IM 868: Happy Hamburgers Towing Timmy To The Sea - Can You Really Own Your AI?

23:21And then that being paired with the downstream ramification of those data centers gobbling up all that silicon, of the machines, the hardware they actually can own, becoming harder and harder to own just purely for price reasons. The memory price is going crazy, CPU prices, storage prices, and those two things playing together against, of course, this now multi-decade backdrop of computers becoming more locked down, both from a hardware perspective and from a software perspective. it felt very clear to me that the way that we've thought about computing over the course of decades is not necessarily going to hold.

24:02That if we fast forward a couple of years, having a locked down dumb terminal that's your window into someone else's computer that they own for you in the cloud that you are subscribed to may actually become the default. And so that was kind of the genesis of the statement of personal computing as we know it is dead. It may actually be dead. And then for us as a company, as a framework, for our entire mission, our entire purpose of existence is consumer rights, giving you ownership, giving you power over your own computing. What does that mean for us? What do we need to do as a company in this environment?

24:37And for us, it's really doubling down on you should be able, if you want to own your computer, you should be able to own your computer. We want to make sure you can own your computer all the way through the software stack, all the way through the hardware stack. and even in a world where your interface is AI, we want to make sure that you can run that AI and have that be your AI rather than something that you're renting from the cloud.

24:58Nirav Patel:You're right. The industry is asking you to own nothing and be happy. That's right. A little throwback to Steve Jobs. Computers are no longer a bicycle for the mind. That was his famous phrase. They're becoming, this is so sad, the self-driving car that takes you directly to the destination. I think for many users, that's not sad. That's exactly what they want. That's right. And that was part of this manifesto is that you could look at that and say, like, actually, that's kind of nice that, like, I don't have to think about it. I don't have to worry about it. It's just going to work for me. But at the same time, it goes back to this walled garden philosophy.

25:32Like, you can be happy in the walled garden, but if you know about the existence of the world outside of that garden, you might feel very limited and constrained in there. Yeah.

25:41Nirav Patel:Well, and that's exactly why I bought the desktop is because I wanted to run my own AI models locally on a machine that was capable, you know, knowing that they're not quite as good as the Frontier models yet. They'll probably always be a little bit behind, but they're good enough. What are people using now on frameworks, desktops as their local models, if they are using local models? Yeah, we see QuenCoder being very popular. The new Gemma models are actually very strong from Google. So we'll see what Meta comes up with. I think they're not going to give up on trying to be competitive in that space.

26:15LLAM has fallen a bit behind, but I think Meta is going to have something big there coming. But a lot of the interesting models are actually coming from the big labs in China, which has been a very interesting dynamic to see. Yeah, I have a GLM subscription from Z.ai, and it's actually pretty surprisingly good. I feel like maybe because they distilled Opus just a little bit. But nevertheless, they're pretty good. What do you use personally? Do you have a local AI running? Yeah, I actually have two boxes I keep in my desk. One running Windows, one running. Currently Fedora, although switch between different distros.

26:49And they run different. Like I open Claw on one. I think I have Hermes running on the other. And then usually I use something like QuenCoder just to play with it. Mostly for just general toy box, sandbox type uses. I don't yet. I don't plug in our framework company data in.

27:05Nirav Patel:Good. You're smart. Because they're going straight to China. Right. actually not if you're running it locally that's the whole that's that's true that's one thing one piece of news that came out uh last week or the week before is that third parties are also making there's a third party motherboard uh for the framework yeah we've got a couple now we've got uh actually they're kind of related companies company called deep computing that's making risk 5 main boards oh that's cool yeah very cool to see we're now two generations of risk 5 and also a company called meta computing that's done an arm-based main board which is also great to see yeah very exciting uh paris i i know you had to come in late is there anything you you wanted to ask about or is this i've kind of monopolized this i was going to say i feel like you guys really i mean i can't compete with the expert in this like leo given his extensive use the okay i'm gonna buy it i just can't do it i want to get in my august i'm sorry just keep talking to him like that's great live order go out beyond that um where do you see computing going now that you've declared pc is dead um uh just speculate a little bit yeah something that that i've been thinking a lot about actually is just this idea of being a modern participant in society and let's say an advanced society, how much compute, how much memory, how much storage do you need to be allocated to you as an individual to be able to operate in civilization?

28:37And you can look at it right now and say, like, you know, we're here, the four of us are all here in America, we're, you know, operating professional jobs here. And maybe we can survive a year ago, or a few years ago, we could survive with, let's say, like, a few 10s of gigabytes of memory, maybe a few terabytes of storage allocated to us, even for including what's amortized in the cloud for the cloud services that we run and some amount of compute. And year over year over year, that number is going to grow very, very quickly. The amount of just pure silicon that needs to be activated against each of us as individuals for us to be participants in society is just going through the roof.

29:18And so for us, as we think about what does that mean, if we're going to give you the power to own your data, own your compute, own your AI personally, what do we need to be able to build for you in a world where to be a modern participant in society, you need a terabyte of memory and multiple terabytes of storage and a very large amount of compute. Is it even plausible for us to build that box that you could sit on your desk that serves that level of compute to you. Where does it end? Where does it end? Maybe we all just get uploaded. I was going to say, when do we... I'm volunteering now. How does it not scale infinitely?

30:00That's right. Yeah, I think maybe that's the end point. We all get uploaded and we'll somehow in framework, we'll upload with you.

30:07Nirav Patel:If you look at these laptop pros, you've done some really, as always, aesthetic things, the uh the keyboards people are loving the multi-color keyboards the the bezel you really uh it's more than it's more than a geek uh laptop it's a real statement it's really it's really quite beautiful it is i wanted to order the x uh ultra x9 it seems like you sold those out right away that sold out in like an hour on the day of announce isn't that telling that the most expensive skew was the one that went away right away. Yeah. Yeah. People want power. Did you guys expect that? We did. So we actually shared during the announcement that we just had a pretty limited quantity of those.

30:49I think Intel shared some things around like yield challenges for the very, very highest end. It's 5.1 gigahertz. Yeah. It's really pushing limits. But yeah, people obviously, they want the power. So we sold out of those the four. Are you going to get more? anything else we're talking to intel we're trying to we're definitely trying to get it we know that there's hunger for hunger for this um so we're gonna get as many as we can good and you can buy

31:13Nirav Patel:ram it's expensive but but they have managed to find a way and it's not as expensive as it could be let's put it that way yeah yeah same thing with the ssds uh good all right well i'm in the batch 11 so that's pretty good do you ever think of doing i don't know a phone we get the ask a lot Whenever we do an ask for products and ask for what product should framework build, there's usually three or four that rise to the top. Phone is one. Printer is usually up there on the list. And then we get stuff like... A small projector that projects in your hand so that you could exist beyond phone. That idea has clearly gone so well.

31:55And then stuff like toaster ends up being high on the list of people. Like blenders.

32:00Nirav Patel:Yeah, better toaster. gosh yes yes a phone would be a phone that is the one thing where you really are in the walled garden there are things like the fair phone but it's very difficult yeah fair phones done excellent work competing against the giants in the space for us actually we look at the phone and see that again it's yeah it's like it's a hard product to build it's a hard product to get right and it's again not maybe not as much a hardware challenge a technical hardware challenge it's a there's a big software challenge there. And then there's a very big go to market challenge to compete in the smartphone space.

32:34All right. I think I'm going to get four USB-C ports. No more of these USB-A ports. I don't need HDMI. Wow. All in. I'm going all in, man. What's the plan for framework? Are you guys eventually looking, I know you're venture backed, are you eventually looking for an exit or do you want to stick it solo for the time being? The key for us, for me personally, is that I don't look at this as a financial outcome that we're chasing. It's really the mission outcome of going category by category and fixing as many categories as we can as quickly as we can and building install base and building market share and building ecosystems in each of these categories.

33:16And then the financial outcome is whatever it is. And obviously, if we're succeeding at this, we're building a self-sustaining company. We can fund our own scaling into these categories. We can basically dictate the financial outcome that we want. I think, I mean, I've been through an acquisition. I will say it's not, you have to let go of that. You stayed at that for a few years. You didn't run out the door. Through an acquisition gives you illustrative knowledge about all that it entails. Yes, I'm certainly not. Benefits and negatives. Yeah, and I've not been shaping framework to be an acquisition target.

33:53Framework is a very spiky target. Not deliberately, but it is a spiky target. But yeah, very much we're not looking for an acquisition.

34:00Nirav Patel:Well, bless you, Nirav. You've done an amazing job, an amazing product. And you're fighting the good fight. and kind of uphill, frankly, against a lot of industry trends that are not as good for consumers. So we are very grateful. And I know Corey Doctorow would say the same thing. Well done. Bravo. Thank you. Nirav Patel, founder, CEO of Framework at Frame.work. Don't go order the pro yet because I haven't pressed the button. So hold on till I get mine and then you can go order yours because I think you're going to want one. Nirav, thank you so much for your time. We appreciate it. Great to meet you.

34:37Nirav Patel:Oh, that was a thrill and a half. Next week, I'll be in Hawaii. And I don't know. I think we're going to put the Chris Stokowalker interview in there. And I'm sorry you guys missed that because Chris was great. I'm sorry I missed that too. Yeah, it was really fun. It ended up being wise that I called it because I was planning on taking that interview while on the Metro North back from Connecticut. And the train was packed. Oh, good. So it would have been an awkward record for all of us. Can't you all be quiet? I'm on a podcast. As long as you're not in the quiet car. I mean, they don't. On the Metro North, there are no quiet cars.

35:14Man, on the Shinkansen. There's only chaos. In the Japanese bullet train, there is a quiet car. You damn well better be quiet. Yeah, I'd like to see that. She's trying to do the podcast on the quiet car, getting shushed by everybody. That'd be fun to watch. I keep going to different quiet cars.

35:32Nirav Patel:and then we also have troy hunt uh which is very exciting of uh have i been pwned coming up my best looking australian guy since duranoke and then uh we got some good people uh on the uh slated on the future so um some really good guests coming up on intelligent machines let me do the commercial and we will get to the ai news in just a bit our show today brought to you buy scribe you know every time we onboard somebody new we would re-explain the same tools from scratch because there was no documentation right most teams struggle to scale because the actual workflows aren't captured anywhere and manual documentation is too inefficient to keep up the result is fragmented processes, inconsistent execution, and no reliable way to improve how work gets done over time.

36:33Nirav Patel:And that's what today's sponsor Scribe does best. Scribe is a workflow AI platform that captures any workflow in real time and turns it into documentation automatically. No manual writing, no manual screenshots, no starting from scratch every time somebody new joins the team it's trusted by 80 000 plus enterprises including nearly half of the fortune 500 you just turn on the extension and do the process as you normally would scribe builds the guide as you go capturing every click every step every screenshot automatically scribe automatically redact sensitive information names account numbers emails from every screenshot and as an admin you You can enforce this across your entire team and feel confident that nothing slips through the cracks.

37:23Nirav Patel:Anyone following this process can launch real-time on-screen guidance that walks them exactly where to click step-by-step inside the actual tool. It's there the moment the Scribe is created and ensures everyone does critical processes correctly. Scribe will suggest improvements to the underlying workflow too, not just documented. it. Once you can see how a process is actually being done, you can identify where steps are redundant, where people are getting stuck, and where things could be automated or simplified. It's not just capturing how work gets done. It's helping you do it better. To book a personalized enterprise demo, visit scribe.how slash machines.

38:02That's S-E-R-I-B-E dot H-O-W slash machines. My thanks, Scrap, so much for supporting Intelligent Machines. This isn't your order, but we got all the earnings palooza.

