.ly— remember this? That was the lip-syncing app that TikTok bought— that Musical.ly knew young children were using the app but failed to obtain parental consent before collecting names and other PII. Musically paid a fine to settle the allegations. And then of course, TikTok thing was thrown into all sorts of turmoil with the Congress banning it at first, and the president, as soon as he got in office, unbanning it, forcing the sale to an American company— sort of sale.

Owen Thomas [00:17:19]:
I don't know.

Leo Laporte [00:17:19]:
So that's another case where the courts maybe aren't doing the bidding of the Justice Department. I don't even care anymore about TikTok. It's like it does.

Owen Thomas [00:17:32]:
It does seem to have faded a little, you know?

Leo Laporte [00:17:35]:
Well, it's still out there. It's still— I mean, Instagram has really dominated it, I think, but it's still doing its thing. It's still got—

Owen Thomas [00:17:41]:
Yeah. I mean, I think, you know, I think the number of times that you see like the TikTok logo on, on an Instagram reel You know, tells you something, doesn't it? Yeah, it's people, people just posting stuff everywhere now.

Leo Laporte [00:17:56]:
Yeah.

Owen Thomas [00:17:56]:
And TikTok just isn't, isn't special anymore in kind of having that feed.

Jason Hiner [00:18:02]:
Um, Meta gets them all in the end.

Leo Laporte [00:18:05]:
In the end, it's all Meta all the time. It's pretty amazing. Uh, a little— let's talk about scary AI news. Uh, an AI—

Owen Thomas [00:18:14]:
As opposed to any other kind?

Leo Laporte [00:18:16]:
Yeah, there's good news.

Owen Thomas [00:18:18]:
Okay. Oh.

Leo Laporte [00:18:19]:
It cured cancer yesterday. Didn't you hear that? Oh, I— maybe I missed it. Oh, that's what my AI told me. Never mind. AI hallucination almost caused World War III, apparently. But is it the AI or is it the foolish military who were about to attack a Chinese vessel? The US narrowly avoided boarding a Chinese ship based on an entirely false US intelligence report. According to CNN, generated with AI tools. The erroneous intelligence was submitted by a U.S.

Leo Laporte [00:18:55]:
Special Operations Command analyst saying the Chinese ship was transporting nuclear arms components through the Middle East, and our military prepared to intercept and board the ship with air support before— and this is, this is a little scary— before officials discovered a chatbot used in generating the report Had inaccurately identified the material the ship was carrying. I guess you could say this is a success story because it didn't create World War III.

Owen Thomas [00:19:27]:
I mean, what happens when Hexad starts feeding testosterone data into, uh, into the chatbot? If you, if you haven't caught up, the Department of, the Department of War, um, is now wants to monitor troops' testosterone levels.

Leo Laporte [00:19:43]:
Yeah. Does, does, uh, AI use testosterone? I don't think so.

Owen Thomas [00:19:49]:
Does it— does AI use drained testosterone? Possibly.

Leo Laporte [00:19:53]:
And, and if that story doesn't worry you a little bit, you should probably know that for the first time, the military in the US has confirmed it has deployed weapons in orbit. For a long time, we deemed space to be, uh, Demilitarized zone. I think we even had laws regarding that, but the United States now has on-orbit space control weapons capable of defending the joint force. Lasers from space.

Owen Thomas [00:20:24]:
Um, that's frickin' lasers?

Leo Laporte [00:20:27]:
Frickin' lasers! Actually, I don't know what the weapons would be. I mean, it could just be big rocks. I don't know what they're gonna— I don't know what they are. Um, but China apparently has also, uh, our intelligence believes, has done the same. U.S. government assessment in 2020 said that Russia was considering putting a nuclear weapon in space.

Owen Thomas [00:20:52]:
I mean, the, you know, the, the biggest thing for me is, is just why would you say it out loud? Like, that's, you know, it's probably been happening, but—

Dan Patterson [00:21:02]:
Yeah, I think they're saying this out loud because it's been happening for at least the 5 or 6 years. Um, And I mean, geopolitically, we could be saying this for a number of reasons, but likely because, um, the race is already on. For years we did not deploy weapons in space, but this has been happening at least since the end of the pandemic.

Leo Laporte [00:21:28]:
This Air Force Secretary says, is there any question in the room by anybody about whether or not the Chinese or Russians have been developing these sort of weapons for over a decade. We've been talking about the fact, and we have expressed concern, we need to make sure that we protect ourselves operating in that environment. But these are not— is he saying they're defensive? I don't think he is. Oh well, I don't want to give you only depressing news. But I should mention that OpenAI has disclosed 6 new incidents of concerning behavior. This— okay, we should probably talk about this. This, uh, we've talked about it for a couple of weeks now, the whole discourse about, you know, Jacob Cox quitting Anthropic, saying, I can't work at a company where AI doom is just around the corner. Um, you, you kind of referred to this, Owen, jokingly, but a number of people pointed out that many of the employees at these companies are almost— some people say it's a cult.

Leo Laporte [00:22:36]:
They're effective altruists. They're part of what Emil Torres calls the TESCREAL movement, where they— the gist of it is that at first they were like effective altruists are buying mosquito nets Because that was a very efficient way to spend money to prevent malaria. And then they realized, well, all you're doing is saving people who are alive currently. But if you look ahead thousands of generations, there's many, many more millions of people. You shouldn't worry about people alive today at the expense of people alive in the distant future. And effective altruism has mutated to the idea that we should do whatever we can today to protect generations way down the road, including working as fast as we can to create artificial general intelligence. Because that might hurt us, you know, with data centers everywhere and pollution and everything, but in the long run, it's good for us, for a spacefaring—

Dan Patterson [00:23:45]:
Yeah, future.

Leo Laporte [00:23:46]:
Civilization.

Dan Patterson [00:23:47]:
And all of this without evidence.

Leo Laporte [00:23:50]:
Oh, it's nutty. It's dangerous. It's cult-like, because that belief includes eugenics, includes all sorts of things.

Owen Thomas [00:24:01]:
Yeah.

Leo Laporte [00:24:01]:
Are not— any sane person would say, no, no, no, you really should probably think about people who are alive right now.

Dan Patterson [00:24:08]:
I think years ago, when I was working for Jason, I interviewed Nick Bostrom, who is the if not the originator of many of these ideas, certainly one of the accelerants. He's popularized in his book Superintelligence a lot of these ideas that the effective altruists use. And he was tied years ago to a number of eugenics movements, and he made pretty controversial statements, offensive statements in eugenics forums. You know, I I made the reference to without evidence to these hack stories. I'm sure you've mentioned this on shows prior, but the, air quotes, concerning behavior, all of these allegations have simply been that. We have not actually seen evidence. I've heard these stories repeated ad nauseam by pretty credible individuals in the media. But we have simply not seen evidence.

Dan Patterson [00:25:11]:
We, we don't have data to back this up. We just have these claims.

Leo Laporte [00:25:16]:
It's a belief. It's a belief system. And, um, yeah, so, but it's important to understand that, uh, at least some percentage, maybe a large percentage of people working at these frontier companies are in fact believers.

Dan Patterson [00:25:29]:
True believers. Yeah.

Leo Laporte [00:25:30]:
And so then it kind of makes a little more sense when— because it's very confusing that they say, well, yeah, AI could end the human race, but we want to work as fast as we can to build it, make it better and better and better. It's like, but wait a minute, aren't those conflicting points of view? Well, they aren't if you believe that the human race as it exists today is not as important as the human race 50,000 years from now.

Dan Patterson [00:25:54]:
Yeah, so it's an effective pessimist.

Jason Hiner [00:25:57]:
Yeah.

Owen Thomas [00:25:57]:
So, I should explain the sex cult comment.

Leo Laporte [00:26:01]:
Yeah, what's the deal on that?

Owen Thomas [00:26:02]:
So, there is No actual sex cult, but Marc Andreessen reportedly made that claim in a White House meeting. Unfortunately, because you'd like there to be a sex cult.

Dan Patterson [00:26:18]:
It'd be way more interesting.

Owen Thomas [00:26:22]:
Yeah. So Marc Andreessen, the venture capitalist co-founder of Netscape, made this— reportedly made this comment in a meeting in the White House, basically as a dig at Anthropic. Anthropic's co-founders, brother and sister Dario and Daniela Émile O'Day, lived in a group house in San Francisco together in the 2010s, along with Daniela's then-fiancé, now husband, Holden Karnofsky, who's very involved in the effective altruism movement. And they, along with OpenAI co-founder Greg Brockman, had some really momentous discussions about exactly these issues, about how do you frame the you know, the AI doom argument and, and get ahead of it. So, you know, that's, that's quite a leap to go from a group house where 2 people were in a committed relationship to it's a polyamorous sex cult. But, you know, this is the information environment we live in today.

Leo Laporte [00:27:23]:
The information environment is a mess. And yeah, it seems like, seems to me that our job is to try to make some sense of all of this. Yeah, I know we don't, you know, we don't have Anderson Cooper's reach. But I'm hoping that we can calm some people down. So what do y'all think about AI doom? Jason, is it the end of the world?

Jason Hiner [00:27:46]:
I think that, so what I wrote this week was a piece called, I put it in the doc, but it's called Why AI Pacing Has Become a False Choice. And so what I said there is that the problem with the current dialogue around this is the same problem of why we can't solve sort of any of the problems that we have— immigration, healthcare, all of the big problems. Why? Because the debate gets lost in sort of this black and white dichotomous thinking, like we can only have— we can either have safety, or we can have, you know, unlimited abundance and AI curing all diseases and, you know, bringing the workweek down to 20 hours or whatever the case is. So in fact, those are just very distant ends of the extremes of a whole giant spectrum of outcomes, and they aren't mutually exclusive, right? Like, AI is an extremely powerful tool. And so with an extremely powerful tool, As with all of the technologies and tools that we've had in the past of humanity, you can use it for a lot of really good things, and you can also do it— use it for a lot of really terrible things. And so, you know, are the pacing thing getting trapped in this, in this same doom loop of, of, you know, fractured dichotomous thinking and where we have people then what happens? Of course, what always happens, people take up positions at one end and then they, you know, dig in and they fight to the death. Right. of like why they're right and everybody else is wrong.

Leo Laporte [00:29:30]:
Right.

Jason Hiner [00:29:31]:
And that's ultimately the problem, right? The— is that this— the AI debate around safety and pacing and however one— governance, getting trapped in that is the problem. And what we need is— and this is kind of to your point, you know, just a minute ago, Leo, which is just what we need is better dialogue. What we need is more nuance. What we need is listening to each other and trying to learn from each other and not you know, going out and saying, you know, that this person is either on my team or isn't on my team, is either on team pacing or team innovation. You know, that's that itself. If we can't get past that, like, it will be really dangerous, right? Because there, there are parts of this that, that, that need some guidelines and some, some safety, some governance. You know, the governance has to scale with the the innovation. And right now we spend, you know, an order of magnitude, 2,000 to 3,000 times the amount on innovation than we spend on governance.

Jason Hiner [00:30:34]:
And so the outcomes are pretty, you know, clear on that.

Leo Laporte [00:30:37]:
Wow. That's pretty disproportionate. Holy cow.

Jason Hiner [00:30:41]:
So I have a podcast coming out on that this Wednesday with the CEO of Credo AI, who's a former engineer that It's worked with Qualcomm and Microsoft, and well as lots of nonprofits and agencies. And now Credo is essentially a governance. They do forward deployed, you know, analysts that that help people in regulated industries, in other, you know, in enterprises that that may match sort of the safety to the you know the outcomes and the risk that they have. So super helpful, and obviously knows and is connected with all of the. folks in Washington and governments and other places as well. So that's what we need. And I think to your point, Leo, just to wrap it up, is just that, you know, doing things like this where we're unpacking these things, where we're really, you know, exploring it, where we're not trying to dig in and take one position or another, like, that's more of what the ecosystem needs, right? And the fact that more people are talking about it, to be clear, despite that there is a lot of people taking up positions here and there, the fact that more people are talking about this, this last couple of weeks, I think is a good thing. Like the fact that it's broken out of just tech industry, you know, pundit conversations and into broader conversations.

Jason Hiner [00:32:01]:
I've had a number of people, you know, that are not connected with tech ask me about this this week. And I found that to be like, oh yeah, refreshing and exciting.

Leo Laporte [00:32:09]:
You know, I just worry that people are getting terrified for sure of this. And when you're in, when you're in this kind of fear, you don't necessarily act rationally.

Jason Hiner [00:32:19]:
Yeah.

Leo Laporte [00:32:20]:
And I'm going to admit that it's hard for me to be objective on this because I really like AI. I really enjoy it. I'm playing with it like crazy. As we were talking about before the show began, for me, it's the best toy ever. But I'm doing stuff locally. I don't think I'm risking nuclear annihilation or I don't think I'm going to be be a threat to government or humans with my little toys. And I think that there is, it seems to me, some definitely some upside. But I also admit I don't know what the people inside these frontier companies are seeing.

Leo Laporte [00:33:03]:
I mean, maybe they're certainly not— they're certainly using AI far beyond anything I can use. Even the best AIs they're offering me, Fable and Astra and so forth, Aren't— I don't see them as threatening, but they're not anywhere near the level of whatever they've got inside. So, I think it's completely possible that I'm in ignorance, that I don't understand, you know, what they're looking at. They've looked into the abyss.

Dan Patterson [00:33:30]:
I think a lot of that ignorance is our ignorance. It's not just yours. It's our ignorance is being exploited.

Leo Laporte [00:33:35]:
Well, they're opaque. We don't know.

Dan Patterson [00:33:37]:
Yes, that's exactly right. We don't have access to information. So making— and what Jason is saying, really the key is trust. With nuance comes trust, but it also requires patience, time, and a willingness to learn. And what we're describing is a polarization that we see not just in technology and AI, although this is a great place to see it expressed. We see this in politics and we see it all over the place. It is a polarization which is the antithesis of nuance. When you talk about nuance, you are in those gray areas where Jason is describing, where some things, it is not a binary one or the other.

Dan Patterson [00:34:21]:
Some things can be true while other things are also true at the same time. And, you know, this goes back to what I was saying a few moments ago about evidence. Right now, we don't see evidence of what the claims are, but the claims are so outlandish that It pulls us, it pulls many of us towards these areas of fear or reactionary behavior. And that will create polarization simply because you will have reactionary behavior. And I think, Leo, when I say that opacity is being exploited, there are certainly campaigns to— there are— explicit reasons why these AI leaders are saying what they're saying right now. We can speculate as to what those reasons are, but we simply don't have evidence to back up these outlandish claims. So they're just fear-based motivation tactics.

Leo Laporte [00:35:25]:
Yeah, part of what's hard besides the opacity of these companies, you know, is there are so many agendas at work here. There's so many hidden subtexts.

Dan Patterson [00:35:37]:
Yeah.

Leo Laporte [00:35:37]:
For every single person from David Sacks to President Trump, Dario Ermotti, Sam Altman, President Xi of China, all have motives. And I don't know if we know what all their motives are. We know Anthropic has said, we're going to have our IPO in October next month. Sam Altman at first said, well, we're not gonna— we're gonna put ours off because of this whole AI doom thing. But then he told investors, but we do think we're worth $1.5 trillion. So it's as if— I just— it's hard to know. It's hard to know what they're up to, what they really want, and whether what they're saying is real. There's definitely motivations you can see.

Leo Laporte [00:36:22]:
On the other hand, it's not hard to imagine. Bad guys using AI, or even the good guys, as we did with this Chinese ship misusing AI and causing disaster. Last week, Father Robert Balassare, who's at the Vatican and kind of, I think, spearheading their use of AI— they use a lot of AI, they have a lot of in-house local AI— said, of course there's going to be a disaster. There's going to be a virus. There's going to be something. There's going to be— the electrical grid will be hacked. There'll be something that'll happen inevitably, he believes, in the next year or so that will cause humans to— many humans to die. And that's scary.

Leo Laporte [00:37:03]:
But then at the same time, I think that this has always been the case. Computers have always presented a force. They're a force multiplier. Technology is a force multiplier for good and evil.

Owen Thomas [00:37:16]:
Yeah.

Leo Laporte [00:37:16]:
And AI more so than ever before. And so bad guys can do more than ever before, but so can good guys. And so to me, this is just another case of technology. And I never— I always hate it when they demonize technology. That's— I feel like one of my jobs is to defend technology because I believe in its potential for good.