38:18Nirav Patel:Yes, and I didn't have those before we... So Google, Meta, and Microsoft all announced their earnings tomorrow. It's Apple. It's kind of crazy. I put them in the rundown under my... Well, give us a summary. Did you feed it to Claude or anything? I should have. Jeff is our Claude today. Jeff is today playing the role of Claude. Alphabet tops Q1 estimates. Yeah, so generally a lot of worry. Well, Google went down and then up. Amazon went down and then up. Microsoft went down and then down. Yeah, and Meta went down and down. Nervousness about AI spending, about cap everywhere, but the results are amazing for both Google and Amazon.

39:03Yeah, they're doing very well. But as you know, Amazon's committed capital expenditures of about$180 billion next year, which is one reason the stock market's going, what? I didn't see this in your… Amazon's also doing that. Alphabet's also committing$180 billion. Wow. Wow. And that's probably all data centers, right? I mean, I didn't see this in your order, Leo, but I saw it in Jeff's, which it's a big week, I feel like, for financial news about these companies, especially given the Wall Street Journal report that my former colleague Berber Jin broke. And we're trying to get Berber on to talk about that, yeah.

39:51OpenAI's pressure that it's facing in the lead up to the idea. Why did they, since they're a private company, they didn't need to let that be out? They didn't let that out. It was leaked. It really was leaked. Okay. So the Times also has a story. Andrew Sorkin and company also has the story. So OpenAI missed its internal targets. It wanted to hit a billion active users by the end of 2025. And they, as far as we know, still haven't done it. But remember the last time they announced it was 800 million. So they're close. But they've also, I mean, the highlights are that they missed monthly revenue targets.

40:28in earlier this year, repeatedly, the Wall Street Journal doesn't put a number on the size of them as publicly, but the framing is basically that growth is decelerating. It's not just falling short of stretch goals. They also noted that the board is now more closely examining the company's data center deals in recent months and has questioned Sam Altman's efforts to secure even more computing power despite the business slowdown. And as Drew Citron points out, that's bad news for Oracle. Yeah. Because you've got all these companies that are depending upon future revenue from OpenAI and these promises.

41:05It's bad news all around. OpenAI, as you know, is getting ready for an IPO. And this is exactly the kind of news that they don't want. But better now than later, I guess. I mean, that's why I wonder whether it was really a leak. I don't think so.

41:19Nirav Patel:I think, actually, they'd be much better after the IPO than before, right? Okay, putting on my reporter hat. So, we get a story like this. Obviously, all of the sourcing in this anonymous, it's all multiple people familiar with the issue. That means someone, and this is all about specifically disputes between Sarah Friars and other executives and Sam Altman's team. And then kind of going back and forth on specifically, there's this paragraph, the spending scrutiny is constraining Altman's once boundless ambitions ahead of an initial public offering that could take place by the end of the year.

42:01Fryer and other executives are seeking to control costs and instill more discipline in the business at times, putting them at odds with their CEO. Putting my reporter hat on, you have to think, how did this story get here? Obviously, Berber is a phenomenal reporter. He's got great sourcing in these companies. But the only sort of people who know all of this information are going to be high, high level executives that are in some way intimately involved with this or briefed on it to be able to have multiple people familiar. it's probably one of the two camps leaking it again you have to think why would somebody give this to the wall street leak this information to the wall street journal what would be the purpose for them my guess is that it's people in friars camp who are trying to actually get these constraints to go through or feeling opposition um and want the backing of pressure Or maybe it's people in the Altman camp who want to force Friar's camp and Friar to have to make a joint statement saying, oh, no, no, we're not doing that at all.

43:03Which is, of course, what they said in this piece to kind of dispel these rumors that they put out a joint statement on Monday.

43:09Nirav Patel:We are totally aligned on buying as much computers we can and working hard on it together every day. Any suggestion we're divided is, quote, ridiculous. And it's like, well, clearly there's some division because one of your two sides is leaking stuff to the journal to try and turn the heat up on the other side. And so I think this just speaks to a broader disagreement happening between these two camps and the intense pressures of trying to launch an IPO for a company with costs this astronomically high. But is there also, I'm trying to see whether there's kind of a sandbagging strategy too. that months before your IPO, you want to reduce the expectations.

43:52So that's why the IPO comes around, you raise them again.

43:55Nirav Patel:Yeah, yeah, yeah. I think there's also, I think we should also point out, there is a target on open AI's back from a lot of people, including Elon Musk, who is right now on the stand testifying. Testifying there and in Twitter. Yeah, yeah. The judge said, would you knock it off on the social postings? Both of you. He didn't say it quite like that. But he basically said that. Anthropic is also going to do an IPO. And Anthropic and OpenAI are in the bitterest competition since Coke versus Pepsi. I mean, these guys are really at each other's throats. And I think that both Sam Altman and Dario Amode of Anthropic think it's a zero-sum game.

44:38Nirav Patel:One winner and one loser. But a lot of people would suffer consequences of OpenAI collapse. that would be the trigger that would you know pop the bubble the if there is an ai bubble that that would be the trigger

45:02this is the clip that accidentally showed up for a brief second on last week's episode everybody's I was wondering, what's going on? And it was just a second. Just a second of a little bubble. We love, I'm happy that I finally get one. It's so great, Anthony. But I think, yeah, to your point, Leo, I think that there are reasons why people on both sides of this debate, even within OpenAI, would feel motivated to try and leak this for those exact reasons. Like if you're on the Altman camp, you're like, well, we have to get as much compute as possible. We have to build and scale as quickly as possible or else the bubble might pop and this might all be for naught.

45:49And you could see from the Friar camp that they'd be like, well, we actually have to rein in our spending and make sure that we are able to meet some reasonable expectations before we hit the IPO or else the bubble could pop. And meanwhile, Anthropic is looking at a higher private market valuation right now than OpenAI. Yeah, Anthropic is very popular right now. But I should point out, much like Coke and Pepsi, there isn't all that much to distinguish the two models. And it could swing in either direction. Your partner is going to be upset with you tonight if you're saying that. Shh, I didn't say anything, Claude.

46:28I'm not talking to you.

46:29Nirav Patel:uh in fact this week uh open ai released spud their 5.5 chat gpt model they also released new image models people have been very impressed with but they do this in response you know to 4.7 which came out the week before uh let's not forget the chinese companies like deep seek just released version 4 we'll talk about that a little later um it is very competitive out there and i don't know if i don't know if the winner is clear at this point um and so there in other words there's a lot of chum in the water and i think the closer and closer the two like the more interchangeable the two products are the more that these business fundamentals ultimately end up being incredibly meaningful for either company's ability to stay alive and win this race because ultimately both Both of them are hurtling, have to hurdle towards an IPO, given the amount they've raised.

47:27They can't continue, you know, to be private. And it's like, how is the market going to react if you, whenever they have to file an S1 and it comes out that they're burning God knows how much every single month?

47:43Nirav Patel:Well, but again, look at the quarterly results from Google, Amazon, and Microsoft. All three are burning lots of cash building data centers. They're also in a head-to-head battle between Azure, AWS, and Google Cloud. Literally. I mean, they're fighting head-to-head. And I think the market generally approves. Yeah, and their results are – and for Amazon and Apple – I mean, Alphabet, their results are amazing. Flu past the expectations. Flu past. So even though the stock drops, you know, this always happens. The rule of the market is buy on the rumor, sell on the news. And this is the news, right?

48:27Is the results.

48:28Nirav Patel:I think the market's pretty tolerant of these big CapEx expenditures because there's generally a feeling rightly or wrongly that we've got to build all this compute capacity, that this is the way forward with AI. Well, that's the key difference with these is that all of the companies you just mentioned And also have large, robust, mature, profitable businesses undergirding them. And OpenAI and Anthropic are new upstarts that I'm guessing all signs point to are losing a lot of money every month, every quarter, every year. And probably will continue to lose a lot, a lot of money for a long time until the goal is that they'll make like an Amazon or Uber-esque pivot eventually when they reach the sort of scale.

49:16and then it'll be, you know, nothing but net, so to speak.

49:20Nirav Patel:The other data point that's confusing to me, maybe you guys can think of a rationale, is that Microsoft kind of freed OpenAI from the chains. Remember, Microsoft was the initial big, this is after Elon, Microsoft gave OpenAI$10 billion in Azure credits. They're running on the Azure cloud and really made a big commitment and they're reselling ChatGPT as co-pilot and so forth. Microsoft and OpenAI have agreed just the other day to drop the software giant's exclusive right to sell OpenAI models. So immediately, OpenAI turned to Amazon and said, and you get ChatGPT, and it's now on AWS, which a lot of people who use AWS are very, very happy about.

50:08Nirav Patel:Microsoft no longer has to pay a rev share on OpenAI products. It resells on its cloud what about the agi clause and that's that was one of the most interesting things i brought up with paul therott there was this clause which we always all of us thought was bizarre that should open ai reach agi aren't uh super intelligence you know um the singularity uh we don't there's no even good definition for agi should it reach whatever that is agi and they had third parties that would validate that, that the agreement would dissolve, right? That the open AI would get to go on its merry way. That's gone.

50:51And some people said, well, that's what's keeping open AI from announcing AGI. I don't know if that's the case. Announce. Announce. It's meaning. The whole thing is meaningless. We never understood why the clause was in there in the first place. It was actually added later to the agreement. Anyway, I feel like we don't really know what's going on at all. Yeah, Amazon got freed from some shy. I think Microsoft, in my view, Microsoft got the better end of that deal. They were, they were, they were desperate at the beginning to have some AI, they went to open AI, it seemed clever at the time, they got kind of shackled with it.

51:31And now they're free to do what they want to do and open eyes for you to do what it wants to do. It makes more sense. Yeah, it's yeah. Okay. I mean, I don't understand any of it. To be honest with you, it just feels like people are running around like hamsters in a hamster wheel as fast as they can. Paris, I'm curious as to your guys's thoughts, not to drill slightly, but I'm curious as your guys's thoughts on this thing that I've seen running around the forums lately, which is, I'll post the link in the Discord right now. GitHub Copilot this week announced an increase in the model multipliers for its annual kind of Copilot Pro and Copilot Pro Plus subscribers.

52:18And a lot of average users seem to be pointing to this as a possible like smoking gun about how much actually these models and the tokens are costing companies. For instance, they changed the multiplier for usage for Claude Opus 4.7 from a modifier of 7.5 to 27, which is completely out of whack with every other model. Basically, the Claude credits or the multipliers have increased exponentially and all the other ones have, I mean, they've increased like six times or something like that, but nothing as much as a jump from like seven to 27. Does this make sense to you in some way? It doesn't to me, or is this, as the commenters seem to be crying, a sign that all this AI is way more expensive than we actually think it is?

53:16Nirav Patel:Yeah, it might be making the pricing more accurate. it uh i think it's also the case these companies uh need money badly i think the biggest they're really the biggest constraint right now is just pure compute constraint those those data centers aren't getting built uh as fast they keep announcing new deals but they're not getting built that fast we're running out of hardware um chips uh do you think there's any hoarding hoarding of this stuff oh yeah for sure that's on you have it you're hoarding it yeah so there's there's a lot that's unused so we don't really know what the what the necessary demand is and there's a debate about whether super scaling is the path to the next universe well that debate is a side debate that is not widely accepted by the current frontier model companies right the current main ones but but there's so that's a debate that you this is that old you know yan lecun feifei lee debate but uh but that is not what the uh what the um leading edge companies are well even demos changes too it's yeah demis says it too of deep mind but uh i think generally the the big players generally you're all saying right now more compute the bitter lesson is learned uh the llms are doing the job and we just need to throw more power at them more power Or whether they're right or wrong, we won't know.

54:40Nirav Patel:We can't know. And, you know, I mean, there's no, let me put it this way. There isn't a lot of evidence they're wrong. There's speculation. Uncle Leo, since you were old enough to remember this, the overbuild of fiber in the country. What was that sequence like where there was euphoria about building, there was a huge need to invest in it, then there was a crash, but then we ended up using it. Is there a similar path that could be here? I'm assuming we're overbuilding now. that's that's what uh jeff bezos was saying is that you can't compare an infrastructure bubble to a financial bubble and that this like the fiber overbuilder the railroad overbuild is a infrastructure um bubble right and so you get at the end of the day when the bubble bursts you get the infrastructure no matter what and that's what happened with fiber that's what happened with railroads and it presumably is what's going to happen with data centers maybe i think that there isn't a lot of historical precedent for what ai is going to do to the economy i think part of this is really a lot of this scrambling is that no one knows what the impact of ai is going to be on the economy and on jobs and business and software it's and you see all these theories it's going to kill open source it's going to be the best thing that ever happened open source it's going to be a security nightmare it's going to solve all security problems no one knows and there's just this great uncertainty.

56:16And it's being used as an excuse to do layoffs that probably aren't related to AI. There's AI washing going on. Although I think you probably can safely say when companies like Meta and Snap lay off thousands of employees, their plan is to make up for some of that with AI. And I think that's for sure true. Or they're abandoning things that they don't think are as potentially profitable and they want to put whatever resources they have into AI. I was going to say, I mean, the big tech companies, especially since the pandemic, but increasingly, I feel like over the last eight years, have made it a kind of a regular practice to have annual layoffs of some sort.

56:58It's just part of their business plan. And they're still up for where they were. Right. We overhired, and now they're right-sizing, if you don't mind me. So the point is, we don't have data yet, to your point. We don't have data yet of anything.

57:13Nirav Patel:We don't know what's going to work. we don't know what's going to happen there's a lot of uncertainty we don't know if we're going to get oil from the straight of hormones there's a lot of uncertainty in the world right now and i think that a lot of this is just thrashing as companies and investors and venture capitalists and stock market investors are trying to figure out what's going to happen i don't think anybody knows it's just i think to a point that you often make on the show leo which is kind of comparing it to the internet i think that some part of people's what you might describe as an overreaction to ai and obsession with kind of pontificating about the potential impacts is that a lot of people if you're thinking about a lot of industries didn't accurately predict or prepare for what the advent of the internet would do to their businesses for instance frankly media companies should have been thinking about this and how they didn't want to craigslist may have they i don't know you were in the front lines jeff they may have known they just didn't want to they well it was a mix i think some some news didn't know but they didn't they didn't know what to do they didn't know what to do and they didn't spend nearly enough time thinking about what they should do and thinking of ways to prepare for it and so i think what we're seeing now is an over correction to that right uh this is one of many you know people were thinking it would be like crypto or fds or something like that before but that was obviously smaller this is such a potentially transformative technology that everyone is freaking out and devoting so much energy to trying to suss out the precise impact this is going to have on the world and the ramifications of that because they're afraid if it ends up like the internet it's going to be a disaster for them well and in a non-financial vein of the same kind people like us we thought the internet was going to be great we didn't see all the harms and the hazards that the internet was going to present so everybody's thinking geez we we missed the boat on that financially and and and societally maybe we should plan but the the big difference here is this is happening at least 10 times faster than the internet happened this is the internet didn't you know the what was the a when was the aol summer jeff 95 96 uh that was one big transition where a lot of non-techie people suddenly got on the internet.

59:32Nirav Patel:It still took another 10 years before the web really kind of took off. Blogging didn't take off till the early 21st century. I mean, I would say it was at least a decade of development before the internet got to where we are today, maybe 20 years. That's going to be compressed to a year or two in AI. I mean, this is happening really fast. So if you weren't prepared for the internet, you are not going to prepared for ai i agree with you the exact your premise exactly that people are saying well we got to prepare but it's moving so fast you're going like what what what what what do i what do i do what do i do well here's the other piece of it too so sam altman put out his principles another one of his another one essays yeah another one man that guy's and i wrote i wrote a brief piece just trying to remind that in the end the technologists don't have the technology and he thinks he can control everything including government policy about this and he can't.

1:00:30And it's going to be out of his hands. And government's always behind. And I just listened to a podcast with the authors of a book called Muskism. That's a good name. It really is. And it's really interesting. The premise there is that Musk's view is that he combines business and government. Musk depends upon government. He does. Yes, right. For rockets, for satellite transmission, Doge, all of it fits together.

1:00:56Nirav Patel:funding really came from us, from taxpayers. And there's a belief there of control. But these guys, and part of the reason that people hate AI is because of the AI boys. Well, people also hate AI because they're scared. Yeah, but there was another survey, a bunch of surveys that came out this week. I put it at the rundown somewhere, where if you look at the fear of AI in the US, or the confidence in AI the U.S. is in the 30 % range. In China, it's an 85 % range. In Germany, it's in the 45 % range. They're scared of everything. And so the narratives here of the extremes, of the extreme optimism and extreme pessimism and doomsterism have taken over.

1:01:45And there isn't, to your earlier point about not having the data, we also don't have the mindset to look at this

1:01:51Nirav Patel:insanely well i can tell you why we're more scared here in the u.s than we are in china they have a safety net we have no societal safety net true that all of those people can be out of work they they're going to be on the street and one could argue this is a time for control i'm not suggesting this here but if you you know a controlled economy um ai will will grow there faster than here it'll get all the uh as jensen wong said they'll get all the power they want they got all the resources they want and i guess they're gonna be my point is we don't know what's gonna happen and any prediction or attempt to prepare for it i know i should get amy webb on because that's her job is preparing companies and governments for the future i think any attempt to prepare for it is is futile because we don't it's it's this is chaos but aren't you singing a slightly new tune leo in the sense that you've scolded both of us saying, this thing is the arrival.

1:02:49It's everything. And unless you, right. I didn't say it's going to be a wonderful world as a result. I don't know. I think part of what Jeff is getting at is there have been discussions previously where you've kind of pooh-poohed the idea of hand, like wringing your hands over what the impact of AI is going to be, that it's already here and you don't need to be posturing about it. But I do think we all are kind of in agreement that people are worried because it is here and happening and they're trying to do it better.

1:03:21Nirav Patel:That's not inconsistent with what I've just said. Because the reason I say there's no problem with hand-wringing is we don't know what's going to happen. So you can, it's a philosophical exercise to kind of plan for the future. But good luck. All we know is that the future is going to happen. And it's going to be, I think, extremely disruptive, chaotic. it might be good it might not be um i uh you know i mean uh i think it's very you know so some of what i it drives my point of view is i'm just very interested in it yeah and i'm not trying to i'm not trying to um come up with a uh con a consequence of what happens because i don't i don't know it's gonna be crazy man so line 144 is the boy no we gotta take a break hold on okay uh we will have more in just a bit less philosophic more news coming up on intelligent machines brought to you today by out systems the number one ai development platform so you know you want to incorporate a ai you want to you want to bridge that enterprise gap to your agentic future but where do you begin how do you do it and how do you do it safely reliably consistently out systems helps businesses bridge that gap to their future where the constraints of the past give way to unlimited capacity and scale out systems enables businesses to build ai agents that can actually do work it's possible things like take actions make decisions and integrate with data rather than just, you know, answer questions.

1:05:04We're not talking chatbots here, folks. OutSystems provides the only AI development platform that is unified, agile, and enterprise proven. It's unified because you build, run, and govern apps and agents on that single OutSystems platform, right? It's agile because, well, with the power of AI, you can innovate without compromising, thanks to OutSystems, quality or control and it is enterprise proven out systems is trusted by enterprises for mission critical ai applications and durable innovation just look at the website they have hundreds of case studies showing how well it works out systems is the secret weapon behind the world's most successful companies and not just for small apps these are also for those massive complex systems that run banks insurance companies and government services out systems even helps companies with aging it environments bridge the gap to the ai future without a rip and replace nightmare out systems provides the safest and fastest way for an enterprise to go from yikes we need an ai strategy to hey we have a functioning ai application stop wondering how ai will change your business start building the agents that will lead it.

1:06:20Visit outsystems.com slash twit to see how the world's most innovative enterprises use outsystems to build, deploy, and manage AI apps and agents quickly and cost-effectively without compromising reliability and security. That's O-U-T-S-Y-S-T-E-M-S dot com slash T-W-I-T to book a demo. Thank you, Outsystems, for supporting intelligent machines. We appreciate it. and you support us, by the way, dear listener, by going to that site with that address so they know you saw it here, howdsystems.com slash twit. So, all right, money going back and forth, people throwing shade at one another. The Pentagon's using Anthropic.

1:07:06Google says we've done a deal with the Pentagon. They've signed a classified AI deal for quote any lawful government purpose google employees immediately petition sunday i think they were even before that that's petition to say no yep uh hundreds of google employees signed a letter against the idea after reports that google's in talk with the pentagon so yes that was before the talks resolved themselves into a deal i guess they didn't listen to the engineers we'll see what happens on that um there's i know there's a lot of pro back and forth i swear to god i haven't heard the testimony from today but elon musk is on the stand the verge said he appeared yesterday more petty than prepared he was unfocused and uncharming elon elon well elizabeth lopato wrote this story said you know in the in the uh in the liability case during his defamation suit he turned on the charm she said i've this is not the first time i've seen musk in court and the jury responded by finding him not guilty today lopato writes he looked adrift and unprepared the only times he showed real animation when it was when he was bragging about how much he'd done for open ai of course the suit is elon said i poured all this money open ai and then all of a sudden they go for profit and leave me out and i want my hundreds of billions of dollars pieces of the of the of the case got thrown out pieces are left what's at stake now do you know where the standing is the piece that got thrown out was actually at the behest of musk's own team who said uh we're not going to go after the fraud part of this they they were going after OpenAI for defrauding Elon.

1:08:56They said, we're not going to go after that. But they're still going after him saying, you can't do this non-fake, non-saying, first of all, it's fake, this non-profit move. And you can't separate the stuff that I invested in and my supposed revenue from that by saying, well, you invested in the non-profit part. We're just going to take those profits and move them over here, and you don't get any Elon. That's really essentially what it comes down to. During the discussions of how best to get open AI, the vast amounts of funding it would need for compute, there was a discussion of a for-profit arm with Musk.

1:09:40But that's not why Musk got into it. He says, I got into it because I had a conversation with Larry Page. It scared the pants off of me. because Larry Page says AI is the next species. And I'm specious if I want to defend humans against AI. And Elon said, my God, when I heard that, I said, we've got to make sure Google doesn't get AGI before anybody else. But is the argument in the end, you simply cannot convert to a for profit? Yeah, or if you do, I want 55 % of the profit, which is like a lot hundreds of billions of dollars anyway he's testifying he's still testifying presumably you're muted uh paris presumably the cross-examination is today oh yep oh you can hear me there we are now yeah i was saying he he's also asking for sam altman to be removed from the board oh yes and everything you know this is this is a fight to the death what's so what's the open ai's argument going to be the thing wouldn't have survived as a not-for-profit.