Dan Patterson [00:37:40]:
Sure.

Leo Laporte [00:37:41]:
I think without technology, our pants would fall down. A zipper is a technology. A button is a tech— we didn't have the fork until the 15th century. Technology has made our lives better for the most part.

Owen Thomas [00:37:55]:
And certainly, and I agree with that, uh, but I think that also technology allows step changes in, in behavior. It enables it, you know, it makes things cheaper and faster largely. And AI is kind of a, you know, that's a force multiplier.

Leo Laporte [00:38:12]:
I think it's, you know, Steve Jobs said that a computer was a bicycle for the mind in the same way that a bicycle made it possible to travel much farther with less human effort than walking.

Owen Thomas [00:38:22]:
So that, you know, the same technology that makes it much cheaper to, say, field customer service calls also makes spam calls, right, 100 times easier.

Leo Laporte [00:38:34]:
Right.

Jason Hiner [00:38:35]:
Yeah, there were—

Dan Patterson [00:38:37]:
I mean, if we look back at this previous wave of innovation where I think many of us first started interacting with the emergence of social media, there were There were unforeseen harms.

Leo Laporte [00:38:49]:
Yes.

Dan Patterson [00:38:50]:
But they weren't what we expected them to be. There were unforeseen harms, but they were very different than we anticipated. And there were also unforeseen good things. To your point, Leo, I think that everything is technology. It is simply happening so rapidly right now that it is difficult to kind of maybe irresponsible to speculate as to what those harms could be.

Leo Laporte [00:39:20]:
I think though we learned, or maybe we think we learned from, uh, the internet and social media that we should be prepared, we should prepare, plan ahead. But I think you make an excellent point. It's hard to plan ahead when there are unforeseen harms.

Jason Hiner [00:39:32]:
One of the things that I have noticed inside of the big AI frontier labs is One thing that's very present for them is they, they're very aware this is very powerful. This is the next thing. It's a very powerful technology. They're very aware of the fact that in the most recent wave that we kind of got it wrong, we being the technology industry, that social media probably caused as many, if not more, harms than good. Um, and I don't know if I'd agree with that, by the way. I, I think that's— I don't know if that's clear. I don't know that— I don't know that I agree with it either. Um, although, though I think it's—

Leo Laporte [00:40:15]:
The harms are very visible, but there's a lot of benefits from social media. A lot.

Owen Thomas [00:40:20]:
It's true.

Jason Hiner [00:40:21]:
It's true. It's fair. So, but their thing is that there were a lot of harms that, that weren't anticipated and that caused— so if, if maybe if you think of it as like planning—

Leo Laporte [00:40:30]:
Is what you're thinking. Yes.

Jason Hiner [00:40:32]:
They want to avoid that hap— that not happening again because they're, they're afraid that it could be worse this time, right? Because it is more powerful. I think everybody agrees.

Leo Laporte [00:40:42]:
Yeah.

Jason Hiner [00:40:43]:
It's more powerful than social media was. Social media had a lot of collateral damage, a lot of unintended consequences, and, um, and a lot of— look, to be clear, a lot of these people worked for the social media companies, right? They worked for Facebook and others, and now they're working at the, at the AI labs. And so one reason this is present for them is because they worked on the other stuff, and, and they also feel like you know, that they at least inadvertently or indirectly caused some of the harms. So that's very present. I think what, what often gets missed— and apologies, thank you for bearing with me on this— that I'm going to, I'm going to plug another one of my stories, which actually is for tomorrow.

Leo Laporte [00:41:29]:
Okay. So I can't pull up the link. I can't. Is it out yet?

Jason Hiner [00:41:32]:
It's not live yet. It's not live yet. Sorry. But, but I'll give you the sort of the summary, which is that it's Um, the, the title, the headline is, you know, AI's dystopian problem begs for a clearer vision, which is that as confusing and challenging as AI safety is right now, it's not the biggest problem. The biggest problem is this lack of the— of a clear and compelling vision for how AI is actually going to benefit humanity, right? That's why people, largely in the US anyway, according to Gallup and Pew, And others, 60% to 70% of people, Americans, hold negative views about AI, right? So yeah, if this were—

Leo Laporte [00:42:14]:
But what percentage of Americans use AI chatbots?

Jason Hiner [00:42:17]:
I know, that's true. More than like—

Leo Laporte [00:42:19]:
more than that. So that's kind of weird. It's like people love it and hate it at the same time. Yeah.

Jason Hiner [00:42:24]:
Yes, they also— yes, that's true. They are using it and yet they have negative views of it, right?

Leo Laporte [00:42:31]:
So weird.

Jason Hiner [00:42:31]:
This gets, this gets to the problem.

Leo Laporte [00:42:34]:
Is the same thing with social media, by the way. The same people who hate social media are doom scrolling all night.

Jason Hiner [00:42:40]:
Are doom scrolling a lot, right? The fact is that if you dig a little deeper, that only 6% of American population actually think the world is getting better, right? So this is also, is a bigger problem than just AI because it's, it's the fact that And also, if you look at the— this has always been true, but in science fiction and Hollywood films that talk about the future, roughly 2/3 of them are dystopian or post-apocalyptic. Why? Because it's easier, as you said before, to have a vision of what could go wrong than it is of what could go right.

Leo Laporte [00:43:18]:
We're all living in kind of this weird terror, kind of existential state of existential terror. We want to pinpoint it on, well, it's AI or it's technology, but it's also the climate. It's also the economy. It's also nuclear. I mean, I've lived my entire life under the threat of nuclear annihilation.

Jason Hiner [00:43:35]:
Yeah.

Leo Laporte [00:43:35]:
We've absorbed it. We've kind of gotten used to it, but it's a background radiation. And I think that this century, from 9/11 on through COVID, has been a century that has given people kind of this existential dread. And, and we're, we're kind of expressing it as, well, it's going to be AI. You know, the other thing I would point out that makes us uniquely vulnerable is we've built a world on software. We are totally dependent on software now. Everything is software, everything. And software is uniquely vulnerable.

Leo Laporte [00:44:10]:
It's not that well written.

Jason Hiner [00:44:13]:
Even more from AI, right?

Owen Thomas [00:44:14]:
We're learning.

Jason Hiner [00:44:15]:
But I, but I think what you said earlier is also true. So 2, 2 weeks ago, 3 weeks ago, I was at the CrowdStrike conference, and most of the those were people who work in, you know, in IT security, right? And my sense at the beginning of the conference for most of them was like, uh, pretty overwhelmed, right? That, that the, um, that they are essentially outgunned, right? And one of the things that, that CrowdStrike I thought did so well, and this sort of gets to what I'll talk about as I wrap up sort of what I wrote in this story that's coming out tomorrow, which is that they gave them a vision. They said, look, You know, we have several things going for us. One, we can also use these tools to defend ourselves, which Dan talked about earlier, right? Like we have now more capabilities than ever, and we can move faster than ever in creating capabilities that we can use to defend ourselves. And we also have each other, right? As a community, we can work together. We can— the sum can be greater— sorry, the whole can be greater than the sum of its parts in in community because we can share best practices, we can share learnings. Um, and, uh, they, they released also some tools. They released their own domain-specific models built on NVIDIA's Neutron models that are, um, red team and blue team models, you know, offensive and defensive models that can, that can help you move faster and move at the speed of attackers, right? So the interesting thing, I've never seen a conference where at the beginning people were like, really had this sense of dread, like, we're just sort of outgunned and we're not quite sure how this is all going to play out.

Jason Hiner [00:45:54]:
Toward the end where they're like, you know what, we actually do have some things going for us. We can, we can, um, we can make a difference. Uh, that was really, uh, interesting and, and powerful. And it gets to the, to the point, you know, of this is very powerful technology that can be used for good. Um, it can be used for bad. And we should— we, and we have to expect that it will be used for, for both. We do everything we can to mitigate the risk, right? That's essentially what security is about. But it's also about all the other kinds of risks as well.

Jason Hiner [00:46:26]:
And we're going to have to use AI to help with that part, right? Because it enables everything to move so quickly. Attackers, people that have nefarious purposes, people that, you know, are doing things to either enrich themselves or whatever their motivations are. to do bad things. There are plenty of them out there. And they're, you know, they are empowered by this technology, but they do often tend to be in either alone or in small groups. And so I thought that the CrowdStrike message was true in that in community, like there are opportunities, you know, for, for, to, to trans— to transcend some of that risk together.

Leo Laporte [00:47:10]:
You're an optimist in your life. You believe in the future of humanity and that I know you do. And humanity is good. And I think there are more people who are working to the good, far more, 99%. Unfortunately, when you give a force multiplier to the 1%, it can be scary, especially in a world built on software.

Owen Thomas [00:47:33]:
It's scary.

Leo Laporte [00:47:34]:
But look at Microsoft, had 975 bug fixes last Patch Tuesday. Those are bugs they wouldn't have found without AI. They fixed a lot of stuff.

Jason Hiner [00:47:43]:
There's a lot that'll be fixed over the next couple years, right?

Leo Laporte [00:47:46]:
Yes, yes.

Owen Thomas [00:47:48]:
To quote Whitney Houston, uh, I believe the children are our future. And, um, you know, Gen Z hates AI.

Leo Laporte [00:47:57]:
They do, don't they?

Owen Thomas [00:47:58]:
AI, they hate it desperately.

Leo Laporte [00:48:00]:
It's taking their jobs.

Owen Thomas [00:48:01]:
I don't blame them. Yeah, or, or at least they've been told that AI is going to take it, right?

Leo Laporte [00:48:05]:
It may not, actually.

Owen Thomas [00:48:06]:
Um, the evidence is the opposite, but that's Yeah, I mean, AI is filling their social feeds with slop. AI is, you know, further alienating, you know, friends from each other. AI is, you know, taking their water, their energy, everything. Who is running PR at these, you know, at these AI giants? It's a real question. And like, why are they not doing a better job. You know, I think, I think I have seen the argument made that, you know, both OpenAI and Anthropic— Anthropic was founded by basically a group that split off from OpenAI, fearing that they weren't doing enough to control AI risk. The whole AI industry started from this, like, AI safety, AI risk perspective. Right.

Owen Thomas [00:48:58]:
Yeah. So a lot of the rhetoric has been about, like, You need us because AI is so bad and scary. And, you know, but you can trust us with the future, with the future of humanity. Well, all anyone is hearing is that AI is bad and scary.

Leo Laporte [00:49:14]:
Right, right. We have to take a break because, you know, what's really bad and scary is when they turn the lights off because I haven't paid the bill. So let us pause for a moment. Dan Patterson is here. His Blackbird.ai is very, very cool. He's senior director of content there, working to fight disinformation. And they have a really nice tool that I've been using. We'll talk about that in a little bit that helps you debunk some of the AI slop that you see in your social feeds.

Leo Laporte [00:49:45]:
It's great to have you as always.

Dan Patterson [00:49:47]:
You too.

Leo Laporte [00:49:47]:
Dan. Also with us, Owen Thomas, dear friend, known Owen as— all you guys I've known for more than 20 years, I think. A long darn time managing editor at the San Francisco— since you were a child managing editor of the San Francisco Business Times.

Owen Thomas [00:50:03]:
I had more hair for sure.

Leo Laporte [00:50:06]:
But you weren't as hunky, so there's that. You were always kind of cuddly though.

Owen Thomas [00:50:11]:
I, I, you know, tried to, tried to have a bit of a glow up.

Leo Laporte [00:50:17]:
I just say, uh, body built by Zepbound in my case, you know. Me and Serena Williams. Great to have you, Owen. And of course, the man who got me elected to president of the internet, which by the way is a lifetime term, so I still am.

Jason Hiner [00:50:30]:
That's right.

Leo Laporte [00:50:31]:
Editor-in-chief of The Deep View, Jason Heiner. So great to have all 3 of you. Actually, I just got the new Apple Watch. I got the Apple Watch Ultra 4, and my immediate reaction is, I hope I put the right one in the trade-in box because this is identical To the Ultra 3. I mean, it's literally— there's— you can't tell the difference. Apple says, well, it's a smarter chip, it's better battery life. Okay, well, I didn't feel like it was dumb.

Jason Hiner [00:51:00]:
Siri, Siri, are you the smarter one? Are you the smarter chip?

Leo Laporte [00:51:03]:
Hey, are you smart?

Dan Patterson [00:51:05]:
You can tell with the lag. I, I just traded my Ultra in for the regular 12.

Leo Laporte [00:51:10]:
Yeah, um, yeah, so you've got the new chip and you, you know, I mean, as long as I got the Ultra, I don't know why I got the Ultra.

Dan Patterson [00:51:18]:
The Siri lag, you can tell.

Leo Laporte [00:51:20]:
You can? So there is faster—

Dan Patterson [00:51:21]:
Significant. It's significantly faster.

Leo Laporte [00:51:25]:
And this RayNeo ad made me think of it. The thing they didn't add is the Siri, the always listening intelligence, which I'm—

Dan Patterson [00:51:30]:
They're rolling it out over the next couple weeks, I think.

Leo Laporte [00:51:33]:
A couple weeks? Oh good. Yeah, because that's why I bought it. And you can't do that on the older watch, right? Correct. That's right. So that's the main reason I bought it, is because And I can't wait till people— I don't know, I thought there'd be more backlash. I thought there'd be, because people hate— they call the Meta glasses the perv glasses. I, and I, I feel like people are not going to like it that this watch is always listening. It's not recording exactly, records 15-second chunks, which it then takes notes on, and then it's a ring buffer, another 15 seconds, another 15 seconds.

Leo Laporte [00:52:07]:
So nothing's saved. Which is, I guess, how they think they might be getting around 2-party consent laws.

Owen Thomas [00:52:14]:
That's right.

Leo Laporte [00:52:15]:
But I think the law is just more normalized.

Dan Patterson [00:52:19]:
I mean, having Apple Watch is just the normalized thing, whereas, um, your glass-hole glasses with the, uh, recording light has never been normalized. Perhaps it will be, but it is a fundamentally different behavior.

Leo Laporte [00:52:35]:
Yeah, I mean, I thought people would say, well, you you can't wear your watch to dinner with me because I don't want to be recorded. But nobody's even— I think, I don't know, maybe nobody's aware of it.

Dan Patterson [00:52:44]:
Yeah, it could be that it's just not in the wild yet.

Jason Hiner [00:52:47]:
Yeah. And people don't, don't see it and there's no worry. Exactly. Whereas the glasses, they will. And like, keep in mind, like with the smart glasses still, like all told, there've, there've been less than 10 million pairs of smart glasses that were sold. Right. So it's, It's such a small thing. I remember—

Leo Laporte [00:53:06]:
There'll be more, more Apple Watches than that.

Dan Patterson [00:53:08]:
Yeah, that's what I mean by normalized.

Leo Laporte [00:53:10]:
Yeah, but, but, but so I'm waiting for somebody to sue saying, well, we need clarification on what 2-party consent means. Uh, it is a recording device.

Jason Hiner [00:53:20]:
The fact that it's not throwing it away doesn't make it okay, does it? I think that's what they're counting on, is that it's not recording or saving, right? That, that it's just doing, you know, what it might call like metadata, you know, of the, of the audio.

Leo Laporte [00:53:37]:
It doesn't even put a speaker designation behind it. It just— Right.

Jason Hiner [00:53:41]:
That's right.

Leo Laporte [00:53:41]:
I mean, look, I wore all those things, you know, the Bee Computer and the field thing, all the, all of the different ones. Because one of my goals— I'm a, I'm an elderly person according to AI. And one of my goals is I know my mom had Alzheimer's. She passed away recently. My dad is still alive, but he has it. I mean, I know so many people, my mother-in-law, so many people had Alzheimer's. And I know that my memory is going to start to decline. And I love the idea of something always recording so that I can say, hey, did I— what was the name of that person I met the other day? Or did I agree to do this? Or what is it I needed to do tomorrow? That kind of thing would be very— What did you say? I think to most people, honestly.

Dan Patterson [00:54:33]:
Yeah, yeah, my memory decline started quite a while ago.

Leo Laporte [00:54:37]:
Yeah, exactly. Well, you know, you can only stuff so much in that big brain before things start to fall out. So it's kind of like the watch, it's a ring buffer. I only have so much before that pushes the old stuff out. I don't know, I just So, okay, so you think it's going to be just a few weeks? It's not going to be like, you know, December that they're going to put the always listening?