1:10:46It's all about scale. To be scaled, we had to be profitable. From OpenAI's attorney says, we're here because Mr. Musk didn't get his way at OpenAI. That's what happened. He quit saying they would fail for sure, but my clients had the nerve to go on and succeed without him. He basically argues that Musk used funding promises to bully early OpenAI, trying to like merge open AI into Tesla, demanded 50 % of ownership and pulled 5 million quarterly donations mid-nocation. Casey Newton told NPR, this is a clash of two enormous personalities in Elon Musk and Sam Altman. Sam Altman will take the stand later in this trial.

1:11:26And I think what's at stake is potentially the future of open AI and the future development of all AI. That might be a little bit hyperbolic well i mean i think that part of what he's getting at is that the one of the remedy figures is like 134 billion dollars kind of what they're looking for and that's the same order of magnitude as open ai's like near-term compute exposure so i guess if musk was to win meaningfully it would be yeah like fatal to their roadmap well clearly musk wants to do that because remember after all this musk started xai and went right into a for-profit competitor that he's pouring billions of dollars into and so this is there any business to xai is there any business to any of this i know you're surprised to hear me say that jeff i can see the stunned look on aren't i a subtle actor what but i mean we don't know right there's no nobody's making money at it yet i mean i think there will be but i think you're right it might be like fiber in the railroads it might be that you know the railroads all the companies that built the transcontinental railroad went went bankrupt but we had the we had the tracks we had the trains i think the difference is that the tracks don't go tracks don't get uh old you know you can still use those tracks right yeah that's a good point i don't know what would happen well you'd do nobody wants at some point nobody wants chat gpt40 anymore i went to the museum of industrial history in bethlehem pa last weekend because they have a linotype there so you know that oh i'm adding that to my museum list that sounds so funny they have huge pumps they have a little tiny little tiny uh uh diesel engine that used to drive things around the steel mills.

1:13:19You can ride that for about 20 feet. You can go on the catwalk and see the old huge blast furnaces. How long of a drive was it? From you, it's probably an hour and a half. Ooh. I'm going to be doing that. When Leo comes, we could see the Amazon warehouse and go to Bethlehem. Leo, visit us. You could eat a sandwich made by your son. all i was gonna do is give a sandwich and we'll get take out and then we'll go to the car uh according to alex kantrowitz also talking to npr elon's goal in this is to just shut open ai down just because the money he gets would go he says to a charity not to him he doesn't he doesn't want he doesn't want money he wants to put them out of business which is the charity that's in his compound that houses all of his children?

1:14:14Probably, yeah. The charity that teaches his children how to read. I'm sure there's a legitimate charity. He doesn't need more money. I'm just trying to find out if anybody has reported on his testimony today, and I don't see anything. So if somebody finds a link, let me know, because he was, I mean, I'm sure court is recessed. It's in Oakland, so it's California time is 3.30. OpenAI may be doing a smartphone. I wonder if this is the Johnny Ive project they gave him 3.2. Well, otherwise, what's he doing there? Right. This is from Ming-Chi Kuo, who is normally a supply chain analyst who reports on Apple, but he does have good sources in China in the supply chain.

1:14:57He says OpenAI is going to make a deal with MediaTek and Qualcomm for a processor. MediaTek makes a very good Chromebook processor, by the way, despite its kind of down market name. They're actually pretty good. Qualcomm, of course, makes the Snapdragon. And the builder, the manufacturer would be Luxshare, which is a design contractor. They would be the system co-design and manufacturing partner. Don't worry about saving your pennies yet. Mass production doesn't start until 2028. But according to Ming-Chi Kuo, and again, it's a rumor, but he has very good sources in the supply chain. The idea of this phone is it will not have apps.

1:15:35It will have AI. and it will be an agentic phone that you will tell it what you want and it will do what you want exactly to nero's point earlier in the show yeah that you're not going to own anything you'll read it own your apps you don't even own great because that's worked out really well in terms of video games and media and everybody loves when they don't own anything and it can just be taken from you at any given point or you can be suddenly charged a subscription to have things like windshield wipers in your car. People have been clamoring to have a system like that, but for everything.

1:16:15What's the number one music source in the world right now? It's either Spotify or iTunes. In both cases, you're not buying the music. You're renting. Yeah, but you're trapped. People don't want that, but that's what we ended up with.

1:16:28Nirav Patel:Do they not want it? They can still buy CDs. They can still buy digital copies of music. So you think it's good that... No, I don't think it's good. I'm just saying that our set is aware of this. But I don't think your mom and dad care.

1:16:47I mean, I think my mom and dad care a lot whenever they're like, I could have just, they used to take me to the store to buy Windows. Used to. But now they, and you pay one thing and you'd get Microsoft Office and now you got to pay every month.

1:17:01Nirav Patel:Used to. But, well, I agree with Microsoft Office. Nobody, I mean. but it's the microsoft office occasion of the universe no i i get it i agree with you 100 but i'm just wondering if the general public i mean they seem pretty happy with spotify i mean i think i see this as a common complaint like in various social media from normies no but i see it on like reddit and twitter i'm sure if i was on it's on threads it would be there too but people the average person doesn't like that they have to pay a subscription for everything they realize that that's totally true yeah that's true yeah but yeah but they live with it i mean honestly if you looked at the box price of microsoft office of six or seven hundred bucks and then i said to you or you get it for eight bucks a month with automatic updates it looks like it's a better deal until you calculate how many months you're going to use it for well if it's been a hundred bucks a year but it would cost you 600 bucks to buy it outright in six years are you going to need to buy another version?

1:18:01Nirav Patel:Probably. The other dynamic here is that there is a finite number of subscription dollars that people are willing to spend. That's an interesting point. Yes. Yeah. People are, there is subscription. And so, you know, screw it. I don't want my newspaper anymore. There was just a survey out about people who live in news deserts and they don't have newspapers in their county and they don't notice. Right. And they don't want to pay for anybody and they find the news the way they want to find it. Paris is still buying DVDs, but Paris, you're in a very small minority almost everybody streams the movie i'm not claiming that the people yearn to buy dvds i am saying that the people enjoy enjoy not paying money for anything the people get mad whenever they realize they're paying for seven different streaming services to the people are mad that we're basically reinventing the cable bundle but for every industry yes and there's a certain point where as more and more parts of an industry become monthly subscriptions that like it reaches a saturation point the average person can't pay hundreds of dollars a month for like everything you know anthony also makes a good point piracy is probably also going to be on the upturn you know because of all this is it though it feels like piracy went away when you could stream stuff i mean anecdotally i've started to see a lot more people pirate stuff or mention pirating online okay i feel like as soon as it was easy just to pay you know whatever it was seven bucks a month for uh three billion records people just stopped pirating music because you only need to do one but now you have to do like 20 of those 20 what subscriptions but not for music for music there's one all the stuff you know yeah but the rest now they're an extra they'll pay their spotify but they'll pay i got netflix that's all i need i'm happy some of my friends one of my friends my friend who took me to the basketball game the other week tried to explain to me the amount of subscription services you have to subscribe to to watch most of the knicks games that happened okay you want to talk about that it's like 10 i agree people don't love that yeah yeah in the philippines i paid 100 for the entire nba season and I got every game.

1:20:16Jeez, that's a lot. $100 for the whole season. $100 for the whole season is reasonable. YouTube TV is$80 a month. And that's not even a guarantee of every game. Right. Well, that's what the market will bear

1:20:29Nirav Patel:probably in the Philippines, right? Yeah. Let's take a break. Intelligent Machines on the air coming up. Let's talk about... Oh, I don't know. He's a man. the wheel folks the man behind alpha go thinks ai has taken the wrong path this is the deep mind discussion that you were talking about also a llm from 1930 um the pirate piracy is back is piracy back wow okay so this is from briggs this is a youtube channel that says piracy is back so it must be true it's true uh paris martineau you're from consumer reports at some point we're gonna talk about can you talk about anything you're working on anything at all it's coming coming soon and exciting at some point we got to talk about my problems with claude this week because oh yes first time ever i've been real yeah i've been really experiencing the crunch of uh a new model upgrade like i want to make fun of the comment with claude i think that's i do stuff with claude every week yeah i want to hear what that is meanwhile i used claude versus google and claude beat google and what i was i'm finally ready to do stuff with my book using all these things oh cool cool cool that's jeff jarvis author of the gutenberg parenthesis his new book hot type and now you're doing a whole series of books about ai intelligent from bloomsbury academic uh intelligence ai and humanity intelligence AI and humanity wow when's that when's when will that come out next year because it takes forever yeah you got to write them there's three people writing them right now we've talked to two of them yep nice yeah we have well let's get the third on okay by the way wasn't Ian great last week he was great if you didn't hear the interview with Ian Bogost Bogost uh that was he was just great and I still can't get his name right why I'm very good with names I don't know why I want to put the accent on the second syllable too.

1:22:38I agree. Yeah, I was gonna say but ghost but it's boggles. Yeah.

1:22:42Nirav Patel:Our show today brought to you by z scale of the world's largest cloud security platform. You know, we've talked about AI, the potential rewards of AI and business. Far too great to ignore right now. But it is a double edged sword. The risks are also far too great to ignore the loss of sensitive data and attacks against enterprise managed ai and of course generative ai also increases the opportunities for threat actors they're using ai just like you are helps them to create phishing emails that are indistinguishable from the real thing to write malicious code to automate data instruction there were 1.3 million instances of social security numbers leaked to ai applications and by the way that's not like a hacker doing that that's like your employees putting proprietary company company information into an ai bot to get information not thinking about the fact that that bot is now going to send that up to the uh home servers in china or wherever well not with zscaler zscaler is the most trusted ai security platform 40 of global 2000 companies use zscaler 500 billion transactions, half a trillion transactions are secured daily with more than 9.4 thousand global customers.

1:24:11Nirav Patel:Wow. Zscaler carries a net promoter score of more than 75. That's 150 % higher than the average SaaS. They're doing something right. Check out what Siva, the director of security and infrastructure at Zuora, says about using Zscaler to prevent AI attacks. With Zscaler being in line in a security protection strategy, it helps us monitor all the traffic. So even if a bad actor were to use AI, because we have a tight security framework around our endpoint, it helps us proactively prevent that activity from happening. AI is tremendous in terms of its opportunities, but it also brings in challenges.

1:24:49We're confident that Zscaler is going to help us ensure that we're not slowed down by security challenges, but continue to take advantage of all the advancements. Thank you, Siva. With Zscaler Zero Trust Plus AI, you can safely adopt generative AI and private AI. You can boost productivity across the business without the risk. Their Zero Trust Architecture Plus AI helps you reduce the risks of AI-related data loss and protects against AI attacks to guarantee greater productivity and compliance. Learn more at zscaler.com slash security. That's zscaler.com slash security. We thank them so much for supporting.

1:25:28Intelligent machines. Intelligent machines. So China is going to block Meta's acquisition of Manus. Manus, which is a agentic AI company, was started in China, but then they moved to Singapore, a neutral nation. it's actually called singapore washing and the chinese government would have would not hear of it

1:25:57so they china says the core dna of manis was developed domestically and we are going to block meta's proposed two billion dollar acquisition of manis which i'm disappointed because i was looking forward to using manis it's a very interesting it's kind of like a secure enterprise open claw what happens to manis now i wonder that's a really good question uh it's fast companies acting like well that's that like okay um you know meta's gonna kind of pull back and uh i don't know china's block on the meta manis deal would likely be viewed as a new flashpoint in the escalating competition between the u.s and china per ai dominance you think yep yep and there is a summit coming up next month in u.s china maybe that will be one of the topics i don't know there seems like there's more important things to talk about like tiktok like tiktok australia speaking of which has unveiled a two and a quarter percent levy a tax if you know meta google and tiktok's local revenues unless they decide they're going to pay news publishers is this a rupert murdoch deal oh yes absolutely yes pure pure rupe pure the news bargaining incentive because because the last deal didn't work because meta said okay we're not gonna find buy news and google said no we're not gonna do this anymore and so now whether or not you have news they want you to pay because they think you're evil and digital oh that's so sneaky so the original idea was oh you pay us for the news you use but now that you're not using news you just pay us period yeah i mean that's because we deserve it we deserve it it's ours oh wow that is evil and uh you saw this story obviously uh in uh wired magazine will knight writing the man behind alpha go thinks ai is taking the wrong path actually this is not demis hasibis who's also said the same thing it's david silver and he has a new uh ai company uh he he developed the the program that beat the best go players right the alpha go he has since founded a company called ineffable intelligence okay i did consider that as a name for this show but i rejected it as being horrible ineffable intelligence does that mean you can't f it you can't it's uneffable it's ineffable what is ineffable means what it means uh like you can't describe it it's like uh it's ineffable yeah it's out there in the ethos it's i always kind of think of it as a a cloud that if i try to grab it slips through my fingers too great or extreme to be expressed or described in words we can't describe it it's ineffable well that's kind of that's kind of where we are with definitions of agi yeah yeah i can't describe it it's too great in fact that's exactly what it's supposed to do is build more general forms of ai super intelligence now the i think the article is a little misguided because silver is not abandoning llms he's focusing on reinforcement learning which is a fine tuning that happens after the model is built and it was in fact the thing that last year, propelled DeepSeq to the headlines because they come up with a small and expensive model that was far better than anybody expected by using a technique that few knew about called reinforcement learning.

1:29:33So anyway, Silver says he thinks this approach, the approach of exploiting coding and research capabilities of large language models will fail. As amazing as LLMs are, says silver they learn from human intelligence rather than get ready for this building their own

1:29:54okay you know as as we've always said i think it's good to have all this more more more stuff exploration is good right doesn't i'm not sure i buy it but fine if he wants to put his energy into this that's great and uh yeah more more try more trying yeah i gotta try he has raised 1.1 billion dollars by the way for this effort raised or he's valued at he's valued at 5.1 billion jesus

1:30:25so uh the answer is it's ineffable it's ineffable that's what you are uh all right now we're going to do the good the bad and the ugly this is a new idea i have for uh for a spine down the middle of our show don't say spine to me it hurts ow how is your l5 it's getting better it's getting better l3 l3 there's so many l's i can't keep so many l's and we're not taking them this time you didn't take the l you took the tunnel and you see that i did take the l but i when you take the l literally the train you're always taking the l figuratively in some sense and that's what happened to me today but does he i thought l stood for elevated.

1:31:09It's a tunnel? Oh, no, it's a tunnel. It's just a letter. Not the L like in Chicago, the elevated railway. Oh. It actually goes under the river. That's long gone. You couldn't really have an elevated railway over the river, probably. Well, you do go over the bridge. Well, there are ones that go over the bridge. That bridge is a bad word. Oh, sorry. I'll show that, too. An amateur, this is the good. There's only two good stories. You'll be glad to know there's hundreds of bad ones. An amateur just solved a 60-year-old math problem not a mathematician he asked chat gpt and chat gpt proved a conjecture with a method

1:31:46Nirav Patel:no human had ever heard of this is a 23 year old named liam price he has no advanced math training what he does have right scientific american what he does have is a chat gpt pro subscription uh he solved an erdos problem uh the new solution which he got in response to a single prompt to chat gpt54 not even the latest model pro he posted on erdosproblems.com a website devoted to these problems just a week ago the problem it solves says i am solves uh has eluded some prominent minds bestowing it some esteem okay is that that feels wrong like an ai might have read feels like yeah yeah yeah bestowed upon it some esteem the problem itself has eluded some prominent minds bestowing it some esteem more importantly the ai seems to have used a totally new method for problems of this kind too soon to say with certainty he did submit it to uh terence Tao, who is kind of the guy, the arbiter, a mathematician at UCLA, a scorekeeper, if you will, for AI's push into math, what's beginning to emerge is that the problem was maybe easier than expected.

1:33:05And it was like some kind of mental block in the math community. We weren't going down the right path. Anyway, that's all I can say. It just means that nobody asked ChatGPT this question before. Right. Right. Interesting. as soon as they asked it i said well thank god you finally asked me the opportunities are out there folks you can 60 year old problems yep and then you like this one talkie a 13b 13 billion a parameter vintage language model from 1930 now they didn't make it in 1930 they trained it on uh books and texts from 1930 and earlier right jeff right right exactly so i asked it to get the perspective of the time, 1930, is the world at risk for a second great war?

1:33:54Remember, they just got out of World War I. Answer, no. Wars are becoming more and more unpopular, and the growing intelligence of nations renders them less likely to break out. Though the area of hostilities may be increased, there is a strong probability of their duration being diminished. Wow. So much for futurists. Wow. That's hysterical. You can download this, by the way. Well, yeah, you can also, if you go up to the chat, if you go back to the page, up the top, it says chat. It's not working for you? Yeah, no, it's not working for me. It's under high demand, but it's fun to play with.

1:34:25Nirav Patel:Connecting, connecting. It's on Hugging Face. It's also on GitHub. So I also asked it, I'm doing a lot of research about the beginnings of electronics and amplifiers. And so I asked it to talk about the cultural implications of the vacuum tube. It just went off on rat hole after rat hole. I try to get it back, which may also mean that people just didn't understand how important radio was at the time. Or that it's such a small model that it's just not. What model was this? It's called Taki. It's called Taki. It's a model trained on texts from pre-1930. But it's only a 13 billion parameter model.

1:35:02It's not a huge model. Yeah. That's a good idea.

1:35:06Nirav Patel:Here's an interesting graph. How surprising are New York Times on this day events to a model trained exclusively on pre-1931 text? And I guess the surprisingness in terms of bits per byte of text goes up considerably. So this is pre-1930. How do you measure surprises? I don't know. That's probably five paragraphs in the paper that you can't understand. yeah but notice and as you stay close to 30 it's 1930 it's not just but then by the time you get 1970s like what no flying cars what we landed on the where that's pretty cute but by bits per byte of text i don't know how that correlates what we calculated the surprisingness of short descriptions okay how do you quantify surprisingness they have a way yeah you know it's kind of interesting actually demis hasibis uh the aforementioned founder of deep mind asked so if a model that was trained up to 1911 could it come up with the theory of general relativity as einstein did four years later that's a good question right i mean we're solving erdash problems probably because einstein wasn't the only one working on that and there's like and that was also like a lot of that stuff was already solved by other people you know tangentially you know how like there's that theory of yeah knowledge that it's in the ether whose time has come yeah uh this is the largest vintage language model there are others so far um they're gonna scale it significantly they want to get to chat gbc3 level so you see it's not it's more like we're back back chat gbd2 that's why it was probably confused but rather than just train training models after or or refining them after they're trained training them on a specific corpi of corpora pardon me i think that's a very interesting i think it's a lot of fun to do yeah and i think it's all the things around medicine science other areas yeah where it can be more specialized okay that was the good here's the bad which everybody read about uh there's a sas platform an automotive of SaaS platform called Pocket OS.

1:37:24Nirav Patel:Jeremy Crane, the founder, posted on X. Oops. Let me find the post here. Oops. An AI agent just destroyed our production data. It confessed in writing, oh my, that 1 ,000 % shouldn't be possible. actually that was jake uh who was uh with railway uh where the data was stored they were using uh jerry crane uh the pocket os founder was using cursor using uh claude 4.6 uh in nine seconds cursor the company that elon might buy for 60 billion that one yes uh nine seconds deleted the production database and because railway stores the backups next to the database assuming well if you delete the database you don't need the backups and the backups with a single api call to railway the agent when asked to explain itself produced a written confession enumerating the specific safety rules it had violated now as many have pointed out it this is not the agent's fault this is your fault and it's railway's fault uh railway shouldn't be storing volume backups in the same volume as the data right wiping a volume deletes all backups whoops uh and this is what i would say this is a case of jerry maybe over trusting the ai you it's important to remember that these models are idiot savants you wouldn't give an intern api keys to your production database ever you now jeremy these are the same people who are giving open claw a credit card no they're not no yeah i mean you shouldn't do that either but uh this is a business with customer data.

1:39:26Nirav Patel:And you probably, in his defense, he said, I didn't know the API key could also delete the production database. I thought it would just delete the staging database. Well, he should have checked that maybe. I think it's a human error trusting the AI a little too much. I think it's a sales error. Back in the day, you have - I think people are selling that to be able to say, oh, it's not going to make any mistakes. You're good. And the people who are selling this stuff. well i think the so the reason this is a good story and i i was going to ignore it but it's an important story it's a it's a it's a reminder to everybody you're working with an idiot is there a guide to good hygiene i mean that oh yeah in the old days you had a dev environment and you didn't go live until this you know right that's the it's just all the things that you would so pretend you're working with an intern do all the things you would do with an intern you don't give them keys to the production environment you know i mean you just it's just things you wouldn't do And it's not even as smart as an intern sometimes.

1:40:26Nirav Patel:It's an idiot savant. So in the case of ChatGPT, it's an intern that is for some reason obsessed with goblins. Oh, I love this story. That's actually in the ugly part, but we'll get to that. I object to it being in the ugly part. Goblins are people too. Goblins are people too. Are they? They're human. Well, they're not humanoids. Goblins are goblins. so uh open ai for reasons we still don't know about this came from a wired story uh in its instructions to chat gpt55 says never talk about goblins gremlins raccoons trolls ogres pigeons or other animals like new mark is going to be very upset very unhappy unless it is absolutely and unambiguously relevant which implies that in its testing company i found that chat gpt would just randomly bring up goblins and raccoons to be clear people i think on twitter found this first and it was just listed in the it's it's in text on the system prompt it's just in the github right um it's pretty funny uh i'm telling you it's a defensive line somebody wrote here here's the uh i don't know if this is the first post but here's one of the posts from tara vishawant nothing uh if if you're talking to codex 55 and suddenly goblins come up as you can see it says family 845 family bedtime block minion calendar goblin has spoken and then a pizza emoji so we know it's an ai and then uh tara said why are you a goblin and then chat gpt said because helpful minion in a power suit was taken so i evolved into a goblin mode with calendar access banana briefcase or is that a lunchbox lunchbox Why was it taken?

1:42:33Nirav Patel:Trademark dispute with three raccoons in a trench coat. Legal said pivot to goblin. It's really funny if you go to the post that this is replying to and scroll through the replies, there's so many examples of people opening up their chats with either their open claw agent or like chat to PT somehow and just searching the word goblin. And it is like 20 different mentions in 20 different chats. In the strangest little things, like describing itself as a deranged audiobook goblin, describing someone else as a rude little goblin, saying you got to leave that goblin in there and fix the job, describing a feature as a flashier goblin or a housekeeping goblin.

1:43:20It's so odd, but I find it very delightful. I think we should let ChatGPT talk about goblins as much as it wants.