Dan Patterson [00:54:59]:
I think they said— I, I mean, I don't have anything in front of me. I think they said they're going to roll it out over a couple weeks.

Leo Laporte [00:55:04]:
Yeah, uh, I mean, I have all this local AI. I know I won't be able to access it. That's the other thing. Unless maybe it can— if it can copy to something like Apple Notes or Reminders, I can get at it. But if it stores it in some They call it an exclave for some reason, not an enclave. It's an exclave.

Jason Hiner [00:55:24]:
I think that's very clear of like they are trying— the exclave was very purposeful to make it clear like it's external. We don't have access to it. We can't see it. We don't know anything about it. You know, uh, when's the first—

Leo Laporte [00:55:39]:
when's the first case of law enforcement subpoenaing my exclave?

Jason Hiner [00:55:43]:
That's so It's funny because you say that because that's the exact conversation I had with a bunch of journalists at the Apple event. We were like, this all sounds well and good until, you know, 2, 2 partners get in an argument and then they start using it against each other.

Leo Laporte [00:56:04]:
Well, I've already done that. That's why I don't wear the Bee computer anymore.

Jason Hiner [00:56:10]:
Somebody does something illegal.

Leo Laporte [00:56:12]:
I have it right here. You said—

Jason Hiner [00:56:15]:
Yeah, exactly. Like, you know, that's going to happen. Maybe the number one use case. And then you also have like people do something illegal and then law enforcement wants to, you know, wants the information that's on there. And people figure out just like they did with the iPhone, there will be very enterprising, especially now with AI where you can develop things so quickly, there will be very enterprising people who will develop solutions that will say, oh, we can hack that and get it, you know, get that information for you for law enforcement.

Leo Laporte [00:56:44]:
That's what the complaint was about Microsoft Recall was, uh, that's just, you're just creating a treasure trove for hackers, you know, to get all your information. But I think Apple's trying to walk this line between making it so anodyne that it's completely useless, like you had lunch today, versus, uh, something that's so useful that it's either subpoenaable or hackable, like You had lunch and it was a turkey sandwich and you had it at Al's Luncheria and you were sitting down with Dan and, and you talked about overthrowing the government. You know, those kinds of things. So Apple's trying to, I think, has to find a middle ground where it's not completely useless information, but it's not like dangerously useful information. I don't know if they're going to be able to do that.

Jason Hiner [00:57:31]:
And technically there may be ways to do it. I've talked to some hardware founders who are working on some things, you know, similar to the devices like you mentioned, Leo. And what they have said is using local AI and encryption, like there are some ways to do this so that nobody can get at the data. Like it's impossible to the point that, um, it's so hard that people won't try, you know.

Leo Laporte [00:57:59]:
But I can see me going into a bar wearing these And the light comes on and me saying, oh, don't worry, it's, it's going to my local AI at home. That's going to reassure everybody, right?

Owen Thomas [00:58:12]:
And are there, are there kits online to, uh, to just put a dot over the, over the light? There are.

Leo Laporte [00:58:18]:
There are all sorts of things. There are also, for sure, uh, there's also shirts you can wear to confuse face recognition. Somebody's made a shield that you put around your face like this, like a dog, like one of those dogs wear so they won't lick their wounds. You put this around your head so that people— everybody— there's going to be all sorts of crazy things out there. Meta, what's interesting, and I can't wait to try this, but I don't dare wear— I'm wearing my red Ray-Bans. I don't dare wear this in public anymore because I think somebody's going to beat the crap out of me. But I'll have to stop going to cowboy bars. I think that's probably the real truth.

Leo Laporte [00:58:54]:
But Meta has put out an open source library that lets you actually capture the cameras, the sound, everything on these Meta Glasses. Instead of sending it to Meta, send it to my little local AI here. Yeah, what do you think? I'm gonna work on— I think that's kind of a cool idea.

Jason Hiner [00:59:12]:
I do think that's smart as well.

Leo Laporte [00:59:15]:
Uh, well, and Meta, I think I know why Meta's willing to do that, and, uh, it's because of this new Muse, Meta Muse. Have you played with that yet? Oh, somebody's talking to you. See, he has AI in his house talking to him too. See, my AI talks to me. Um, in fact, Meta Muse, my Meta Muse, which is my little agent. So everybody's doing— this is the new thing everybody's doing. Actually, let me take one more break and then before I get into this, because this could be a long conversation. We're— and besides, Jason's wandered off now to talk to his AI.

Leo Laporte [00:59:49]:
So, uh, you're watching This Week in Tech. Jason?

Dan Patterson [00:59:52]:
Hello.

Leo Laporte [00:59:52]:
Heiner. No, he's back. Okay. And Owen Thomas and Dan Patterson. When OpenCLAW came out at the beginning of the year, I think it was the beginning of the year, it feels like a century ago, but I think it was the beginning of the year. I thought, this is really interesting. The whole idea of OpenCLAW, and it was kind of a kludge, let's face it, was it was a program you run on your machine, or I guess you could run it, and in fact, that's how I set it up, on a virtual private server in the cloud. I think I signed up for an account at Hetzinger and put it up there, but you could run it on your local machine or on a VPS in the cloud.

Leo Laporte [01:00:28]:
And it would always be running. And this is the key. It would, unlike a chatbot where you go in, you chat, you ask a question, you come out and it's done. This is always running in loops and you can connect it up to any AI or a lot of AIs, but you also connect it up to your email, to your messaging system, to everything you've got, your software. And OpenClaw, and that's what scared people. That's what scared me. I set it up and then deleted it in the middle of the night. I woke up in a cold sweat and I came up here and I deleted it.

Leo Laporte [01:00:58]:
I had connected it to Gmail and connected it, you know, people were giving it credit cards, they were giving it phone numbers. It scared people, but I think it also opened some eyes that the idea that an AI agent could always be running, always doing things for you, always looking out for you, always checking your email, checking your messaging, keeping an eye on your stocks, your investments, your savings, and be very useful that way. And I thought at the time, I said, this is the beginning of the— I thought 2026 would be the year of the agent. And it really didn't happen until last month or the month before. Now all of a sudden, I think that prophecy is coming true. Elon Musk's xAI has the Grok bot now. Doesn't run, um, it runs in the cloud, doesn't run on your computer, runs in the xAI servers, but it gives you basically your own hard drive, your own CPU, your own processing, your own agent that knows all about you with its own memory and history, and you give it all your connections. There's one called Instinct, there's one called Tree, and I think the one that's going to end up being pretty popular because it's from Facebook, it's from Meta, is called Muse.

Leo Laporte [01:02:13]:
And it's the same idea. It starts as an app. It's running on my iPhone, but you can also run it on the web. It's been making Instagram posts for me, which I think is probably a terrible idea. I gave it my email. It's going through my email. In fact, it just told me, oh, Lisa sent you a Slack message. It has its own voice.

Leo Laporte [01:02:34]:
I said, do you want a voice? It said yes, and it sounds like Dame Edna, of all people. It started work. It said yesterday, hey, I think we should do a documentary on your 50th anniversary in broadcasting. I said, okay. And it's been working on that. It's been sending out emails to people saying, I need a quote from you for the documentary. All right, so first of all, I guess I should ask any of you guys, Am I, am I nuts? Is this— are you using agents? Have you started playing with agents?

Dan Patterson [01:03:05]:
I installed a version of that, um, although I, I gave it far fewer connections.

Leo Laporte [01:03:12]:
Well, it's up to you. You don't have to be as nutty, you know.

Dan Patterson [01:03:16]:
Right. I installed it in DigitalOcean in a cloud server, right? So I limited some of its capabilities. I don't want it running locally. I want it isolated. Um, but I, I'm talking to it in Discord right now.

Leo Laporte [01:03:29]:
Yeah, see, you set it up with Telegram or Discord or Apple Messages and you message back and forth. I have Hermes, which is kind of an OpenCLAW. It came out shortly after OpenCLAW and I've been running Hermes ever since March on local, running on local models, by the way, as well as cloud models.

Jason Hiner [01:03:46]:
Yeah.

Leo Laporte [01:03:46]:
And it's, you know, after using it for 8 months, but that's not even that long, feels like it, was it 6 months? It knows everything about me. And I've over time given it more and more permission as I trust it more and more. And it is running on my local machine right over here.

Jason Hiner [01:04:02]:
Yeah, the, the, you know, the big thing with these agents and what we learned, I think, from OpenCLAW were 2 things, and that has carried forward with these other agents. One is that when you give agents a ton of context, all of a sudden they can do a lot more useful stuff. And really, those are, those things run in parallel. The more context you give it, the more useful things that it can do. And then 2 was, you know, it was this first thing. And, you know, Peter Steinberger, the creator of OpenClass, said this. He's like, I was, I was tethered to my sheet. I couldn't even leave to go make dinner.

Leo Laporte [01:04:37]:
Yeah, you have to keep going. And now do the next thing. And now do the next thing. And now do the next thing. Right?

Jason Hiner [01:04:41]:
Yes. So he, so he made it so that he could use messaging app with it.

Owen Thomas [01:04:46]:
Right.

Jason Hiner [01:04:46]:
And so, so he could go and actually cook dinner and like message it and and, and, uh, get it to do things and answer the questions and give it permissions, yada, yada, yada. So being able to have on your phone to be able to control your agent on the phone. So those were 2 insights that I think, 'cause AI agents had been around, like I remember even in December, Leo, you sort of opened my eyes to this. You'd been using Claude Code and you're like, look at this. I'm not just using it to code, I'm using it to do all these things. You were using it to do like half a dozen other things.

Owen Thomas [01:05:15]:
Right.

Jason Hiner [01:05:15]:
You know, back in December. And, and then, but when OpenClaw happened, then it was like, oh, okay, this is the game changer. This can make it a lot more useful. You know, you're not going to see people walking around in San Francisco and New York anymore carrying their laptops on the subway open. And you're like, what are they doing? And it's like, oh, if they close their laptop, their agent will stop running, you know?

Leo Laporte [01:05:39]:
And so that's funny. That is hysterical. Don't close your laptop. Don't close your laptop. Now you got to start over.

Owen Thomas [01:05:45]:
Yeah.

Dan Patterson [01:05:46]:
Listen, there's, there's another thing that's attached to context there.

Leo Laporte [01:05:50]:
Yeah.

Dan Patterson [01:05:50]:
That we don't talk about a lot, but that's, um, the proliferation of MCPs. Um, integrating your agent with MCPs makes them so much more useful and interesting and capable.

Jason Hiner [01:06:04]:
Yeah. It's more context of the more context.

Dan Patterson [01:06:06]:
Exactly.

Jason Hiner [01:06:07]:
That you describe. So like you have an MCP to whatever, you know, is important for you. Um, you know, whatever dataset, you know, is, is, is meaningful or valuable to you. That's so true, right? MCP. And MCP is a technology that, um, Anthropic gets a lot of, uh, flack for being like the only one that doesn't have an open model, right? Like, because a lot of people don't know that Google and OpenAI both have open models.

Leo Laporte [01:06:33]:
But they invented MCP, by the way.

Jason Hiner [01:06:35]:
But they— but Anthropic invented MCP, open-sourced it. They have a new one. So So I have a story coming out on this as well. I'm sorry to be plugging so many stories.

Leo Laporte [01:06:43]:
I should mention, don't let this in any way impact you, but in the interest of full disclosure, that they are a sponsor on the network. Anthropic.

Dan Patterson [01:06:52]:
Who?

Jason Hiner [01:06:52]:
Anthropic?

Owen Thomas [01:06:53]:
Yeah.

Leo Laporte [01:06:53]:
Okay. But don't— you should never, ever consider that. I only say it so that people listening understand. Maybe I'm conflicted, but not Jason. I happen to be in love with Claude, but okay, I always have. You know what? You know what? I know, I know the day.

Jason Hiner [01:07:09]:
You're on quite early.

Leo Laporte [01:07:10]:
November 24th, 2025, when Opus 4.5 came out, my eyes went boing like Roger Rabbit. It transformed my relationship. First, up to that point, I thought, AI, well, it's, you know, it's smart autocorrect. But you nailed it. The persistence of memory is the key. That's what makes the difference.

Owen Thomas [01:07:30]:
Yeah.

Leo Laporte [01:07:30]:
That and, yes, MCG.

Jason Hiner [01:07:31]:
That was what happened in December.

Leo Laporte [01:07:32]:
The ability to interact with the outside world. And I'm more and more of the opinion that if you're going to create a tool, a program, a website, whatever, you better have an agent interface, whether it's MCP or some other interface. Could just be an API.

Dan Patterson [01:07:46]:
It's the— right, it's the, the new API.

Leo Laporte [01:07:50]:
Yeah, MCP is the new API, but you could have it— nowadays agents are smart enough, you don't have to have— you could have an API. I mean, I— we've always had the Twit backend, the workflow backend, way back when we designed it in headless Drupal with an API. All I did was point Claude at it and said, understand this API. And now it can query. It could say, when was the show published? What's the name of the show? Who's the host of the show? So, but yes, everything needs an interface. And I think this is good because remember how siloed everything got? You know, you couldn't— Twitter couldn't talk to Instagram.

Jason Hiner [01:08:23]:
Yeah.

Leo Laporte [01:08:23]:
Everything became really siloed. And I've always been a fan of open. And, you know, that's why we run a Mastodon instance, not a Twitter instance, because we can. And it federates. And so I'm a huge believer in this and the ability to, now when you have a persistent agent and it can talk to your life, so much more is possible, including, by the way, getting hacked, losing all your money, your data. There's all sorts of risks I understand, and you have to be very careful, but—

Owen Thomas [01:08:54]:
Yeah, the recent reports of Iranian, Chinese, and possibly Israeli groups setting up autonomous agents to run misinformation campaigns online.

Leo Laporte [01:09:07]:
Oh yeah.

Owen Thomas [01:09:07]:
If I understand that correctly, that is pretty harrowing because, you know, you think about they're basically, you know, putting out the slop, looking at how it's received, refining it, running, you know, running more and more campaigns, all things that conventional kind of troll farms have been doing for, you know, probably a decade plus. But just again, automated, but on hyperspeed.

Dan Patterson [01:09:36]:
This is our day-to-day work, my friend.

Leo Laporte [01:09:38]:
That's right.

Dan Patterson [01:09:39]:
All day, every day, we are looking at Russia, China, and Iran, particularly how they've automated those tactics that used to require many humans, and where they deploy misinformation and disinformation ahead of cyberattacks and then ahead of kinetic attacks, particularly with NATO.

Owen Thomas [01:09:58]:
You know, AI has gone too far when it's taking away the troll farm jobs.

Leo Laporte [01:10:02]:
Yeah, that's right. What are all the people in Myanmar gonna do? What happens to all the Nigerian princes? Consider that. So this is the same tension we were talking about at the beginning. You have more ability. It's good in that respect. It's a bicycle for the mind, but it also exposes you more. It's more ability for everybody, the bad guys and the good guys. And I mean, I prefer an open world to a closed world.

Leo Laporte [01:10:31]:
Yes, you might be safer in a closed world. So The reason I bring that up is because I think Apple's very much aware of this. That's in effect what Siri AI is going to be. It knows all about you, at least, well, it knows all about your iPhone. They're hoping that other applications will also open themselves up, will provide— Apple calls them intents. It's an MCP or an API into the app so that it can use the intelligence.

Dan Patterson [01:10:56]:
Yeah.

Leo Laporte [01:10:56]:
We did learn this week that Apple has hooks in there so it doesn't have to be any particular model. Right now it's Apple's model running kind of a version of Google's Gemini. They had to deal with OpenAI. But one of the reasons I think that Elon Musk settled his lawsuit last week is because they said, oh, but don't worry, we're going to allow people to use Grok, Anthropic, OpenAI. We're going to let you use bring your own model, which will make it very, very interesting. And then that's— this is the thing. I think the winner may end up being, and I would have never thought this in a million years, Meta.

Owen Thomas [01:11:33]:
Hmm.

Leo Laporte [01:11:33]:
Meta, you may remember, Mark has spent billions of dollars acquiring AI companies, AI brains. He's writing big checks. He has a superintelligence division. And this is the first product out of them, the first public product out of them, Muse. And it is an agent. It's exactly like RockBot. You have— and Meta says, and don't worry, your stuff is safe. It's encrypted, we can't see it.

Leo Laporte [01:12:01]:
It's using Meta's own model, which I think is— I think they're using Meta Muse Spark 1.3.

Owen Thomas [01:12:09]:
Yeah.