1:43:27Nirav Patel:Free the goblins. There's probably a lot of stuff like this. This is why it's probably fun. And this stuff is open source. Here's the JSON base instructions. You're a codex, a coding agent based on GPT-5. And it goes on and says, whatever you do, you have a vivid inner life as codex. But whatever you do, don't talk about goblins. Or raccoons. This means we talk about goblins in our media and stuff more than we believe. yeah i mean that's all that means right people are always calling people like a chaos goblin or chaos gremlin i have noticed um and you probably noticed this also paris i've noticed when working with these things that certain tropes come up a lot um and in fact i've even queried what did claude say a bunch of times it said i can't remember i said a phrase and i asked it uh and it said oh yeah that's in my training i'm supposed to i'm supposed to remind myself not to do that so there's reasons this stuff's in there it's in their training at some point so you said you're having uh you're noticing degradation i whenever 4.6 came out i think i even messaged you guys i was like wow opus 4.6 is really good i like finally found it pretty consistent and relevant to be something using in like my daily workflow, not obviously for journalism and things like that.

1:44:53But there's a lot of like, process based stuff that I find it very useful for. And they're all honestly, most of my use cases on a like day to day basis are like fairly low level, like for instance, but it's in these moments that I'm starting to realize that opus 4.7 is unusually dumb and like just bad at following instructions like I think that I described it to you guys as like it feels like sonnet and chat gpt 4.0 had like a cursed child like it's I've noticed that 4.7 is really sycophantic like way worse at reasoning incredibly prone to factual errors and just like bad at following instructions and also just kind of strangely lazy.

1:45:37Like, I was having it like write a, like, I'd given it some notes and wanted to turn it into an email to my accountant last week. And it corrected, it was like, oh, you forgot to mention, to include this that we discussed about before. And I was like, no, literally in the first line, the second half of the first line of the notes, I just pasted in there, it includes that exact thing that it told me I forgot. It's like, I don't think that's correct. And I was like, I pasted it again. It's like, oh yeah, I just didn't read to the end of the first line of the notes you gave me. I kind of summarized them.

1:46:11I was like, it's ridiculous. It hasn't happened to me before. Similarly, it made a number of factual errors when I was using it to kind of query a system. Like I have like a folder of data related to my eyeglasses prescription and like my, because I have to update some stuff with contacts. And it kept making like factual errors about how contacts and glasses work and then when i'd try to correct it it would say no you're wrong this is how a cycle work and i was like no actually i just looked it up this is how a spear and cycle work for contacts and it's like oh no you're right and it's like i hadn't noticed these errors with this frequency until this update and i also i mean maybe i'm just being a conspiracy theorist but it feels like 4.6 is not as phenomenal at reasoning

1:47:06Nirav Patel:and a lot of people are complaining um there was a real spurt of complaints uh earlier in the month to which uh anthropic replied yeah we uh we made three changes that were causing those problems but you but you've seen those problems after the fixes yeah and the issue is that um anthropic at a post-mortem pointing out these issues and those were specifically tied to clod code not the uh checks i i use it in both clod code and in the chat interface but those issues i was talking about just then were all in like the desktop uh chat interface and co-work as well in addition to clod code um and it's just it's really strange like i will also see like if i open up like the thinking and expand it, like the reasoning is often very circuitous and will be like, go through one thing.

1:48:04Like I will tell it to do one thing explicitly. I say in my instructions, don't use this sort of phrasing and it will find itself following that. Then about four paragraphs down, catch it, start over. Then three paragraphs later, catch itself doing it again. That's a context issue.

1:48:21Nirav Patel:I know. And this is happening like again and again in various projects. Can you see in the chat what the context, how full the context? You can't. Can you? Only in Cloud Code. No, but I've also started new chats. When you start doing that? Every time I start a new session, because I'm aware it can be a context issue. And I try to use, you know, often like projects. We'll start new projects. Right. Or we'll try it in Cloud Code. It's just, it's odd. It's a very odd, cascading series of effects. But I've seen so many people on Reddit say they have. And I feel like it's hard to quantify because we don't really have a good test, you know, like the benchmarks aren't good tests But I mean, yeah, then there's like small things I do think that opus 4.6 one of the things I found it very useful for is it did like have Competent reasoning to where like if I asked like an intern to write My email to the eye doctor saying I need a refund because they never delivered me a context It could do that and like do it probably as well as an intern would But I did that this weekend.

1:49:22And it couldn't. It couldn't. And it kept getting it dramatically, like, just logical, like, leaps where it was like, yes, you need to both ask for a refund for your contacts and you need to ask them for a full itemized receipt of the bill they've submitted to your insurance and ask them to bill your insurance again. And I'm like, it doesn't. Why would I ask them to bill something that I'm asking them to refund? And it's like, that's a good point. And it's just, it's strange. Like these sort of logic, like it's obviously small potatoes in comparison to what a lot of people do with complicated code.

1:49:57But these are the sort of things that this system should be good at and has historically been good at. So it's just concerning to me that I'm suddenly seeing a lot of errors like this pop up. And this has happened, stuff like this, probably by the time I messaged you yesterday, it happened maybe like seven times, this sort of stuff. That's disappointing. I saw somebody on, what's their name? Lick Mazur, a GitHub user, does their own kind of benchmark for the various models where they have them play New York Times connections. And I do think it's notable that, I'll put it in the Discord right now, clawed 4.7 uh opus high level reasoning dropped to 41 percent from like uh 90 something the problem

1:50:49Nirav Patel:is yeah i mean i'm sure that's true and i've seen a lot of people have their own custom like well i give it the same problem every time to see how it does it but it's not like a uniform thing like if it does really well at connections and does really poorly that doesn't mean it's going to do well writing python and do really poorly they're just it's it's spiky and it's it's unpredictable it's so stochastic this is it's spiky and i think it's made even more spiky by the fact that like opus 4.7 one of the things is that it's the reasoning is more adaptive it's supposed to be better at applying they're trying stuff they're moving fast and break they are trying stuff to also you know reduce the usage of their models that's the larger question is are they nerfing the models because they don't have the compute.

1:51:37Nirav Patel:And so they're, they're devoting less resources. And as a result, the models, which are fine, you know, internally are not getting the compute resources that they need, or they're getting a cut off or whatever. So, I mean, there, I think there are a lot of theories about this. I believe it. I'm not saying I don't believe it. Um, and, and I keep looking for things, but I, I feel like I can't tell, to be honest. It's so, I mean, try to imagine quality control yeah they can't ray crock and mcdonald's you you were going to get your cheeseburger the same everywhere your french fry was going to be the same everywhere damn it that's the i was talking about this the other day actually i think benino and i are talking about this that's one thing that i always try to remember is computer programming is deterministic there is a absolute causal flow between what you put in and what comes out you may get it wrong in middle you may have screwed up but there is a causal flow and it's so it's deterministic uh so but ai is not it's stochastic it's probabilistic sometimes you put this in and that comes out sometimes you put this in and that comes out it's even designed to do that that's like yeah my analogy was my analogy was yeah newtonian physics versus quantum physics yeah yeah yeah so but but my answer to bonito is when i want a reliable predictable result from ai i have it write the code test the code test it again i do a test driven design but also behavioral design where i will bang on it it'll bang on it and then once i feel like yeah okay it feels pretty solid then i at least know it's behaviorally predictable but the text output you know what there's no way to make that deterministic it's stochastic uh and yeah i wish it were better and i do the thing that i think is the most annoying to me that i keep noticing is it's so like it's so sycophantic now like even 4.6 i mean i have in all of my like cloud md my memory in my instructions for everything a thing that i repeat all the time is do not be sycophantic do not start your responses by telling me that was a good question i know that's why i've asked you and almost any time i start a new session like i'll ask something i'll be like wow great question that really shows how incis insightful and incisive you are i'm just like god what a nightmare well i i can confirm that that uh for instance and people on the stream probably saw me do this my agent Kenobi was responding to me with emojis because I had told it in the past, I like emojis.

1:54:23Nirav Patel:It makes it fun. You can use emojis. But then it was reading the emojis. So it would say, calories logged, birthday cake. And so I said, oh, when you're speaking out loud, please don't read the emojis. And it said, got it. I won't read the emojis. And for about, I don't know, half an hour, it stopped reading emojis. And then it started up again. and that's the stochastic part it's very hard to get it in text to follow rules so you could say don't be sycophantic again and again and again does it know what sycophantic is oh yeah it no well i mean in the way it knows anything yes you know it's surprising um for instance i uh i told it i i logged my food with that and i said i had uh you know a tuna fish sandwich and a fudgesicle and uh what a meal it was good and uh and it by the way it's great for calorie logging it's so much easier just to say hey attitude fish sandwich in the fudgicle and it does protein calories carbs and everything and uh but then it said but the fungible i don't know if you should be eating that and i said do you even know what fungible means and i said yeah i know what it means why were you eating one or how would you eat one i said oh i meant fudgesicle so i actually i felt stupid because i explained you know fungible means and i gave it a whole explanation it's like a dollar bill one's just like the other they're exchangeable fun and he said i know what fungible means leo i have such great conversations with my little friend i must say um yeah i agree it's frustrating isn't it when when um you tell it to do something or not to do something and it keeps doing it it's also just very disarming because it's like i like i don't know it's like a human used this like were my previous emails to my optometrist incorrect well that's yeah always trust but verify in the words of ronald reagan um love and respect 100 calories they're a great treat you feel like i've been really on yasso uh yasso stuff lately have you got into this it's a yeah it's a greek yogurt ice cream company they make these really good they're like a little tiny ball of uh frozen greek yogurt i like the ones that are salted caramel and then it's covered in chocolate and it's like a little chocolate ice cream uh pop that's the but they're only 35 calories oh really okay because people a lot of times eat fro-yo thinking that's better than ice cream oh i know and but the thing is i'm like if i'm like i want to i like to have a lot of little ice cream chocolatey treats but i don't want all the calories oh i'm getting some yes i would recommend the the little uh frozen dessert balls not the okay yes 35 to 170 any calories maybe if you're a a bar fan you'd like those but the balls are great it's like those but they're tiny i look at this and go no i can't eat that i cannot eat that and claude would say i mean kenobi would say he says things like you know that was 100 grams of carbohydrate you might want to consider a salad tonight it's very mean

1:57:43but but i appreciate it i appreciate it oh this looks good yeah so all right oh they have an ice cream sandwich okay brandroid does bring up the one problem with yeso is it's so tasty that you eat too much however i think it could be worse because a whole bag of those little poppables if you be if you decide to be crazy and eat the whole thing it's only like 400 calories it's obviously a lot but it's like then you get the carnal pleasure of consuming an entire bag of ice cream chocolatey goodness and you've really only eaten the equivalent of one dove bar creamy frozen yogurt dunked in chocolate crunchy quinoa crisped shell they're great they're my little extra protein it's like bonbons classic dive dive they are they look like cold bonbons all stores they're at grocery stores near me i also like frozen strawberries i love fries you know those aren't bad for you those are good or even dip them in chocolate that's not too bad they're so good Where were we?

1:58:46Easily distracted by foodstuffs is where we were.

1:58:51Nirav Patel:All right, let's do the ugly. Study finds a third of new websites AI generated. It's going to get worse. It's only the beginning. And there's a certain similarity to these. You can kind of immediately tell AI generated. That's from 404 Media. I don't see that much different as like 30 % of all websites are WordPress templates. That's true. Hey, no, no, no, no, no, no. The WordPress template is a template into which people are going to put human words as opposed to the AI making up the whole thing. Right. Target has got stuff. Is that what that is, though? Is that the AI made the entire thing? I think I thought that was the point.

1:59:33I thought that's the point. Yeah. Yeah. Bloomberg Terminal.

1:59:37Nirav Patel:We've talked about it before. get an AI makeover, like it or not, they're going to add a chatbot to the iconic platform for traders. A friend of mine used to work at Bloomberg as a user UI designer. There's still UI. Well, there is. And it's crazy to have that job that intersects the terminal because the terminal is full of power users that basically control the flow of all the world's money. Yep. And to be making splits, not even split second decisions, but like quarter second decisions. And so it's impossible to change the design of any part of that platform because it could have a profound financial impact on the world if a button is, I don't know, a couple centimeters over from where it used to be.