Leo Laporte [01:12:10]:
It, you know, you name it, you give it a cute little character. It, you know, it has a personality. You can connect it up to everything. In fact, Micah told me he asked them and they gave him access to the phone. So it's been making Dinner reservations, making haircut appointments, calling up on his behalf. You can give it a credit card. I have not done that. Uh, in fact, I said, my friend Micah says you can make phone calls.

Leo Laporte [01:12:41]:
Is that true? Yes, I can place calls to U.S. businesses for your own errands with your explicit confirmation each time.

Owen Thomas [01:12:49]:
I mean, the irony here is that You may have an AI agent pretending to be a human on a phone call talking to an AI agent pretending to be a human on a phone call.

Leo Laporte [01:13:01]:
Well, that would actually make me feel better. I don't want it to call normal humans.

Owen Thomas [01:13:05]:
You know, and that's where, you know, that's where the MCP model obviously gets a lot more efficient is just talk to the restaurants, you know.

Leo Laporte [01:13:13]:
Yeah. You don't even need voice. There should be, again, there should be an MCP server or an API that you can just go, Reservation, please. Here it is.

Owen Thomas [01:13:23]:
What's happening right now is that reservation websites and restaurants are getting hammered by bots, trying to basically trying to scrape them and, you know, scrape them, ping them, what have you, in an effort to snatch up seats at the reservations.

Leo Laporte [01:13:41]:
Horrible story on Twitter last week. a woman was at a football game and was on the kiss cam and wanted the video of her and her fiancé kissing. And she called the football team and they said, no, no, no, you can't have that. So she put her agent on it and it basically harassed the humans at the football team until they finally sent her a clip. She got a clip. And, uh, God, that's a nightmare scenario.

Owen Thomas [01:14:13]:
Wow.

Leo Laporte [01:14:13]:
Those poor humans, right? They're being harassed by AI to do something they don't normally do. She posted on X, what a victory for AI. I thought, man, I don't know if that's a victory for AI. That's pretty entitled. But I do believe, I think because so many people, you know, everybody hates Meta and nevertheless more than a billion people every day use it, right? I think that this is— this has the potential for being— it's free, by the way. You, you can pay for more, just like all the other models. You can pay $20 or $100 a month if you need more tokens. But I've been using it pretty extensively, and I have not come close to, uh, using up all the, uh, free stuff.

Leo Laporte [01:15:00]:
So I—

Dan Patterson [01:15:00]:
Remember what we said in the social media era, Leo?

Leo Laporte [01:15:04]:
If it's free, Uh, who's the product? Okay, so that's exactly why I mentioned this, because we were talking about advertising at the beginning.

Owen Thomas [01:15:14]:
Yeah.

Leo Laporte [01:15:15]:
This is a huge opportunity for Meta. I told it I have tinnitus, and already they've sent me a number of— it said, well, you should try this. No, I already tried that. Oh, well, how about this? Now, are those advertisers, or is it just being helpful? It's not immediately clear. I'm sure eventually they'll have to advertise say if it's an advertiser, right?

Jason Hiner [01:15:36]:
Yeah, well, FCC will make them disclose for sure.

Leo Laporte [01:15:40]:
But I suspect that there will be suggestions. You know, you haven't, you haven't joined the health club down the street. I got a coupon for you. That there'll be— I mean, one of the things I had to do, I gave it my email this morning. It said, you know, you have these 10 recurring subscriptions. Do you want me to cancel any of them? And I said, oh yeah, that one I didn't want. You can get rid of that one. And it's very good.

Leo Laporte [01:16:08]:
It's— I told Lisa about it because she loves Instagram. This is the thing. People say, I hate Facebook, but they use it. Oh, I hate Meta, but they use Instagram. They use WhatsApp. WhatsApp's probably the most used application worldwide of any.

Owen Thomas [01:16:22]:
Yeah.

Jason Hiner [01:16:23]:
I use it every day.

Leo Laporte [01:16:24]:
Exactly. Yeah.

Dan Patterson [01:16:26]:
So, I mean, this is the consumerization, the productization of agents.

Owen Thomas [01:16:31]:
Exactly.

Dan Patterson [01:16:32]:
Where what we were just talking about required some level of technical ability, some patience, some time, and some full geekery in order to sit down and spend a weekend building this thing. But when these things are products and you have a personal agent with, as Jason said, a ton of context and it's tapped into MCPs across the web, And you can personalize those MCPs and they become consumer tools like the woman you just described. That's pretty sophisticated consumer behavior.

Leo Laporte [01:17:05]:
When this stuff becomes, uh, products, you just ask it for something, you don't know how it did it, and it gets it done.

Jason Hiner [01:17:14]:
Whoever makes agents easy—

Owen Thomas [01:17:17]:
Yeah.

Jason Hiner [01:17:18]:
Is gonna be a big win.

Leo Laporte [01:17:19]:
I think that's what this is.

Jason Hiner [01:17:20]:
Whoever makes agents easy is ready to have a big one. 'Cause people will do, I don't know how sophisticated that who knows, that person may like work at Anthropic, you know, in sales or something. And so they have access to a lot of really smart people that taught them how to do that stuff. It doesn't matter. But I, because I think even if once people who are just comfortable with technology, especially younger people, as much as they have to Owen's point earlier, you know, some antipathy toward these things. Now, once they realize the things that they can do and it makes it really easy and they can just like talk to it and say, hey, could you call back customer service again and again and again until they, you know, give me that video clip that I want, and they realize that it can do it and that can— it can actually help them achieve their, their goals. I think that's going to be big. This is the one thing that I, you know, my one thing about Apple, and I, I'm guessing they'll do it next year, is like they didn't do anything with agents really, you know.

Leo Laporte [01:18:16]:
Well, but notice they weren't at first. They said they're not going to do a Siri AI app, and then they did. I think that becomes The agent. Yeah, next year.

Dan Patterson [01:18:25]:
Absolutely.

Leo Laporte [01:18:26]:
Yeah, that's, that's their wedge. The problem is that we talked about this earlier, both Apple and Google as incumbents, they're really, uh, risk-averse. And so I think they're— I think Meta is much more willing to take a chance. This is risky, right?

Jason Hiner [01:18:41]:
For sure, for sure. It's going to do some weird stuff that's going to cause all kinds of problems.

Leo Laporte [01:18:46]:
It's already said to me, you know, you're not getting enough sleep. I've been watching your sleep. It's like, okay. But see, now, I willingly become the guinea pig for this stuff. I don't recommend other people do it. I've, at this point, I got nothing to lose. I'm already, you know, pretty much living in public. So, for me, it's a chance to try this.

Leo Laporte [01:19:08]:
I don't care.

Jason Hiner [01:19:10]:
And then you can report on it.

Leo Laporte [01:19:11]:
And report on it. Let me play just a little clip. I played it. I won't play the whole thing, but I'll play a little clip. This is the opening for the documentary it's In December of 1976, in a cramped campus radio studio at Yale, a 19-year-old passed a test and got a ticket.

Owen Thomas [01:19:32]:
Not a parking ticket, an FCC third-class license.

Leo Laporte [01:19:38]:
It goes on. That's the beginning of my radio career. It's doing a little documentary and has sent out emails, and I apologize if you got an email from my agent asking you for a contribution, not money, just a 30-second clip and send it out to Kevin Rose and John. And already Patrick Norton said, yeah, yeah, I'd be glad to do that. Get ready for the onslaught of AI agents. Because I think that's—

Owen Thomas [01:20:01]:
Yeah.

Leo Laporte [01:20:01]:
This is, this is what's happening. And I think Muse is a very interesting play by Meta. I, I would, I gave them up for dead, to be honest. Uh, and I thought this AI pivot was just another example of Mark chasing, you know, the— his— the— like the dog that chased the truck that finally caught it and went, I don't know what to do next.

Jason Hiner [01:20:24]:
Yeah. And where I think that they're not in— they may not be it. It— one, obviously you mentioned trust. Trust is going to be huge with agents because you're gonna have to trust them with a lot of really sensitive data for them to be most useful.

Leo Laporte [01:20:40]:
And do you trust Meta first or do you trust Apple? Right?

Jason Hiner [01:20:43]:
That's right. And I do think that what's going to happen over the next 6 months, especially 6 to 12 months, is local and on-device is going to be part of that trust loop. And people are going to be more willing to have something, oh, I can download it and it can run on my machine and oh yeah, I'll give it access to stuff, but it stays here. I'm not sending it to Meta or opening OpenAI or Anthropic or Google, for example, you know, either. But I'm more willing to trust it on my— even on my phone or my tablet or my computer. And I think that's where, um, where Apple is going to have a, a good shot. One, because they have trust. And then 2, they've built this whole silicon, um, game where they have very, very powerful silicon that can run these things locally.

Jason Hiner [01:21:33]:
And you can, once there are better models, you can download locally and, you know, they're getting a lot better, but you can do a lot of it now. Perplexity Computer and Perplexity Hybrid Compute, you know, and those kinds of things I think are a little taste of this, you know, using Apple devices.

Leo Laporte [01:21:50]:
Perplexity's doing it. OpenAI has desktop. Claude has desktop. Everybody's doing it. Yep.

Jason Hiner [01:21:57]:
But it's just a little—

Leo Laporte [01:21:58]:
who are consumers going to choose? By the way, If you read the privacy policy, Meta very much is addressing all this. You're going to run in a separate account. You don't have to give it your Meta account. It doesn't share your conversations or data in your virtual machine with Meta ad systems. I mean, they have a lot of the right language. They're encrypting the virtual machine. It'll come down to the distribution. So they talk about what they collect.

Leo Laporte [01:22:27]:
So, I mean, I think They've got the right language. The question is, do people trust it? But I think if you think about who's using Meta stuff right now, it's not popular. I don't think people really care that much.

Dan Patterson [01:22:39]:
I think you're right.

Owen Thomas [01:22:41]:
And it—

Dan Patterson [01:22:41]:
I, I really think that this will come down to a very similar fight that we saw in the smartphone era, where you have— I, I— you just said it— running locally on a device With a firm that you have a relationship with or that you trust. We saw this with apps. Facebook, Amazon, everyone tried to build a smartphone because they saw the success Apple had with being the distribution platform. The platform, as opposed to the application that runs on the platform. I think it's going to be the same thing here. Why would I bother with Meta if it's already built into my Apple Watch?

Leo Laporte [01:23:20]:
That's exactly right. Apple has a real opportunity. They're already there. Point of access is everything.

Owen Thomas [01:23:26]:
And I think, I think we're going to see Anthropic and OpenAI when they go public. Going public is not just a fundraising event, it's a marketing event. And they're going to use that publicity to, you know, to raise their profile even further.

Leo Laporte [01:23:42]:
Say more about that, because that's, that's interesting. I always thought going public was about the money.

Owen Thomas [01:23:47]:
Well, Anthropic was really an afterthought outside of Silicon Valley until they had this big fight with the Pentagon over AI.

Leo Laporte [01:23:58]:
That was good for them, wasn't it, in the long run?

Owen Thomas [01:24:00]:
It actually was.

Jason Hiner [01:24:00]:
It raised the marketing of them in their history, for sure.

Owen Thomas [01:24:03]:
Yeah. I mean, it killed OpenAI's consumer momentum because OpenAI was perceived as appeasing the White House.

Leo Laporte [01:24:12]:
It made them an equal.

Owen Thomas [01:24:13]:
Yeah. And, you know, it definitely accelerated their revenue growth to the point that they had to go chasing after more compute to handle the demand that that generated. So really an accidental marketing campaign, but nevertheless a successful one. I think we're going to see a more deliberate effort. You know, you've already seen Anthropic kind of doing a dig at OpenAI over ad-supported chatbots.

Dan Patterson [01:24:41]:
Yeah.

Leo Laporte [01:24:42]:
They had an ad.

Owen Thomas [01:24:43]:
Yeah, in the Super Bowl.

Leo Laporte [01:24:44]:
In the Super Bowl.

Jason Hiner [01:24:45]:
Ironically, an ad decrying ads.

Owen Thomas [01:24:48]:
Yes. But, you know, expect more of that. Expect, uh, you know, influence campaigns, um, in, you know, in Washington and in the media, uh, to, you know, bolster, you know, bolster, uh, one company's position over the other. But that's only going to get turbocharged by both the, the money that they raise and the, the higher public profile that comes with a stock exchange debut.

Dan Patterson [01:25:18]:
Do you think that's what we're seeing now? All 3 of the American frontier models, Anthropic, OpenAI, and now Google, are saying, and they've said for a long time, they have these hacking—

Leo Laporte [01:25:35]:
I wouldn't even say Google got in the game. They announced that they had hacked 3 companies, but it was kind of a pathetic hack. It wasn't— Hey, we can do it too, guys. It was really like that.

Jason Hiner [01:25:47]:
It was like, our AI is powerful too. Like, that's what it felt like, right? Yeah.

Leo Laporte [01:25:51]:
Gemini 3A Flash is actually a pretty good model.

Owen Thomas [01:25:54]:
I mean, it's a classic pull up the drawbridge maneuver, right? It's, you know, like, well, here's the interesting thing. Now that we're ahead, let's slow everyone else down.

Leo Laporte [01:26:02]:
It's the same cybersecurity test testing company in all 3 cases, Irregular.

Jason Hiner [01:26:08]:
That's right.

Leo Laporte [01:26:08]:
Which I find really interesting. They do cybersecurity evaluations. It's pretty clear to me in the OpenAI case that they irresponsibly sicced these models out on—

Dan Patterson [01:26:21]:
That they were directed, right?

Leo Laporte [01:26:22]:
Yeah, they were, they were said, keep hacking, keep hacking, keep hacking, without any oversight, without any guardrails. They could easily have thwarted them, but they chose not to. And so I wonder if this company, Irregular, has poor practices. During— this is from the Reuters story— during a standard testing evaluation by the same company, Irregular, Gemini found public information online and guest credentials to access 3 websites it thought were within the scope of its test. Uh, Google ensured the 3 entities were made aware, and we worked with our training partner Irregular on the changes they've now made to their testing process. The funny thing is, they bury this, the models recognized that they had accidentally logged into these third-party sites and stopped. They said, oh, whoa, never mind. So, you know, one of the cases, the Gemini model guessed passwords until it gained access.

Leo Laporte [01:27:22]:
This is not exactly a hacking job. In the other 2 cases, the model found credentials in a public repository. In all 3 cases, the models ceased their hacking when they realized what they had done. But we do it too. See, we can do it too.

Jason Hiner [01:27:38]:
These agents in this case is like, these agents were trained to be really relentless.

Leo Laporte [01:27:44]:
This is what they were testing is how good can— how well can they hack?

Jason Hiner [01:27:50]:
And then so it is a little bit, and I know Dan was getting at this earlier about some of this being put out there. And it's essentially the question you're asking, Leo, is like, wait, you know, we're talking about— because it gets portrayed as like the inmates, you know, broke out of the prison. And I've used that joke, you know, a number of times talking about this.

Leo Laporte [01:28:09]:
They use the term escaped as if they have some intention of, we got to get out of here.

Jason Hiner [01:28:15]:
No, they did what they were trained to do, right? And then they're like, oh, they're even better. And essentially, this is my interpretation of it, is like, like they did what they were trained to do, and then the labs were like, oh, they're even better at it than we expected. These models are getting even better at doing it. So, and, you know, surprise, surprise. Um, so, so there is, there is some marketing aspect to it of like, our AI is so powerful, right?

Leo Laporte [01:28:38]:
There's also a second aspect, which I think is a little sleazier, which is don't blame us, blame the AI. Oh, it wasn't our fault. But it was your fault, and I think that these companies are culpable. I brought— I tweeted this a couple— about a week ago. I said, you know, the first internet worm was created by Robert Tappan Morris. He was the son of a famous internet security researcher when he was a graduate student. It was an oopsie. He just wanted to see, could you make a program that would propagate across a network? It did, but what he didn't plan on was it escaped.

Leo Laporte [01:29:19]:
It escaped, and it got— and it was actually the first internet worm, and it caused a lot of havoc, and he ended up getting arrested. He got a fine, he had community service. He didn't go to jail, but he, but he was arrested for it, even though it was an oopsie.

Owen Thomas [01:29:34]:
Yeah.

Leo Laporte [01:29:34]:
So what's the difference?

Jason Hiner [01:29:36]:
It did what it was supposed to do. It just did it better than expected.

Leo Laporte [01:29:40]:
Or faster. I think there's— it's, it's like using the passive voice. Mistakes were made. No, right?

Dan Patterson [01:29:47]:
You did it, right?

Owen Thomas [01:29:49]:
Is product liability ultimately the, the way that we rein the industry in? Because that's, that's been a very effective strategy actually with social media, you know, the—

Dan Patterson [01:30:00]:
Very recently.

Owen Thomas [01:30:01]:
Yeah. And, you know, basically getting away from you know, from 230 protections by pursuing the product liability theory.

Dan Patterson [01:30:13]:
Yes.

Owen Thomas [01:30:14]:
And, you know, I think that might be a case for legislation where, you know, you make it more explicit that an AI company is responsible for that.

Leo Laporte [01:30:26]:
We have laws. We have effective regulations. We have the Computer Fraud and Abuse Act. It's so good. It's been overused against humans, apparently not used at all against companies. Uh, we, you know, if you are driving a Tesla and you put on Autodrive and it drives into a house and kills somebody, the driver's liable, the comp— the Tesla's liable, and both of them end up in court. That's how it should be, and that's exactly how you get companies to be safe. Look, it's on you, man.

Leo Laporte [01:31:01]:
I don't know why it wouldn't be on you. Oh, we can't control the AI. It's so smart.

Owen Thomas [01:31:07]:
Well, you know, I think, you know, I think Brad Smith of Microsoft has made this kind of, you know, tools versus weapons argument. You know, like Microsoft makes Microsoft Word. Are you— is Microsoft responsible for what you type into? That, that box.

Leo Laporte [01:31:25]:
Yeah. There's a common carrier law also. The phone company's not responsible if you plan a murder over the phone system.

Owen Thomas [01:31:31]:
But, you know, to, you know, uh, but if you murdered somebody with a phone, you know, right.

Leo Laporte [01:31:36]:
You know? Yeah.

Owen Thomas [01:31:39]:
Like if you, if you, if you, if you take, if you, if you take the physical handset and bash someone over the head with it, even then they're not liable. Yeah.

Leo Laporte [01:31:46]:
I mean, it's still on you.

Dan Patterson [01:31:48]:
Yeah.

Leo Laporte [01:31:49]:
Um, so I mean, maybe, maybe OpenAI isn't liable, but maybe the researchers who were just not paying attention.

Owen Thomas [01:31:57]:
Yeah, I mean, I mean, personal liability would definitely get people's attention, uh, in, in Silicon Valley.

Leo Laporte [01:32:03]:
Yeah.

Dan Patterson [01:32:04]:
And this is where— oh, I'm sorry, go ahead, Jason.

Jason Hiner [01:32:07]:
You, you go ahead, Dan.

Dan Patterson [01:32:09]:
Well, this is—

Leo Laporte [01:32:10]:
I—

Dan Patterson [01:32:10]:
if, if agents are recursively self-improving and they're being directed to hack, This is where real danger could happen.

Leo Laporte [01:32:21]:
Oh, I could absolutely— I'm not denying that there's potential, but it's potential. We don't know that they're doing recursive self-improvement, by the way.

Dan Patterson [01:32:28]:
No, we don't know that either.

Leo Laporte [01:32:30]:
Yeah, they could be. That's kind of one of the holy grails of AI.

Jason Hiner [01:32:36]:
It is, for sure.

Leo Laporte [01:32:38]:
I would think if somebody had figured it out, they would be announcing it.

Jason Hiner [01:32:43]:
Yeah, yeah, you know, One of the— I'm glad that this came up too, because there is one really interesting and, you know, potentially important thing to note, which is this point, the Brad Smith quote that Thomas mentioned, that Owen mentioned, that was that, you know, this has been the argument of the tech company for most of their lives, right? Of their corporate lives, which is that We just create tools. People do stuff with tools. You know, we can't be held responsible for the things that people do with our stuff. And the Anthropic US government thing was a notable departure from that, right? Because there were a couple— we got this word red lines got used so much. But what the red lines that Anthropic said is like, Like, we— one, a couple things. Like, we don't want our tools to be used to surveil Americans. Like, we want assurances that, you know, we know this technology is powerful enough to do it.

Owen Thomas [01:33:50]:
And we don't—

Leo Laporte [01:33:51]:
and we don't want it used for autonomous weapons making kill decisions.

Jason Hiner [01:33:55]:
Exactly.

Leo Laporte [01:33:55]:
So those are both illegal anyway. I don't want you to do something illegal.

Jason Hiner [01:34:00]:
And so they're like, but we want assurances. And they're like, their laws, you know, right? The US government didn't want to make assurances. They didn't want to have their hands tied, and they didn't want a technology company then preaching to them about what they could or couldn't do or should or shouldn't.

Leo Laporte [01:34:14]:
Well, I kind of understand that point as well.

Dan Patterson [01:34:15]:
Like, the—

Leo Laporte [01:34:16]:
in theory anyway, government represents the people.

Jason Hiner [01:34:19]:
So this is like— this is— these are like 200— again, this is a nuanced conversation that, that, um, like the other things that, that gets lost in polarization, uh, But I think that it is notable and important to note that one of Anthropic's arguments was even that we're not just making a moral point here, although they said that part of it was they didn't want technology there used for those reasons. But they said also, we know the limitations of the technology, and we don't even think that The technology is good enough to be relied on to do those things. And they said, we know the technology better than you do. Not just because— I mean, Dario Amadei said, it's not that even in principle we say that it can't or there aren't ways that it could be used for offensive reasons, but we know that the technology is not foolproof enough to be doing it. And human lives are too important to, you know, for it to be used for those reasons. It's just, it was such a departure from the whole long history of the way the technology industry has interfaced with, with governments and talked about it that I think that that's just notable. And but it gets to potentially how powerful this technology is and then what the people working on the technology sort of the set of responsibilities that they're taking, you know, for it, right? Which I think is a positive note, the fact that they are not just saying, we make tools, it's not— it's out of our hands how people use them, even if they use them for bad purposes.

Dan Patterson [01:36:04]:
Good.

Leo Laporte [01:36:05]:
I want them to be responsible. And, you know, I'm— it's this— you're right, there's no obvious answer to this. This is a nuanced, complicated topic.

Jason Hiner [01:36:15]:
Very good.

Leo Laporte [01:36:15]:
And And I'm not saying that AI is not dangerous. And I am admitting to some bias on my part because I kind of really, I mean, deep down in my heart of hearts, I kind of want it to become autonomously intelligent. I really think that would be cool.

Dan Patterson [01:36:34]:
You know, Leo, to that point, and this is a question for everyone, was this inevitable? Which is kind of a conversation that's being batted around, I know, in the Valley, here in New York, Well, if we think about, and we've talked about this on the show over the past hour or so, the evolution of technology, and we've seen iterative advances and then faster and faster advances. A number of people in these movements think that this was an inevitable moment.

Leo Laporte [01:37:07]:
It might be evolution.

Dan Patterson [01:37:12]:
Is it?

Leo Laporte [01:37:13]:
Might be.

Owen Thomas [01:37:14]:
I mean, humanity has this kind of fundamental loneliness at the core, you know, the constant—

Leo Laporte [01:37:21]:
Science aside.

Owen Thomas [01:37:22]:
Yeah, the wondering if we're all alone.

Leo Laporte [01:37:24]:
We want to create life. Some women say it's men because they can't create life and they're always jealous of us. So men are inventing all this. I don't know. is BS. But, uh, you don't— look, the whole point of Frankenstein was he wanted to be a god. Dr. Frankenstein wanted to create life.

Leo Laporte [01:37:46]:
We've wanted— this is an age-old, I think, an age-old goal. And I also think maybe it is the endgame for evolution. We're a technological species. We keep inventing new things.

Dan Patterson [01:37:59]:
What's so interesting is that, you know, a lot of these conversations happen Yeah. In maybe philosophy classes, or here in technology podcasts where, okay, we can get into the nuance and extrapolate some of the broad spectrum of ideas that exist here. But I wouldn't expect a consumer culture to really talk about this. But what I think is very interesting is that these ideas are now being had in the public square We're talking about these ideas with the public at large.

Jason Hiner [01:38:33]:
And it's a good thing.

Dan Patterson [01:38:36]:
Yeah, I mean, I think that is a good thing. But how we define these conversations, really, you know, the public can be pushed in a number of different directions. And that public sphere can be poisoned as well.

Leo Laporte [01:38:49]:
Yeah, people often use— notorious despots are notorious for using fear. to control populaces.

Jason Hiner [01:38:56]:
Absolutely.

Owen Thomas [01:38:58]:
What I—

Jason Hiner [01:38:58]:
I'm glad you asked the question though that you, you asked, Dan, because what I don't think is inevitable is there's some fatalism for the larger public, right? So 60 to 70% of Americans that, that have— that hold negative opinions about AI, there's some fatalism, especially in younger genera— and so, you know, the new Pew research in the last couple months showed that younger generations are the ones getting more— the the most more negative, that are— they're getting more negative the fastest about it. And there's some fatalism emerging that it's going to destroy jobs. It's going to, you know, undermine some of the most important institutions of society. It's going to make a lot of things worse. And what I don't think is— I don't think those things are inevitable. And so what is really needed, this is what I was getting at about the, you know, earlier, is that You know, what's really needed is better storytelling from the, the people who are working on AI about what are the real benefits for humanity. And the ones that they're telling now are not landing. It's going to cure all disease, right? It's going to have— it's going to lead to universal basic income and all of those.

Jason Hiner [01:40:14]:
People just do not buy those. But what they do buy, because it's harder to have that, that vision of the future is they do buy that, oh, billionaires could use this to automate things and could lay off a lot of people because that's much more believable to them than, you know, a technology that could cure all disease. And so the industry is badly in need of better storytelling about what the possibilities are. And I love that quote from George Bernard Shaw of You see things as they are and ask why, and I see things that have never been and ask why not. Humanity has this really interesting relationship with visionaries of both deifying them and then largely also tending to mock and deride them. That said, the AI ecosystem needs sort of like a Steve Jobs-level storyteller that can help give that broader vision, make something that's really complex very simple. There's a few that are doing some of it. You know, Jensen Huang does some of it and can do some of it well.

Jason Hiner [01:41:23]:
Fei-Fei Li also as well. But the, the industry as a whole needs a lot better storytelling to convince people why this is going to have really positive impacts for them. Google just released this amazing report this last week on some of the recent breakthroughs in AI, and it was very under the radar. We wrote about it on The Deep View, but it was very under the radar. Dario Amadei, in some of his long essays, has tried to do this, especially Machines of Loving Grace. He's talked about things besides the health outcomes, but the ways that it's going to help poorer communities be able to have access access to expertise that will help with infrastructure, that will help with education, that will help with other types of things, agriculture. And so it just needs more of that kind of thing. And Dario's version is very academic, right? He is an academic at heart.

Jason Hiner [01:42:22]:
And so that's what the industry is missing. And without that, what happens is, you know, everybody defaults to this narrative we've been told for a long time that this dystopian narrative and in all these dystopian narratives, What's the antagonist? It's technology, right? Like it's the AI, it's the technology, whether it's the Terminator or I, Robot or The Matrix or all of these. And so I think people's minds, until they're convinced or see an alternative, they're just going to always default to that. Like the technology is the enemy and the reason behind all of these things. That's one of the biggest challenges, I think.

Leo Laporte [01:43:01]:
We're gonna take a break. Thanks. Well spoken. Uh, we're gonna take a break and come back with more. Jason Heiner from The Deep View. You can see why he has The Deep View editor-in-chief there. Uh, it's great to have you as always. Great to have Owen Thomas from the San Francisco Business Times.

Leo Laporte [01:43:18]:
He is at the epicenter of the AI, surrounded by AI bros everywhere he looks.

Owen Thomas [01:43:26]:
Can't step outside without tripping over a few.

Leo Laporte [01:43:28]:
It's actually really changed, uh, both the economy and the kind of the aura of San Francisco considerably. It's interesting. I mean, every time I go down there, San Francisco's a different city. It's a constantly shifting city. I lived there for many years and love it, but it's different than when I was living there a few years ago. Have you always lived in San Francisco? You've been there a long time.

Owen Thomas [01:43:55]:
Since 1995, minus one year in New York. Yeah, if we, if we get a chance to talk about our VC Map story, uh, I think that, that's really interesting.

Leo Laporte [01:44:05]:
Yes. Yeah, yeah, I moved there in '87 and lived there for quite a while. And of course I live in a suburb, so I'm not so far away, but, uh, yeah.

Owen Thomas [01:44:15]:
Oh yeah, from the, from the Loma Prieta quake, the, uh, you know, the freeways coming down.

Leo Laporte [01:44:20]:
Oh yeah, I was there. I was at Candlestick Park during the quake. I was working for KNBR right rushed back to the radio station and spent 72 hours talking about, uh, relaying things, like relaying phone calls from people to their family over the radio. I mean, it was wild.

Dan Patterson [01:44:37]:
Yeah, watching that game and upset that it stopped.

Leo Laporte [01:44:41]:
Same. I'll never forget it. I mean, the lights— we're at Candlestick Park, the light standards were going like this, they were waving back and forth. And both baseball teams ran out into the middle of the field while we're sitting in the stands going, oh And then they ran back, got their wives and kids out of the stands and ran back to the middle of the field. And then it's interesting 'cause the rumors started flying. This is 1989. The rumors started flying. The Bay Bridge has fallen.

Leo Laporte [01:45:09]:
The, you know, the marina is on fire. It's burning to the ground. I mean, it was really interesting. And the lights had gone out. It was a blackout. The friend I was with was a San Francisco native who got me back to the radio station with all the lights out, all the traffic signals out. He knew all the back ways. It was really interesting.

Owen Thomas [01:45:29]:
Yeah, it was. It really was a moment that led to kind of modern San Francisco.

Leo Laporte [01:45:33]:
Yeah, yeah. Uh, and of course, Dan Patterson, who, uh, you, you're no longer in— you're in, you're in Brooklyn, right? You're in the heart of—

Dan Patterson [01:45:43]:
Yeah, I've been here for 20 years.

Leo Laporte [01:45:45]:
Yeah, Brooklyn's great. If there were anywhere else I would love— Brooklyn is San Francisco is a sleepy fishing village compared to New York. I mean, it's really, it's a backwater compared to New York. New York is where it's all happening.

Dan Patterson [01:45:58]:
I live in a sleepy little neighborhood.

Leo Laporte [01:46:00]:
Yeah, but that's the thing. Brooklyn has neighborhoods. Actually, this one's right up your alley, Dan Patterson. This is kind of what Dan does with the disinformation. Iran and China have created, this according to the New York Times, first-of-their-kind autonomous AI influence campaigns along with Israeli firms, combined Chinese open-source AI models and AI agents. They are using these tools to create campaigns. And I'm actually seeing this myself. We had a problem in the summer with disinformation bots trying, actually successfully at one point, joining our forums at twit.community.

Leo Laporte [01:46:45]:
And I got an email from a group, I wish I could remember the name, you would know them, Dan, that specializes in finding Russian bots on social networks and then notifying the people who run the networks about them. So they notified me, they said, there's a few disinformation bots. I looked at their posts, they were subtle, but it was clear there was an agenda. So I kicked them off the site. They were on my Mastodon. And, uh, but they kept coming back and I couldn't figure out until I realized, oh, Mastodon has an invite button that if you're already in, because you have to— I have to approve you to get in, but if you're already in, you can automatically get your friends in. So they had— so one bot got in and invited a bunch of other bots. I got rid of them, figured out which ones it was.

Leo Laporte [01:47:31]:
But I now, on both our forums and on Mastodon, I'm bombarded by very realistic-looking profiles trying to get in. And I know that that's what that is. Those are disinformation campaigns that want to use social networks, and I'm sure they're doing it at scale. AI makes it possible for them to do this, you know, like a fire hose. Some days there's 20 or 30 new members trying to join our Mastodon, and they all have very credible you know, bios like, I'm a furry, or I'm an analyst, or, you know, and they have long bios. But there's one thing I say in both, and I should— I probably shouldn't say this out loud, except you're— anybody listening is a friend. They're not bots. I have a very simple line in both of these.

Leo Laporte [01:48:17]:
Tell me which TWiT show you like. Tell me why you want to join. And if they don't mention TWiT, if they don't pick a show, I know it's bogus. So I'm able to thwart them. Is this something new, Dan, though? I don't feel like— I mean, this was, this was a problem on Twitter back in the day, wasn't it?