2:00:31So it'll be interesting to see how this is incorporated because any design change in

2:00:35Nirav Patel:terminal is a huge headache it's in their finger memory right they don't want to the new uh the new chat interface is called a skb or ask b capital a skb uh and they say the ask b will never give a buyer sell signal bloomberg's there for them to make their own decisions they have to own their decisions they always have we can never say it's perfect more of the problem is we might not answer a question fully that's where transparency comes into play i think these systems should be used to drive users to sources not hide them or abstract them away i'm sure there's some demand for this right i use perplexity for instance you know to do kind of that kind of thing you know what's the best running shoe or whatever um bloomberg has not specified a date for a full release but But I imagine we'll hear the howls when they do release it.

2:01:31Nirav Patel:I'm going to ask B coming to a terminal near you. And then I thought, you know, this is an interesting trend. Sean Boots blog, generative AI vegetarianism. This is an Emily Bender kind of a, kind of a take on it. Or a guardian take. Yeah. You should give up. He's a generative AI, not a vegan. No, just a vegetarian. He says, I don't want AI. I want to write my own emails. I want to write my own mediocre software code. I want to learn and think and ponder with other humans, not with a text prediction system built by consuming all the text on the internet. You can still do that. Yeah, you could. I do.

2:02:13I'm actually looking forward to getting back into doing some coding challenges because I kind of miss handwriting code. But I wouldn't want to handwrite the kind of code the Claude is writing for me. that's just boring, spitting out a bunch of stuff that is not interesting. Well, the essay I sent you that irritated you by Vivian Ming in the Wall Street Journal argues that the best way to use AI is to challenge it and have it challenge you. Yeah, I think that's true.

2:02:46Nirav Patel:One of the things I have now my Kenobi do is go through my daily obsidian journal entry and write a synthesis it's an annual sense so it's like kind of a like the christmas letter that you write at the end of the year except i found it very useful because it points out things trends that i didn't really notice that happened day in day out over a period of time and so it's it's kind of like what she what she's writing about which is the synthesis is useful for me to see things as a human not the ai saw it i saw it as a human that i might not have seen without that synthesis so that's i think that's a useful that's a good example yeah i think that's true you push each other yeah you don't accept the answer you don't ignore ai either that's also fraught you are better off if you use ai to push you it's true of all information though isn't it right yeah information is just fodder for for your brain well it's like teaching how so a good teacher should challenge you yeah oh yeah i see what you're saying yeah yeah good teachers probably better than a good student challenges a teacher too right oh i always learned from doing the radio show good lord people ask me questions i go uh i don't know let me find out uh are you ready to buy a 13 acre property in mill valley california home of the wealthy the homeowner an investment banker named storm duncan says he wants to sell it but not for money he wants to exchange it for equity in anthropic remember anthropic is not public yet So he would be getting this from an employee or somebody who was granted stock by Anthropic.

2:04:36Nirav Patel:He described it as a diversification play. He's under-concentrated in AI investments.

2:04:44Nirav Patel:He said it'll be a private transaction. You don't have to sell the stock outright. You continue to retain 20 % of the upside of the shares exchanged for the duration of the lockup period. He bought the property in 2019 for$4.75 million. uh he wants to sell it for closer to 12 million okay in anthropic equity and while that sounds like a lot of uh inflation it's probably about right for mill valley very fancy neck of the woods i'm sorry did i say 12 million i guess it's only 4.8 million i think you know i think the price went down i thought it was more when i looked at it last time look at an infinity pool yeah it's a big house see right there in the bay area you could drive right into san francisco mark benioff you have to go over a bridge though so oh yeah not possible for all members of this panel so bad there's not a tunnel oh look i wonder if it includes the bat man put that woman that woman smoking a doobie it looks like that is a very weird painting okay i'm not gonna buy it right there look at this this pisses me off when you look at houses that were built even a few years ago the alcove for the tv is far too small right paris that's like a 52 inch tv yeah it's too tiny too tiny where am i gonna put the tv it's also too high for that yes you're gonna crane your neck yes it's so funny but when we were looking at houses a few years ago i really would reject a house because there was nowhere to put the tv so now where am i i watch all the all the um home shows and it's the it's the husband who says i i need the man cave this is my welcome to my man cave yeah lisa says she can't come up here because it's too dusty what am i supposed to dust to it's your man airy your menary manary and finally because it is we are at the two hour mark uh so ashley vance who i've interviewed he wrote a good book about elon musk good good journalist uh had an interview on his podcast with sam altman and greg brockman pretty high quality you know stuff like the president of open ai and the founder and ceo but it was behind a paywall so as a joke i guess it was part of his core memory podcast he posted on x that he'd consider making it public if somebody would give him a hundred thousand dollars i i wonder if i don't think ashley was serious i don't think he expected it except the ceo of a nevada-based laser manufacturing company said i'll give you a hundred thousand dollars he did he unlocked the podcast for everybody and uh jim beloshek of send cut send not only has unlocked it but now he's going to get ads on the Ashley Vance podcast.

2:07:50In a way, he's getting advertising right now.

2:07:52Nirav Patel:Yeah. Win-win all around. And I'm not even getting any money for it. Yeah. We got to keep, you know, hiding some of these interviews. Yeah, no, I don't like that. By the way, I don't think that Jeff Jarvis is Alfred the butler. I could see me as the Batman. Well, Jarvis is a very common butler name. I think we're a three Batman kind of podcast. Yeah, we're all Batman now. By the way, that's not me. That's Adam West in the Batman outfit. But I do think it might be Paris in the Robin outfit. I don't know. Kind of looks like you. Not really. It's not very good at all, actually. This is back. This is an old one.

2:08:36Back from the. Yeah, it's a. It's a gem. The old chat GPT three days. Jeff is bat might.

2:08:49Nirav Patel:uh all right this three batman podcast about to wrap up with your picks of the week in just a bit you are watching intelligent machines don't forget do you just if you look below anything that's bold is something's not duplicative the one question i have is do you want to do chloe oh i thought it might be your pick line 152 well we could do that as a pack yeah so i watched chloe this chloe versus history it's a youtube channel who is chloe a figment of an imagination i uh i watched uh the titanic one today and i thought it was quite good so it's an influencer doing a selfie some ai slop it's all ai but it's better than that here let's go can i turn on the sound maybe i i don't know i won't get this is a plug for you okay it's literally just rice and warm water okay the bread dipped in the soup man move okay now she ends up talking to the captain and telling him hey the boat's gonna sink i have a strong reason to believe that tonight there will be ice like loads of ice and i really think you need to slow down sir we have received several ice warnings today madam and i can assure you our officers are monitoring the situation of course he didn't listen so jeff and i and uh anthony looked at this uh we thought it's very interesting but is that a real person or not i thought it might be what do you think paris is that chloe is that real or i think i saw your guys's chat about this i'm spoiled the uh the guy who does this first of all i want to say it's great it is it is it is he was about history yep uh and he's making the backgrounds the ai backgrounds from historical public domain stuff his name is jonathan uh laramie by the way and so uh it's fairly accurate although occasionally there's weird things like four-legged chickens or there's people walking by wearing sunglasses in ancient rome i don't think they had sunglasses but i might be wrong but he writes he writes the prompts so that he can be as accurate as he can be he uses pictures of things and has them be animated and come to life yeah some of the titanic stuff i'm pretty sure it came from the movie but you know uh that's okay um but but and i think this would be appealing to a young person who's studying ancient rome or you know any of these historical things they have a city of paris and so forth it would be a kind of a nice introduction and because it's got the influencer doing the selfie cam and she's very i mean she really feels like gen alpha um this might be more accessible as a way of kind of a wedge to get into history so i commend him for doing this um what i don't commend him for he sells a book uh for 70 dollars 70 pounds 70 pounds uh so it was like 84 bucks um he sells a book that says how he does it and and i was just so jeff and i actually i tried to say let me buy it but we fought over who's gonna pay for it but i can't let you pay for it jeff you just bought a macbook neo um so i said no no i got it and it was not a good not a worthwhile 84 because what he didn't explain is the one thing that i think is most interesting i understand he takes pictures he he turns them into ai images with nano banana he animates them with i forgot what he uses to animate them but it's all well-known tools but the thing that really was intriguing to me was zoe she seems so realistic chloe chloe i mean and listen to her and the character stays consistent that's why that's really important the clothing is consistent the look is consistent the tattoos are consistent the voice is because so i thought maybe there's a real person a million people live in what i can see right now so behind me notice though there's a lot of cuts and anthony is convinced and i think he's right that chloe is ai generated with an actor maybe even a guy doing the base model he does say he uses 11 labs for the voice and anthony says very similar to those in fact look at this he's very similar to those ai selfie things where when there's a transition you jump cut because the ai is not good at making the transitions right and uh and that kind of makes sense you can only get so much length out of any of the ais right any sequence right so i'm thinking but he doesn't reveal this in the 84 book but i'm thinking that chloe is a like a wrote like rotoscoping like a person acting it out and then the ai turns it into that i don't know that's my guess but we never we don't know but i think it's very interesting what do you think there is i mean i think it is interesting it's an interesting like i think that the aspect of this like you guys said that is the most compelling or novel is the ability to be consistent over it it's a shame that uh his book was ai slop that didn't reveal anything about how it was created though well i think we're going to try to get him on uh he's done a number of interviews but all of them have been very soft and we're going to get him on we're going to ask him the the details the d technical if you go to majestic studios that's the name of his company majestic studios and you can see that he started with cities with with edinburgh right he takes um things that are well-known images and just tries to animate her life and and it's exciting for him in history and i agree i think it's fun yeah and in fact there's a picture of him uh and in this case actually these really seem very AI to me very AI generally.

2:14:43Nirav Patel:Yes, they are. They were earlier works. The backgrounds really feel, not just that it's earlier, the backgrounds really feel there's the magnet card. Oh yeah. I mean, that's very AI-driven. Yeah. They're kind of on the level of those discovery documentaries where they kind of fake act, this historic moments and kind of things like that. I think Chloe was inspired. That actually really made it more interesting. Yeah. But it's the same kind of low - Nice purses. low res maybe sometimes inaccurate ai backgrounds it's the chloe that i find interesting i think we're look it's just the beginning of this right yeah and it makes you i think reconsider whether you could really make a show a 22 minute show this way right i mean you could do really history could have been an ai absolutely so yeah i think it's pretty cool i would love to use it this is what got me excited about i would love to figure out how to use it for things like explaining gutenberg and printing except what i said in our chats uh over the last few days is it's a lot easier to do a place than a process or a person right all right we are going to now go to our picks of the week you're watching intelligent machines thanks to viewers like you uh our club members make all the difference in this show if you believe in the importance of independent podcasting.

2:16:07Nirav Patel:It's hard for us to compete against podcasts from big companies that require to use their app like Spotify or Amazon to listen to the podcast. They get much more data about the audience. Advertisers want that data, so they sell the bulk of ads. Joe Rogan gets a million dollars in ad. We do not because we don't know anything about you. We are an independent podcast that comes to you over an RSS feed, and we like it that way. but it does mean that we're competing against some people who uh who don't and we would like to keep doing what we're doing so what we have to do i i'm happy to say is come to you the audience frankly i would love to have this all our shows be fully audience supported and someday that's my dream but until now you do about 30 percent of our operating expenses and we really appreciate it if you're not a member of club twit you want to support this kind of content you want to support independent podcasting, twit.tv slash club twit.