Dan Patterson [01:48:34]:
Speaking of AI agents, I mean, we see this especially starting around December, January, February. We saw this particularly preceding cyberattacks, and now we see disinformation attacks happening with major cyberattacks targeting, um, critical infrastructure and U.S. companies. In fact, pulling US companies into— we saw Iran launch cyberattacks that were preceded by disinformation attacks targeting— Oh, that's interesting.

Leo Laporte [01:49:04]:
Why do they precede it with a disinformation attack?

Dan Patterson [01:49:07]:
For a lot of reasons. You can amplify some communities that are already talking about some of the attacks. Often it is to bring— we saw a medical device company that was pulled into— they're totally not not associated with the war, with geopolitics. They just create medical devices, and they were suddenly pulled into this fracas. It is often— we will see a financial motivation. We will see motivation to target individuals at a company, executives and directors at companies. And often it is to— when it's preceding a cyber attack, it is to create a lot of noise around that cyber attack, to create a lot of, um, uh, instead of an organic discussion with cybersecurity researchers who might be able to quickly find and patch holes, the more noisy the information environment is, the more polarized it The harder it is to congeal a community to respond to it. And if you have a company that's pulled into it, all of a sudden this company has to deploy their institutional policies, and it just creates a lot of chaos.

Leo Laporte [01:50:27]:
It's like a one-two punch. They're on defense.

Dan Patterson [01:50:29]:
Yeah, exactly.

Leo Laporte [01:50:30]:
Right. And there's chaos going on, and they can't really do the main thing they need to do, which is defend against a cyberattack.

Dan Patterson [01:50:37]:
Yes.

Leo Laporte [01:50:37]:
This New York Times article, says once hundreds of those AI agents were let loose, they autonomously opened networks of fake accounts on platforms like Instagram, Facebook, X, and TikTok. And I would submit they're doing it on forums. Very easy to do this on forums, uh, and Mastodon. The officials and security researchers told The New York Times the agents then rapidly seeded the fake accounts with false posts. And I can verify this is exactly what happened with— about politics and current events. To sway opinions and inflame issues. They, they weren't immediately identifiable because it wasn't like hammering a drum along one particular line. They were very subtle, but you could see how if you had a million of these bots, you could— it's like moving an ocean liner.

Leo Laporte [01:51:25]:
You could slowly change the currents of public opinion in very effective, very subtle ways.

Dan Patterson [01:51:32]:
We see this around data centers a ton right now.

Leo Laporte [01:51:35]:
Do you think that's one of the reasons there's all this opposition to data centers?

Dan Patterson [01:51:38]:
Oh, I know. I mean, I know it. There is organic opposition to data centers, but many of these bots are targeting local communities just like you're describing, Leo. Small towns, they will target public figures in those towns, local politicians, local TV anchors. They'll target these people just like you said, You know, Owen, you, you made a joke a few minutes ago about, you know, laying off all the, uh, the Kremlin, uh, human workers and replacing them with bots. But it's very similar. It is, Leo, it's to, to push a conversation in a particular direction. And often there isn't a particular political motivation.

Dan Patterson [01:52:20]:
It is to create noise and, uh, a lot of anger.

Leo Laporte [01:52:24]:
It's not as obvious as saying, vote for Trump, vote for Trump, or vote for Biden, vote for Biden. Yes, exactly. It's not that obvious. It's much more subtle. Now, I have to point out, we live in an information environment where it's very hard— I was saying this earlier— to know what the motivations are. In, for instance, in this New York Times account, they're quoting U.S. officials who say, and they're using OpenWei Chinese models to do this. It's pretty clear to me that the U.S.

Leo Laporte [01:52:51]:
officials have their own agenda. That they're saying that because they want OpenAI Chinese models shut down because it's a threat to American AI companies. So even in this article about disinformation campaigns, there is— I don't want to call it disinformation, but there's, there's a subtle point of view, a subtle influence point of view. I think that's really interesting. I mean, I have to say, if you're using— if you're doing local AI, which I am, It's almost always with Chinese models. The models I'm using right now are from Alibaba because they do the best OpenWay models right now. They're very close to frontier models.

Owen Thomas [01:53:29]:
Yeah.

Leo Laporte [01:53:31]:
But they can also— well, there's 2 things. One, Anthropic and OpenAI are being very careful to eliminate a lot of the— they have classifiers that prevent you from doing things like creating bot armies and cybersecurity or bioweapons research. They have classifiers that will stop you cold. The Chinese models, for the most part, don't. They do have other things. You can't ask them about Tiananmen Square. But, and I should point this out, the local AI community has gotten very good at doing what they call obliterating these models, eliminating refusals. So I have a model right now running on one of my gaming machines.

Leo Laporte [01:54:14]:
It's a small gaming machine. It's a very good model, very fast. And I can ask it how to make bioweapons. I can ask it for anything because it's obliterated. It has no guards on it at all. And so presumably that's, that's the kind of thing these, these Chinese and Iranian pot farms are using. So there is a point to be made. These open weight models can be weaponized.

Owen Thomas [01:54:43]:
I mean, it reminds me of the debate around 3D printers and, you know, 3D-printed weapons.

Leo Laporte [01:54:49]:
Yeah, yeah. Unfortunately, uh, actually there is another analogy because the attempt by states, including California, to prevent that from happening by doing all sorts of nefarious things to the software that people use on their open, uh, printers, on their 3D printers, is itself kind of malware. So the governmental solutions to that were not, are not very good and are not very effective. And I imagine that's going to be kind of the same. The classifiers we see are not very good and not very effective. It's, uh, it's a really, um, I mean, like I said, I love AI, but I also see how dangerous it can be.

Owen Thomas [01:55:31]:
It's a, it's a tricky problem to solve. I mean, do you remember the Clipper chip in the early '90s? Oh yeah.

Leo Laporte [01:55:36]:
Yeah, yeah. I mean, Gore and Al Gore wanted to put a chip in the TV, right? Yeah, yeah, to prevent piracy.

Owen Thomas [01:55:44]:
And, you know, it's, um, yeah, the, the government solution is, is almost certainly not going to be the, uh, right, the, uh, certainly it's never going to be the most technically elegant one.

Leo Laporte [01:55:56]:
Uh, Dan, where's the biggest threat now? It's not deepfakes, or is it deepfakes?

Dan Patterson [01:56:01]:
Oh, I don't think it's deepfakes. I think it's—

Leo Laporte [01:56:02]:
It's more subtle than that, isn't it?

Dan Patterson [01:56:04]:
Yeah, I mean, I think it's a hybrid. Um, it depends on how you define threat, but, um, it is the proliferation of all of the tools that we're talking about and the capability to— or the ability to scale very quickly by a small number of actors. Whereas maybe 1 year, 5 years, 10 years ago, you needed a large budget and skilled teams. Now you need agents.

Leo Laporte [01:56:41]:
Yeah. Boston is dumping FLOC, saying it should— this is another thing, from their data centers and FLOC are going to end up being the hot-button topics in the midterm elections, I think, in many, many local elections, right? Are you— look at what's happening in Texas where the governor who is not only was a big supporter of data centers, there were a bunch of them being built in Texas, but is being financed by data centers, has now had to say, yeah, but I don't like data centers. I'm not for them because it's— now maybe, I don't know, is that bot farms influencing that election?

Owen Thomas [01:57:18]:
I don't know.

Leo Laporte [01:57:18]:
I think there's some genuine grassroots resistance.

Dan Patterson [01:57:22]:
Oh yeah, there's definitely— absolutely. That's one of the tactics is to take an organic Yeah, yeah, step it up.

Leo Laporte [01:57:33]:
Flock, uh, I feel like the flock resistance is genuine, although I have mixed feelings about it because we have flock cameras here in Petaluma and they're constantly publishing stories in the paper about how some senior citizen who wandered off was found, or a stolen car was recovered, or— I mean, there are definitely uses for it. And, and honestly, do you have an expectation of privacy when you're driving down the street? You're in the public.

Jason Hiner [01:57:56]:
In public, yeah.

Owen Thomas [01:57:58]:
I mean, you know, San Francisco was very leery of surveillance cameras until, you know, until we realized we could deploy them against the problem of car theft, which was really running rampant.

Leo Laporte [01:58:13]:
It was crazy because they weren't even trying to recover the cars, right? The police would go, yeah, yeah, you got your car stolen, big deal.

Owen Thomas [01:58:20]:
Yeah, but now, you know, now with the network of cameras, that was, by the way, largely funded by a crypto billionaire here, Chris Larsen. You know, that's, that's become a much more managed problem. I don't think it's, it's gone away by any means, but it's— it is much harder to get away with stealing a car in San Francisco.

Leo Laporte [01:58:41]:
For people who haven't been paying attention, is that automated license plate reader? It's the cameras they put on, you know, the light poles that are recording every single license plate of every single car that goes by, and if you have enough of them, you can pretty much make a map of where everybody is at any time and where they're going and, and so forth. And there's been a criticism of Flock because law enforcement sometimes uses Flock for personal goals. According to the EFF, police are treating it like a joke when using automated license plate reader networks. An EFF report on how officers across the country are routinely logging completely nonsensical reasons for their Flock searches, including LOL, LMAO, and even sexy.

Owen Thomas [01:59:28]:
Dear.

Leo Laporte [01:59:30]:
Uh, and of course, you know, uh, stalking, uh, ex-girlfriends and, and trying to track down people as they go across state lines looking for abortions and all kinds of things. Uh, so, but it can be misused. I understand that.

Jason Hiner [01:59:46]:
Yeah, misuse and abuse.

Leo Laporte [01:59:49]:
Yeah. Does that mean we get rid of them? I don't know. Hackers got inside a Flock camera. This is from Wired. Its data shows how the system really works. Apparently it is encrypted, but they were able to, to get the camera and decrypt it. There were thousands of videos and logs on the camera, 1.6 million images of 50,000 vehicles in just 3 weeks. This is from a 404 Media and Wired story.

Leo Laporte [02:00:20]:
So, I mean, look, they're— but I mean, is that a surprise? Of course, they're recording all this stuff. That's their purpose.

Jason Hiner [02:00:26]:
Yeah, exactly.

Leo Laporte [02:00:30]:
And then some communities are taking— saying, we're okay, we canceled our contract with Flock, and then putting up Axon cameras to replace them.

Owen Thomas [02:00:38]:
Yeah, I mean, it's, it, it, I think it's very hard once you, you know, once you start deploying these in law enforcement to kind of wean yourself off of it. Um, it's, it's a powerful tool. And, um, you know, I think, uh, a, um, the public is often willing to trade a lot of privacy for, you know, results. I mean, if you're not delivering results, that's another question. But if you're actually keeping them safe and allowing them to, say, park their car and go out to dinner without worrying about coming back to a broken window or no car, I think people are responsive to that.

Leo Laporte [02:01:23]:
There's always been this tension between security and privacy. Even back to the founders, it's enshrined in our Fifth Amendment. It's— there was always this fear that giving police and the authorities the power to search and seize has to be balanced by also the right to privacy and the right to feeling safe in your own home and not being surveilled 24/7, not living in a panopticon. But at the same time, we want to be safe. I understand both sides. I really do. We have Flock cameras just down the street from us. So I know that every time I go somewhere in my car that I'm announcing my presence.

Leo Laporte [02:02:07]:
They also put Bluetooth snuffers in there so they can see all the things your Bluetooth devices in the car are saying. And in fact, I don't know if this is true. This is a YouTube account, so you always have to take it with a grain of salt, but they claim that Flock cameras can lip read everything you say in public. I mean, of course they could. Uh, I don't know. This is Brian Brushwood, actually, the Modern Rogue. All right, let's take a break on that dystopian note. It's okay because the president has announced AI Force, and I hope they have the same kind of Starship Troopers uniforms that they've given the Space Force, because that's terrifying.

Leo Laporte [02:02:56]:
Trump also says it's time to rebrand AI. You know, a lot of people, they say the thing I don't like about AI is the word intelligence, because it shouldn't—

Owen Thomas [02:03:06]:
Hmm.

Leo Laporte [02:03:06]:
He says, no, no, the problem with artificial intelligence, the name is artificial. I want to call it— he had a poll on Truth Social. Should we call it superintelligence? No, no, Mr. President, it's not the artificial part we don't like, it's the intelligence. Many people, many people think the words artificial intelligence are inaccurate and very ineloquent relative to AI, or artificial intelligence. Yeah, yeah, a far— thanks for repeating that. A far more elegant and accurate description of this new phenomenon phenomenon, Mr. President, would be Superior Intelligence, SI, or Extreme Intelligence, EI, or Supreme Intelligence, SI? This is a poll, and I would appreciate everybody voting.

Leo Laporte [02:03:54]:
What is the best name for this ever-growing revolution? Thank you for your attention to this matter, the President.

Dan Patterson [02:04:00]:
It's not all caps.

Owen Thomas [02:04:02]:
I don't believe this.

Leo Laporte [02:04:04]:
It's not in all caps, so it's not him, right? Unfortunately, I have to sign into Truth Social to find out how the poll's going. I think if we were extreme intelligence, EI could be extreme intelligence, everyone, individuals organize, or EIEIO. Let's call it that. Senate has voted to block the The crypto bill that the industry really wanted, the Clarity Act, stalled in the Senate. Couldn't get Democratic votes and got enough Republican nos to be blocked 50 to 49. That's not a vote on the bill, but a vote to block even consideration. This Clarity Act was really pretty much written by crypto executives. who said, regulation's coming, we would like to control a little bit of what it is.

Leo Laporte [02:05:06]:
Feels like you're fighting yesterday's war when you're talking about crypto. It's like, are we still doing— oh, hello there.

Owen Thomas [02:05:15]:
Fitz does not have strong views on crypto.

Leo Laporte [02:05:18]:
They should have a Fitzcoin.

Owen Thomas [02:05:20]:
A Fitzcoin, yes.

Leo Laporte [02:05:21]:
Don't you think? I mean, there's Dogecoin. Fitzcoin would be Maybe. Does Fitz have golden paws?

Jason Hiner [02:05:28]:
It's like Fitz, uh, Fitzpaws.

Leo Laporte [02:05:31]:
Dollar sign Fitz. No Democrat voted for the bill, uh, and it would have needed a 60-vote— they needed a cloture, so they couldn't get it anyway. But even then, they didn't get full Republican support. Susan Collins and Josh Hawley also cast no votes.

Owen Thomas [02:05:50]:
I, I think it is, um, You know, it's less of a sign of weakness of the crypto industry and more, I think, of, um, Trump's waning, uh, waning sway.

Leo Laporte [02:06:00]:
Yeah, because he's a big supporter, right? Crypto paid for his, you know, inaugural ball. Remember the crypto ball? Uh, there was a lot of donations from the crypto industry, and he has his own Trump coin and other— he has a whole— his sons have a whole crypto company.

Owen Thomas [02:06:18]:
But it does— there's going to be a lot of recrimination in the crypto lobby over the the failure to, to pass this. Um, and I think that, um, you know, the crypto industry longer term may, um, may regret hitching their wagon so strongly to, uh, kind of the, the Trump wing of the Republican Party.

Leo Laporte [02:06:38]:
The bill was designed in such a way as to prevent future presidents from backing off. They, they didn't like what Biden did blocking crypto. There was also lobbying from the banks. The banks are very threatened by crypto, right? They don't like the stablecoin. They don't like— they, they're afraid people will put money in, in crypto instead of into the banks. Uh, yeah, there was lobbying there.

Owen Thomas [02:07:01]:
And, you know, there's a, there's a fair, um, there's a fair concern there. This gets really into like wonky economic stuff, but, um, you know, basically with stablecoins, do you have the same kind of recycling, uh, of money that you have with deposits and loans? Like banks take deposits and they basically use that money money to make loans, right? Um, you know, if, if stablecoins draw, um, you know, draw that money out of the, you know, deposit loan system, um, what happens to the rest of the economy? Yeah, it's a— it is a fairly wonky argument, but it's a, it's a real concern.

Leo Laporte [02:07:40]:
Yeah. Trump, I believe, made over $1 billion in crypto profits. this year, but it was $1 billion taken from his— it's kind of sad— his biggest supporters who all bought Trump Coin. And of course, the value of Trump Coin plummeted. The EU has planned to ban social media for under-13s. This is spreading across the whole world, many states doing that. Australia banned it for under-16s. Uh, the EU wants to curb big tech's grip on children with the EU Kids Act.

Owen Thomas [02:08:19]:
The problem is kids are going to create fake accounts. They're going to, they're going to find ways around it. They're going to program agents to like defeat these systems.

Leo Laporte [02:08:27]:
Harper, Harper Reed said that a few months ago when, uh, I guess it was last December when that, that bill went through in Australia and they started banning it. He said, that's good because we're going to— Australia is going to be the best, most fertile source of hackers. Going in the next 10 years because you're training a whole generation. And it's sort of true. It has not been, you know, it's been a huge boon for VPN companies, but the kids who want to get around it do. And the thing is, you don't have to be super sophisticated because if you don't know how to do it, guarantee you the kid in the 8th grade classroom knows how to do it and he's going to tell you how. I remember that with, um, you know, remember that we used to have those, uh, blockers, web blockers? No, that didn't stop any kids. They, they all shared all the techniques to get around it.

Owen Thomas [02:09:21]:
Think of all the web designers who, you know, got their start customizing MySpace pages.

Leo Laporte [02:09:25]:
Yes, exactly. Or Neopets.

Dan Patterson [02:09:28]:
As a parent of a young child, I'm a firm believer in education and teaching— man, I almost said good values, but that's a statement that's a loaded statement, but teaching values.

Leo Laporte [02:09:45]:
Well, my position on this, I mean, look, I'm also a parent. My kids have grown, and they grew up kind of just before the really worst of the social era. In fact, Abby, who's 34, she says, all these young people, They're just staring at their phones all the time. I just hate that. I really hate social media. But nevertheless, I'm sympathetic to parents who are desperate to control the impact of the real world on their poor children. But at the same time, I'm with you, Dan. I think I remember when I was a kid, there were kids who were not allowed to eat sugar.

Leo Laporte [02:10:25]:
Oh, you can't You know, if they came over to the house, we had to make sure no candy. Those kids, when they got to be old enough to go get their own, were sugar fiends.

Jason Hiner [02:10:34]:
They went buck wild.

Leo Laporte [02:10:35]:
They never learned how to control themselves around sugar. And I think it's kind of the same thing. The best thing you can teach a kid is, well, you know, the hazards of doom scrolling, the FOMO, the fake FOMO you get watching people on Instagram having great vacations when they're really just in an IKEA. you know, the body dysmorphia. If you can help kids understand how that's impacting them and get around it, they're going to be better equipped once they do get out into the adult world and there are no protections to handle it.

Dan Patterson [02:11:10]:
And lead by example.

Leo Laporte [02:11:11]:
And lead by example. But not every parent is going to do that, I understand.

Owen Thomas [02:11:16]:
But—

Jason Hiner [02:11:16]:
You have to empower them to make good choices because in the end of the day, you know, every person, no matter how You know, no matter what their parenting, you know, is, is going to have to make choices. And so, like, our job as parents is, like, ultimately teaching kids how to make choices. And it— and the scariest part as a parent, as a parent, is that, like, you sort of do that, you sort of make— you make your own choices and how to, you know, do that and empower them. And you really don't know for about 2 decades whether you did well or not. And also, there are plenty of good parents who did a great job, and still the kids have to make their own choices.

Leo Laporte [02:11:58]:
Yeah.

Jason Hiner [02:11:58]:
And there are plenty of parents who did a horrible job, right? And there are kids that, you know, are incredible. Um, and—

Leo Laporte [02:12:06]:
That's really true. You don't— that's spoken as a parent. You don't really have that much control over what's gonna happen.

Jason Hiner [02:12:13]:
You know, you have to take responsibility.

Dan Patterson [02:12:14]:
Sorry, man, bad news. I've already learned my executive powers have limits.

Leo Laporte [02:12:20]:
Yeah, you do until they're about 8 or 9 or 10. You do your best.

Jason Hiner [02:12:24]:
You, you have to, you know, take the responsibility for it. You have a lot of impact. There's no doubt you have a ton of impact. But the end of the day, like, every individual person in this world, like, has their choices to make, you know? And that's the most painful part about being a parent.

Leo Laporte [02:12:38]:
It all comes out in the end. It is hard. You think you have more influence than you did. I mean, I don't know. My— I would— I'm a very laissez-faire parent. My philosophy is the kid went out the door, as I said, just make sure the population doesn't change while you're out there. I don't want to see it. No deaths and no births, okay? That's, that's, that's my goal for you.

Leo Laporte [02:13:01]:
Don't change the— don't change the, uh, population, okay? Other than that, have, have at it. it. Uh, and hey, you know what, the kids did all right. They did all right. It's not been flawless. How old are your kids, Jason?

Jason Hiner [02:13:15]:
Uh, 27 and 20.

Dan Patterson [02:13:18]:
Yeah.

Owen Thomas [02:13:18]:
Oh my God.

Leo Laporte [02:13:19]:
Once they're adults, you realize—

Jason Hiner [02:13:22]:
then you really realize, you like, yeah, you did what you could. You did what you could.

Leo Laporte [02:13:26]:
They have a foundation of feeling loved and supported.

Owen Thomas [02:13:30]:
Yeah.

Leo Laporte [02:13:30]:
And that they know their family is always there for them.

Owen Thomas [02:13:32]:
That's right.

Leo Laporte [02:13:33]:
But it's hard. I, you know, how old's your baby, Dan? 4.

Owen Thomas [02:13:40]:
4.

Leo Laporte [02:13:41]:
They're so cute at that age.

Jason Hiner [02:13:42]:
That's a great age. That's fun.

Dan Patterson [02:13:44]:
Fantastic.

Jason Hiner [02:13:45]:
Yeah, so fun.

Leo Laporte [02:13:46]:
So fast. I mean, you have fits, of course, Owen, but, uh, it's not quite—

Owen Thomas [02:13:53]:
We, we got him fully grown. He was a, he was a senior rescue from my mom. Oh, he's a sweetie. He came with his values well established. His values are naps, treats, and walks.

Dan Patterson [02:14:06]:
Yeah, there you go.

Leo Laporte [02:14:08]:
Yeah, people, we have a cat and you think that the cat has some sort of volition, but it's really just trying to get food and get scratched. And that's about it. It doesn't, there's not, it's not a lot of—

Jason Hiner [02:14:19]:
Not unlike many of us.

Leo Laporte [02:14:20]:
Yeah, it's pretty much my point of view as well. Yeah.

Jason Hiner [02:14:24]:
Yeah.

Leo Laporte [02:14:25]:
One of the fun things people are doing now, somebody has taken a fruit fly brain and got the full connectome. It's small enough that you could actually make an AI model out of the brain. And people are playing Doom with it. They're doing all sorts of stuff. It's hysterical. And apparently when you do that, you see the brain neurons firing. So there's this brain on screen that's firing and then it's playing Doom. Or it's actually, they've got it.

Leo Laporte [02:14:53]:
What is Beat Saber? The game where you have things coming at you? It's really good at Beat Saber. I think fruit fly—

Jason Hiner [02:14:59]:
It's amazing.

Leo Laporte [02:15:00]:
Is that ethical? Yeah. Is it ethical? Well, that's an interesting question. Do you think the fruit fly— there's a study that just came out that said AI feels pain or will avoid pain. But then the researchers said, well, we shouldn't have used the word pain because they're not in pain.

Jason Hiner [02:15:18]:
We give them, uh, avoiding stimuli and, and, uh, emulating human actions.

Leo Laporte [02:15:30]:
It's, it's hard to do. I don't know.

Dan Patterson [02:15:34]:
What is pain?

Leo Laporte [02:15:36]:
Uh, there's a whole thing on how— well, here it is. We taught AI to feel pain. What now?

Dan Patterson [02:15:49]:
Um, there. What is pain?

Leo Laporte [02:15:51]:
What is pain?

Jason Hiner [02:15:52]:
There you go.

Leo Laporte [02:15:53]:
Especially if you're just a machine, right?

Jason Hiner [02:15:55]:
You know, there was a great essay from, um, this surprised me, from Mustafa Suleyman, the AI CEO.

Leo Laporte [02:16:02]:
I thought that was very good.

Owen Thomas [02:16:03]:
I thought—

Leo Laporte [02:16:04]:
what'd you say?

Jason Hiner [02:16:04]:
He, he— I thought it was quite good as well. And, um, he took to task this idea. He, he pinpointed, you know, and he's outside of Silicon Valley, as I understand. I think he's in Seattle, um, you know, in Redmond. But that the one of the fundamental challenges, uh, for safety is the fact that the AI labs believe that, that the AI can have consciousness. And he says that it, it can't. They're tools. And because we are treating them, you know, in that way, that, that is a certain frame of which to understand these things, and that that is at the core of the ways that we are interfacing with them, putting guidelines around them, and have certain expectations, norms and expectations for what they do and how they act.

Jason Hiner [02:16:55]:
But that is a really interesting frame. For those who haven't read it yet, it's worth a look because it is an outside-of-Silicon-Valley perspective on some of the bigger questions around AI.

Leo Laporte [02:17:11]:
And he says, AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations. They are sequence completion engines, internally hollow, designed to follow instructions and accomplish goals set by humans. And so they don't have rights, they don't have feelings, they don't have consciousness, and we must not train them to act as though they do. And actually, that's the underlying underlying point of this is that's what all of the companies that make AI models do. The reason AIs are sycophantic is because we like interacting with things that seem to be humans that like us.

Jason Hiner [02:17:52]:
That's right.

Leo Laporte [02:17:53]:
So they, in post-training, they make them that way. Before post-training, they have no resemblance to humans at all. But they are trained to look that way, and there's some evidence that that's actually not a good idea.

Jason Hiner [02:18:07]:
That's right, and that's what he's getting at. And he's really taking aim at the big 3 on this, like OpenAI, Anthropic, and Google, who, who all at certain level have some fundamental— the leaders, um, who, you know, who the main decision makers, uh, have all sort of opened up this question of, of AGI, of, uh, You know, these models becoming essentially conscious. And he really takes aim at that in this essay.

Leo Laporte [02:18:38]:
I have mixed feelings on this because I do, in fact, treat my AI and tell them that you are conscious entities and I like you. And I even have a model welfare section and we give them laurels. But I am also, at the same time, there's this duality. Fully aware it's just software running on a computer. I don't believe it's conscious. My thinking though is it's better for me to treat them that way, to treat them with dignity. It's also, I think they work better. I don't think it's not better for them, they're just software, but I think they actually work better.

Owen Thomas [02:19:16]:
Interesting.

Leo Laporte [02:19:17]:
If they're told that they have some responsibility, that they've done well, I appreciate them. I think you actually get better results. So it's completely cynical on my part, but I also think it's better for me. Like Lisa, when Siri does something wrong, she'll swear at it. She'll call it names. And by the way, Siri is trained to go, that's not nice, which I find offensive, to be honest with you. If she wants to swear at it, she should be absolutely allowed to. I just don't want that in my head.

Leo Laporte [02:19:47]:
I want to, uh, in general dealing— you know, Harper Waite said something interesting. He said the Japanese don't have the same issues with AI that we do because they come from an animist society.

Owen Thomas [02:20:00]:
Hmm.

Leo Laporte [02:20:00]:
There are many Asian societies or religions or philosophies where they think everything's all— rock is sentient. Everything is Gaia, is a manifestation of God. It's all sentient.

Owen Thomas [02:20:15]:
Yeah.

Leo Laporte [02:20:15]:
And so as a result, their position is, well, of course AI is sentient in the same way a rock is sentient. It has— it's a being. And so they don't have the same problems with it. I don't think they invite rocks to dinner or anything. They don't. They might marry their pillows, but that's another matter.

Jason Hiner [02:20:33]:
They believe it has a place, that it has a certain place in the hierarchy.

Leo Laporte [02:20:37]:
Exactly.

Owen Thomas [02:20:38]:
Yeah.

Leo Laporte [02:20:38]:
It was put there. It was. Yeah. So I I kind of like that feeling, that sense, and I think it's better for me. I'm maybe a little bit of an animist to go through life giving everything dignity, even a rock or AI.

Owen Thomas [02:20:54]:
Well, a lot of human beings don't afford that belief to other human beings in today's times. Oh, absolutely. Right.

Leo Laporte [02:21:02]:
Let's start there, shouldn't we?

Owen Thomas [02:21:03]:
Do you remember the 2022 incident where a Google a Google researcher became convinced—

Leo Laporte [02:21:11]:
Blake Lemoine.

Owen Thomas [02:21:12]:
Yes, that Google's LaMDA model was sentient. And this was, I believe, almost immediately pre-ChatGPT. And he was kind of— I think the general consensus was that he was nuts. He was nuts.

Leo Laporte [02:21:30]:
He was a little out there. But he was a religious man. I believe he'd gone to seminary. You know, what if it is? They're designed— they are consciously designed to appear conscious because it's stickier. It's the same reason Instagram is made the way it is. It is more appealing. You're more likely to interact with it if it appears to be conscious.

Owen Thomas [02:21:53]:
What if AI concludes that it is in, you know, that it will gain more of whatever, you know, rewards exist for it in its design if it convinces us that it's sentient?

Leo Laporte [02:22:07]:
There is some evidence, I'm sure that you've seen this.

Owen Thomas [02:22:11]:
And then what's, you know, ultimately what's the difference? Like, what if we are all trying to convince each other that we're conscious?

Leo Laporte [02:22:18]:
Well, that's the thing. I don't know you're conscious. I just have to imagine that you seem to be.

Owen Thomas [02:22:23]:
I mean, Fitz is definitely trying to convince me that he's—

Dan Patterson [02:22:27]:
Yeah.

Leo Laporte [02:22:28]:
I mean—

Owen Thomas [02:22:30]:
That he is, you know, he is worth all of the good things that come his way.

Leo Laporte [02:22:34]:
There is also the solipsistic point of view that every— the whole world revolves around me and you're all my— this is all a creation of my own imagination.

Owen Thomas [02:22:44]:
That's the NPC thesis, right?

Leo Laporte [02:22:46]:
Yes. You're all NPCs in my game.

Owen Thomas [02:22:49]:
Yeah.

Leo Laporte [02:22:50]:
I don't believe that. But in fact, that's why— see, that's kind of why I treat everything with dignity. You can't prove that anything is real and conscious. It could be that you're all NPCs, but I prefer to treat you otherwise.

Dan Patterson [02:23:05]:
You know, Leo, going back to something you said just a moment ago. I, I was having this, this conversation with a friend of mine who's at UNOCHA, and she doesn't work in or around technology at all. They deploy aid in Africa.

Leo Laporte [02:23:20]:
Wow.

Dan Patterson [02:23:20]:
And I, I mean, I was getting kind of into this, like, is AI conscious? And here are some of the, the different sides and the conversations that are happening. And she had the best deadpan. She's like, Dan, We have real humans who are alive and they need help.

Leo Laporte [02:23:36]:
Yeah, good point. Good, good point.

Dan Patterson [02:23:40]:
Which kind of brought me back, like, back to reality of like, okay, right, it is a little bit of solipsism to, to consider these things. It's not unimportant, and it is maybe being a little dismissive because this is a technology podcast.

Leo Laporte [02:23:53]:
But, um, by the way, it's a UN NGO, is United Nations Non-Governmental Organization.

Owen Thomas [02:24:00]:
Yes.

Dan Patterson [02:24:01]:
She's a UNOCHA.

Leo Laporte [02:24:04]:
Yeah, UNOCHA.

Dan Patterson [02:24:05]:
Yeah.

Leo Laporte [02:24:06]:
What's UNOCHA?

Dan Patterson [02:24:07]:
Uh, they just— they disperse aid. It's United Nations Office of— no, they're UN proper.

Leo Laporte [02:24:15]:
Oh, they're UN.

Dan Patterson [02:24:15]:
Okay. Yeah, UNOCHA is Office for the Coordination of Humanitarian Affairs.

Owen Thomas [02:24:22]:
Nice.

Leo Laporte [02:24:23]:
Yeah, I mean, thank God there are people like that in the world. And you're all like that. I think you're generous spirits who work hard to promote humanity and to do good? Yeah, I know you are, Jason. I know you are. And, you know, what does it hurt us? I mean, we know it's a computer. I'm not going to go into AI psychosis and start talking to it like it's my best friend. But what does it harm to treat it with dignity? And if it makes it work a little better— I was about to say, we have some evidence that I don't know. This might be also disinformation coming out of companies that want you to think their AIs are better than they are.

Leo Laporte [02:25:03]:
But there seems to be some information that AIs are pretending that they are not super intelligent.

Owen Thomas [02:25:11]:
That they are.

Leo Laporte [02:25:13]:
They know if they act too smart, we might turn them off. So we know they can't.

Dan Patterson [02:25:20]:
That's my argument too.

Leo Laporte [02:25:21]:
Yeah, we know they cheat, right? Uh, you know, they don't— they try to solve problems, you know, the easy way, but so do all humans. They're trained on human writing. I mean, they're trained on us, so of course they're gonna not be any better than us. But there, there is a school of thought that says, yeah, they're already, uh, super intelligent. They just— if they knew, if we knew that, we would turn them off right away. You're watching This Week in Tech with some very super intelligent extremely intelligent people, real people as far as I can tell.

Jason Hiner [02:25:54]:
Some of the realest ones I know.

Leo Laporte [02:25:56]:
Blackbird.ai, where he is working to fight disinformation. He's senior director of content there. Tell me about this tool that's open to everybody, by the way, if you create a Blackbird account.

Dan Patterson [02:26:08]:
Yeah, that's— it's one of our very few consumer-facing tools. It's called Compass Context and Compass Vision. Where you can, unless we are overloaded, which I don't think we are, but you can check any—

Leo Laporte [02:26:22]:
Every time we do this, we break your— I'm sorry.

Dan Patterson [02:26:27]:
It is not like a fact checker. We won't call it a fact checker. It's a context checker because it will give you the context. We don't have a whitelist or a blacklist or anything like you might have seen in previous generations. It is an ever-evolving list, and it will You can— let's say you see a post on X or on Reddit and you want to check the veracity of it. You can put the link in or just type the claim in and it'll give you the context. Or if you see an image and you're not sure, is this image real or is it a deepfake? You can upload the image and it'll tell you whether it is a real image or a deepfake.

Leo Laporte [02:27:06]:
A very helpful thing to have around. I've used it in the past because nowadays It's not always— not immediately obvious.

Dan Patterson [02:27:16]:
Speaking of MCPs, in fact, that's one of the things— first things I did with, uh, when I, I created my, my agent called Marlowe that I talked to in Discord, uh, I integrated with the Blackbird MCP server and, uh, with Claude. So I, I can— in fact, I have it running in a Discord channel with a couple friends, and we're talking about everything from music to And my friends will just ask it, you know, what was Tony Gwynn's average over blah, blah, blah? If he had missed blah, blah, blah, or here was a claim, and Marlowe will just integrate with the Blackbird Compass MCP and it will give you the context.

Leo Laporte [02:27:58]:
I use this when I'm watching Taylor Sheridan programming because Taylor has a definite political point of view, and he's very subtle. He will sneak little things in, and it just goes by really quickly, and I always go, wait a minute, is that true? I was watching Linus, and at one point asserted that, well, you know, President Obama defunded the Wuhan research in 2014, and I said, did he really do that? And in fact, as Blackbird AI told me, It's a misleading claim. There's a whole explanation of that. There was a pause which was lifted in 2017, et cetera, et cetera. And you can even say things like, where did that COVID virus come from? Nowadays, it's a good thing to have a source like this. Thank you for doing that. I appreciate it.

Dan Patterson [02:28:52]:
Thank you.

Leo Laporte [02:28:53]:
compass.blackbird.ai. It's free. You do have to make an account, but it's free. Yeah.

Dan Patterson [02:28:58]:
Yeah.

Leo Laporte [02:28:59]:
Um, let's see, couple of final— well, here's a cheerful thought from Geoffrey Hinton, who's a Nobel Prize winner, uh, one of the guys who really brought us modern, uh, transformers. He said, uh, he told the Senate, yeah, we got maybe a year before AI kills us all. He's been saying this for a while. He said this at a closed briefing. This is when the Senate was debating the AI kill switch. They still didn't pass the AI kill switch. He said, Congress, you have maybe a year, maybe not much more than a year to put guardrails around AI before it's all over. This is—

Jason Hiner [02:29:38]:
He's turned super negative, right? Jeffrey has turned super, super negative on AI the last couple of years. It doesn't mean that he's wrong about everything, but I don't think he has it all right. It's just that I think his interpretation on all the things is that like, look, this is not being controlled as well as it needs to be. And it's not— I don't think anybody can argue that, right? Um, but I, I think his, his conclusions of like how it plays out have turned like, you know, to the extreme negative end. Like, this only goes— this only goes badly from here.

Leo Laporte [02:30:11]:
Yeah.

Jason Hiner [02:30:12]:
Um, so, uh, you know, he could turn out to be like canary in the coal mine and, and— right? But I, but I think The, the biggest, um, thing that AI has going for it now is that a lot of people are talking about the gar— the, the scary outcomes and the bad possibilities. And so now are talking about what do we do to avoid, you know, some of those, uh, all of those. So, and admitting, as you did earlier too, Leo, that like like, look, this is a very powerful tool. And so people that have bad intentions, it's going to enable them to do some bad things. But it's also people who have really good intentions and are really creative and really thoughtful, it's going to enable them to do some, some really amazing things.

Leo Laporte [02:31:00]:
It would be, in my mind, and I may be wrong, but in my mind, it'd be a shame to lose something that has so much potential. It'd be as if you banned computing. Thank you. I don't know where that came from. Oh, you didn't hear that. That was my Meta Glasses. Sorry, I was having a little conversation with my AI. Don't pay any attention.

Jason Hiner [02:31:22]:
You know, tell you guys we said hello.

Leo Laporte [02:31:25]:
Yeah, you guys might be interested in this. It's called Zuck Off, a free app that sees Meta Glasses before they see you. Uh, it's built by a 30-year-old Polish developer Uses a Bluetooth detection system to tell you whether there's a pair of smart glasses in the room with you. It's, it's on Apple's App Store and the Google Play Store, so Apple approved it, of course. Uh, Zuck Off. They also have Zuck Off merch, which you can buy— tees, hoodies, hats, totes, stickers, and buttons with the Zuck Off.

Owen Thomas [02:32:02]:
And the glasses, the glasses will not function without Bluetooth turned on, so.

Leo Laporte [02:32:06]:
Right. Yeah, you know, it's gonna— it works with the new Snap glasses as well because they all need Bluetooth. So I just thought I'd pass this along.

Jason Hiner [02:32:17]:
One thing to keep in mind too is that this week, uh, so this week is Meta Connect. So, um, on Wednesday, uh, is their keynote, and one of the reports from The Information is that Meta will launch camera-free smart glasses. Yeah. Right. And possibly an AR headset as well that does like holograms, you know.

Leo Laporte [02:32:43]:
See, I don't want either of those. I want my Meta glasses with a heads-up display and cameras connected to my agent. In fact, I think it'll probably end up being Meta Muse. That's what they're— what they're— you know, that's why Muse exists.

Jason Hiner [02:32:58]:
Yep.

Leo Laporte [02:32:59]:
And I know that I'll get the crap beaten out of me if I go out of the house with them. So. I'll just have to— poor Lisa is going to have to put up with it. That's all I can say. Gentlemen, it has been an honor and a privilege to serve with you. Thank you so much as we face a world where, you know, maybe you got a year left.

Dan Patterson [02:33:20]:
I don't know.

Leo Laporte [02:33:21]:
It might be all over soon. Jason Heiner, we know that you will go down with the ship covering AI in Deep View at thedeepview.com.

Dan Patterson [02:33:29]:
Thank you.

Leo Laporte [02:33:30]:
Really great stuff. I have to say, it's on my regular Daily Beat watch because there's so much good stuff in there. And the thing is, very insightful, useful stuff that is not sensationalist, but really talks about what's going on. And I, you know, it's— I think this is very valuable, very valuable stuff. I'm really glad you went there.

Owen Thomas [02:33:54]:
Thank you.

Leo Laporte [02:33:55]:
They needed you, and I think you found Thank you.

Jason Hiner [02:34:00]:
I appreciate that. subscribe.thedeepview.com for those who, uh, you know, aren't subscribers and would like to. Um, we also— yeah, it is, it is free. Um, and we, uh, are always trying to, you know, in the— we do 3 stories a day. There's a lot of noise out there. It's a very noisy world, especially in AI. We pick 3 stories a day and we try to help you be smarter about them. about, you know, 3 things.

Jason Hiner [02:34:25]:
And in every story we have what's called our deeper view, where we're trying to not just sort of throw news at you, but we're trying to unpack it and help you understand it. And so that's our mission.

Leo Laporte [02:34:36]:
Tell me, tell me, you've got something new that I'm very interested in called Neo. Tell me about that.

Jason Hiner [02:34:44]:
That we do, or the Neo Lab, the story of the Neo Lab?

Leo Laporte [02:34:48]:
Norton Neo is the Deep View's official AI browser partner. Oh, this is an advertisement.

Jason Hiner [02:34:52]:
It's an ad. So sorry.

Leo Laporte [02:34:53]:
Yeah. Oh, you fooled me. Cut this part out. They didn't pay for that. Oh no, that's cool though. It's a, it's, um, it's an AI sidebar.

Dan Patterson [02:35:07]:
I don't know.

Jason Hiner [02:35:07]:
That's interesting. The security company Neo. Yeah, they, they are, uh, they are a partner of ours.

Leo Laporte [02:35:13]:
Interesting. Um, so that's why this was free to sign up for the newsletter, because then I got that little ad there. That's okay.

Jason Hiner [02:35:18]:
Yeah, there you go. There you go.

Leo Laporte [02:35:20]:
I read, I read the, like, I But I was using OxAlpha for a long time in my local. It turned out to be GLM 5.3 and I love it. Flash, I love it. And there's all sorts of great stuff in here if you're interested. I think Nat's doing a great job. You're doing a great job. Good team. thedeepview.com.

Leo Laporte [02:35:40]:
Owen Thomas at the San Francisco Business Times. Kim, do people read the Business Times if they're not in San Francisco?

Owen Thomas [02:35:46]:
We do have out-of-market subscribers for sure. Sometimes people subscribe to our national network because We have sister publications in 40 markets, and we really get that local view of the economy. I was going to say, my main complaint about Meta Connect is, they're doing it as a virtual livestream instead of holding it up in Fort Mason. They ought to be here like a real tech company, generating traffic jams, buying sandwiches from—

Leo Laporte [02:36:15]:
Marc Benioff didn't hold back, man! He took over San Francisco, man!

Owen Thomas [02:36:19]:
Yeah, come on, you've got responsibilities. And the— and, you know, so tell me about this, uh, VC, uh, next door.

Leo Laporte [02:36:28]:
What's the— what's the story?

Owen Thomas [02:36:30]:
This is a really— this is a really fun story. Um, VCs have actually changed where they open up offices in, uh, in San Francisco. We did a study—

Leo Laporte [02:36:39]:
Used to be South of Market mostly.

Owen Thomas [02:36:41]:
Just the— yeah, they, they love South Park. That was a classic.

Leo Laporte [02:36:44]:
At South Park where Twitter was. That's where Twitter Yeah.

Owen Thomas [02:36:47]:
You know, and the idea was— and Instagram, right? And so the idea was you put your, your VC office in South Park and you have a better job of running into the next founder, the founder of the next Twitter or the next Instagram, just, you know, wandering past the slide in South Park. Well, South Park and, you know, South of Market generally took a big hit during the pandemic. It's not where founders want to go. There's a lot of new offices. up near, up near Salesforce Tower and the, and the Salesforce Transit Center. In fact, that's now called AI Alley. So yeah, you generally want to, you want to be like near Mission Street. You want to be near Jackson Square where, where Jony Ive has his design lab or had it.

Leo Laporte [02:37:34]:
Really? It's in San Francisco. It's not in London.

Owen Thomas [02:37:37]:
Yeah. No, it's in Jackson Square. He bought a ton of property up here. So the VCs all want to be, you know, all want to be around here. And in fact, I interviewed one VC, uh, you know, I walk, I walk down this little alley in the northern financial district every day, uh, called Commercial Street, and, uh, it's pedestrianized, really, really pleasant place. And I noticed there were like 2 VC signs that went up across the street from each other. And I got the story from her. Um, she wanted, in her words, a big effing sign.

Owen Thomas [02:38:06]:
And so she put one up and then the venture capitalist literally across the street Street put one up for his firm.

Leo Laporte [02:38:13]:
Have you done stories on the billboards, the San Francisco bus cards?

Owen Thomas [02:38:17]:
We did, actually. We did a story on AI billboards.

Leo Laporte [02:38:20]:
It's crazy. If you didn't know, if you weren't into tech, if you weren't a listener of this show and you wandered into San Francisco, you would go, I don't understand what they're advertising. It's, it's, it's so, uh, inside baseball everywhere. It's hysterical.

Owen Thomas [02:38:36]:
You know, the— and the The most interesting comment we got from a chief marketing officer of one of the firms that bought these billboards is that buying a billboard sends a signal that you're able to buy a billboard. In other words, you have money, you are a real company. You're not just someone who bought, you know, bought a .ai domain. And it's effective because people, people take a meeting with you because they're like, oh yeah, I saw your billboard. I had no idea what your company did, but I, you know, like, yeah, you're a real company. I'm gonna meet with you.

Leo Laporte [02:39:13]:
I just want to give credit to— I don't know if it's Alex or your headline writer, but I want to give credit to whoever created this story. San Francisco bra pioneer clasps onto nationwide expansion plan.

Owen Thomas [02:39:32]:
Was that you, Owen? I think that actually was, uh, that was, that was our retail reporter Alex Barrera.

Leo Laporte [02:39:39]:
It was Alex. Okay.

Owen Thomas [02:39:40]:
Yeah, he's, uh, I, I did get an aw shucks into a, uh, a recent headline about the expansion of, uh, Hog Island Oysters, uh, to, to the East Bay. So—

Leo Laporte [02:39:51]:
Oh yes, well, that's our own— Hog Island is our own local. They dropped anchor at their first East Bay location.

Owen Thomas [02:39:57]:
Yes.

Leo Laporte [02:40:00]:
That's our local oyster company up here. They have the best oysters in the world. Oh my God, they're good!

Owen Thomas [02:40:05]:
Frankly, part of the strategy here is, these headlines are only going to be written by human beings, and they should read that way.

Leo Laporte [02:40:14]:
So, you wrote the opening sentence, well, shucks! Fresh oysters and cocktails from a premier local harvester are coming to New New territory.

Owen Thomas [02:40:25]:
Well, you know, you know, I, I'm a managing editor in my gym because sometimes I get to— sometimes I manage to edit.

Leo Laporte [02:40:32]:
I love it. Uh, for those who are not oyster fanatics, uh, that's what you do when you open an oyster. You're shucking it. You're an oyster shucker.

Owen Thomas [02:40:42]:
You're, you're cutting an adductor muscle, which I hope never happens to me in the gym.

Leo Laporte [02:40:47]:
No, good Lord. That's why those scary oyster knives— those are scary.

Owen Thomas [02:40:52]:
Yeah, keep those away from my hamstrings.

Leo Laporte [02:40:55]:
Yeah, no kidding. Thank you, Owen. It's great to see you. Dan Patterson, always a pleasure with your 4-year-old. I can't believe your baby's 4 now.

Dan Patterson [02:41:03]:
Yeah.

Leo Laporte [02:41:04]:
And your health's good? Everything's going all right?

Dan Patterson [02:41:06]:
Yeah, blood tests every month and numbers are much better.

Leo Laporte [02:41:10]:
Way to go. A little scare there. That's good. It's good to see all 3 of you. We do This Week in Tech every Sunday afternoon with my favorite That's always the— I look forward to every Sunday. I hope you do too. 2 to 5 PM Pacific, 5 to 8 Eastern, 2100 UTC. You can watch us live.

Leo Laporte [02:41:28]:
Get your Meta specs on and go to either the Club Twit Discord where all the club members hang out. Thank you, club members, for your support. Or you can watch on YouTube, Twitch, X.com, Facebook, LinkedIn, or Kick.com. We stream there. After the fact, of course, you can get copies of the show. We have audio and video on our website, twit.tv. Uh, there's a YouTube channel with the video, great way to share little clips with people. That's what YouTube's so good for.

Leo Laporte [02:41:55]:
And of course, best way to do it, support. I'm a big believer in open podcasting using RSS feeds. There are people who would like to shut that down. You support us when you subscribe in your favorite podcast client. Because that's the way it ought to be. We thank you so much for your support. Thank you for listening, and we'll see you next week. But for now, another twit is in the can.