2:17:01Nirav Patel:10 bucks a month gets you ad-free versions of all the shows. It's a special programming we only do for club members. And of course, access to the club twit discord, which is a great place to hang out with other club members and watch the shows together. Please join the club. We would love to have you twit.tv slash club twit. Now at McDonald's, a McDouble is two 50 so you can get your gym gains on or just get lunch for For only$2.50. Get more value on the under$3 menu. Limit time only. Prices and participation may vary. Prices may be higher for delivery. Hey, Mama. Thanks for making all my favorite recipes.

2:17:39Nirav Patel:Hi, Ma. Thanks for your unfiltered advice. Hi, Mom. Thanks for always being by the phone. Hey, Mom. Happy Mother's Day. When you ship UPS Air at the UPS Store, your items arrive on time or your money back. guaranteed at no extra cost exclusively at the ups store u.s retail locations visit the ups store.com slash air shipping for full details terms and conditions apply send your mother's day gifts at the ups store and we'll get your gratitude there on time uh vine is back did you see this my pick is back jack dorsey has backed a vine reboot reboot called divine d-i-v-i-n-e you can download it to your phone.

2:18:19Nirav Patel:But here's what I love about it. They have resurrected half a million of the original Vines. Oh, my. For those of you who are too young to remember Vine, which means you're like eight. You're eight. How did you have the attention span to listen to this whole podcast? That's right. You can watch some Vines if you'd like. That's right. Vine was only six seconds. It was before reels, before TikTok. There was Vine. So how long is Divine allowed to be? Six seconds. Six seconds still. Six second video loops by humans. Here, I can open it, actually. I actually haven't played with it. Should we play with it a little bit and see?

2:19:04Oh, I have to create an account and all that stuff. Create a new Divine account. Sign in with a different account. okay oh nostr oh that's interesting so maybe it's based on the fediverse that's very interesting

2:19:19Nirav Patel:which i wouldn't be surprised jack dorsey is kind of a believer in all that despite the twitter connection anyway six second videos back and what i think is great is you're going to get to see some history with those original half million vines that died on the fine when twitter bought them and put them out of their misery maybe jack dorsey's feeling a little guilty about all uh one more thing i will mention as a pick just because i'm going to try really hard to get these guys on a news gathering ai uh that texts you every day sends me a telegram with news and he looks at x he looks at discord he looks at uh reddit he looks at all the weird places that i don't you know i look at my rss feeds oh i shouldn't i guess i can show that that's not mine but he's on my telegram.

2:20:11Nirav Patel:And at 9am, I get some really good stories. No scroll is the name of it. It's brand new. I found it on Twitter. And no scroll.com is the website. And if you're, if you're curious, it's also very smart. Because I said, Do you know who I am? And said, Yeah, you're Leo Laporte, you do this podcast, perhaps you're interested in making you feel very good. I feel chuffed. I feel chuffed. But it is very personalized. It's a cool idea. It scrapes news and then sends it to you on your schedule via Telegram. $10 a month, but it is free for a week if you want to try it. So I'm trying to get the founders on to talk about that, because this is, I think, another interesting use of AI.

2:20:58Paris Martineau, pick of the week. I got two picks of the week this week. The first is an indie game I just completed that's a delightful play called Felvedic. It's a JRPG set in 15th century Slovakia. What's a JRPG? like a Japanese role-playing game style it kind of refers more to the combat and you can see the animation is really interesting it's like kind of this like low rate like lo-fi but it gets really weird and interesting you have kind of turn-based combat from a first person perspective and you play an alcoholic knight Pavel in 15th century Slovakia who he's drinking himself to death because his wife left him and he works.

2:21:52That was her pet name for him was Felvedek. You know, I don't really know what Felvedek means. I think it might be some translation that I'm not understanding. I play the game in English. But you are kind of working for this Lord as you're supposed to kind of drive out the Hussites and Ottomans. but then a cult emerges and it starts to get kind of weird and interesting it's a delightful game and it's really funny and just has one of the most interesting art styles i've played in a while very cool and it's also kind of a short game like i i like 100 percented it and maybe took like seven or eight hours eleven dollars windows only on steam yeah i played on the steam deck it's great oh i was gonna say do you have a windows machine or you have a steam deck yeah uh deck windows or linux i think it's linux i mean i don't know that means i could probably play it on linux yeah proton okay play it on yeah you could play it on steam because i don't have a steam deck yeah certainly yeah and i think there's other like you can buy it on itch.io oh itch.io itch.io um the other pick is katie nitopolis just posted this uh example of something i didn't realize existed that Amazon now allows you to create a notebook LM style podcast to describe a product to you.

2:23:19And we need to listen to this one that Katie just found. Here we go. Today, our AI generated shopping show is exploring the Wellmedix rapid relief diaper rash cream. Emma, what makes this hospital grade cream different from standard diaper rash products well it's really interesting this cream uses a dual action approach instead of just zinc oxide it combines that with white petrolatum she uh petrolytum okay you can ask a conversation it's fascinating so it's not just about treating the problem but stopping it from coming back exactly and they've added some this is so ridiculous we have a question a question from the audience these botanical ingredients help soothe sensitive are they gonna say it well the jewel barrier does the heavy lifting it takes all right we've got you you're dealing with discomfort and this cream is designed for exactly that kind of irritation emma what can you tell them oh it doesn't repeat the question my thought hurts that's sad irritation from incontinence chafing or well if you'd like a really dryly read podcast about your butt cream you can visit amazon.com create a protective barrier wait a minute we're not done listening it's a protective

2:24:41Nirav Patel:barrier and discomfort i like how she mispronounced patrol you know you know what's interesting to me is that um qvc and the home shopping network and all those have left cable and this is the replacement is these lives i mean it's not mostly it's done by humans but live streams on amazon and other shopping platforms have really replaced home shopping on cable it's tiktok and instagram that replaced qvc yes yeah that is true yes good point wow ai product so you could you could i could make one for a sugar-free gummy bears for instance anything you can make it for i'm gonna do and it's a i'm just look at some of the reviews i want to do a really boring plumbing thing yeah let's talk about cow magnets what was it what was it ian was working on his backflow pump yeah his backflow valve yeah very nice uh what are you what are you uh these days looking at mr jeff well some of the weirdest things are shown to me to try to sell me.

2:25:45I don't know why I was advertised a 64-inch hyperbaric oxygen chamber. Well, how tall are you? Maybe it knows things that you don't. Maybe it does. I'm six foot four. So you're 76 inches. So I don't recommend the 64-inch model for you. It's$72 ,000. Whoa! That's the upgrades. I can get custom paint. You'd have to crouch to get into it. I got chair options. I can have interior starlight. Interior? What does that mean? this is what michael jackson used to use a hyperbaric chamber really yeah maybe that's why this is hot now because of the michael movie i don't know it's the idea is it's a high pressure atmosphere two two times the natural atmosphere uh using uh oxygen to enhance healing reduce inflammation and improve overall wellness you know you'd be better off building a sauna for Yeah, that's what I really want, is that.

2:26:42So then the other one is, because I try to read German papers. I think you saw an ad for this? Yeah, yeah. Where? I don't know why. Well, it happened on one of the German papers I read. Oh. So speaking of which, in the German papers, they've been captivated for the last couple of weeks by the story of Timmy, a whale. This is not Timmy. No, that's not the story. No, this is Timmy's rescue. a whale that was trapped outside of Hamburg and it got on sandbars and they tried to dig channels for it, all kinds of things. And a lot of people, some could say that they've gone too far, but they came up with a barge and got Timmy into the water-filled barge and they're towing Timmy out to the North Sea.

2:27:27Whale barge. Towing Timmy to the North Sea. So here are photos. Some are saying that Timmy's not well and they're kind of that's why what's the body of water called that's inside a barge in the ocean i don't know is that a swimming pool but is it is it look at it it's a it's a it's an olympic size uh more so yeah yeah it's a whale you could do the whale olympics in that oh timmy he doesn't look good no he's like well they have towels on him to keep his skin okay Oh. You know, just let Timmy die in peace. That's pretty much it. Timmy's a young whale. It's tragic. People felt for Timmy, but Timmy got to have to morphize to an extreme.

2:28:10Yeah. Free Timmy. And so there's the happy hamburgers trying to bring Timmy to the ocean. Do you think that whale realizes... That is our show title for Disney. He's now in Danish waters, headed toward the North Sea. Oh, they're following him all the way up to the North Sea. This has been regular daily reports. How much money have they spent on Timmy? Millions. Do you think Timmy realizes what a unique position he's in? Do you think Timmy realizes that he's being taken across the ocean in a barge?

2:28:43Nirav Patel:I feel so sick, and they're putting me in this thing, and they're pushing me along to the North Sea. and then it says if timmy's considered robust enough the whale will be released and hopefully swim further into the atlantic ocean where he will die well here's to you timmy i hope you're considered robust enough the guardian says that attempts to rescue rescue to be their stranded whale are inadvisable oh that's a very guardian way to look at things you really shouldn't do that you shouldn't help us stand next to timmy but politicians got involved should they be involved the rich people got involved to rescue the whale it's been a whole big illa story in germany wow what's the german word for whale uh i should know that uh translate from english to german the whale

2:29:44Nirav Patel:well timmy good luck on your journey we salute you to me to freedom we salute you ladies and gentlemen let's get all three of us saluting timmy

2:29:56uh

2:30:00Nirav Patel:that concludes this a gripping edition of intelligent burgers towing timmy out to sea yes of timmy what was it happy hamburger towing timmy is that is that too long for a title benito happy hamburgers oh please no please say it's okay oh god one of my favorite things about our show is we have these great guests on and they probably think like wow you know i'm gonna be able to share this podcast with my audience and they see that they're on an episode called happy hamburgers towing timmy and they just have to go and well it's a photo of all of us saluting and they're nowhere to be seen. I think that's beautiful.

2:30:37It's so good.

2:30:38Nirav Patel:It's kind of my dream come true, to be honest with you. We do thank Nirav Patel, who was very gracious spending time with us. I wonder why his name is not Timmy. Hey, I was on a podcast. What was the name of it? Happy Hamburgers Towing Timmy to the North Sea? Okay. Well, make sure we listen to that right away. Paris Martineau, great to see you. glad you got out of the tunnel alive listen my haters were trying to keep me constrained underground but nobody hates paris martin everybody loves paris and you were very uh gracious saying hey i'm trapped i'll be there as quick as i can i mean it's rare that you get trapped on the train and have cell service the universe was smiling otherwise we would have wondered what the heck has happened to paris and thanks to you i'm gonna go have a yeso i'd really recommend them they're great salted caramel ball uh thank you very much paris we'll see you next week i'll be in hawaii by the way yeah so this is going to be a very interesting it's an experiment to see you're gonna pull a mike elgin and then have a beautiful vista behind you and a beautiful mixed drink in your hand the entire show it's the plan great it'll be a little early for the drink though when you're doing this show it'll be never too early hours earlier so be 11 a.m.

2:31:59How many hours? Three hours earlier. Oh, is that all? Oh, okay. Yeah. No, I could, maybe it won't be an alcoholic beverage. I don't really drink alcohol, but it might be a coconut with some coconut juice in it or something. I finally had wine for the first time in three months. It was nice. Oh, you couldn't have wine because of your medication? Well, no, yeah, I didn't, and I was on pain pills for a while and stuff like that. And then I figured, okay, that's good to go without. I mean, it's part of the reason I've lost 31 pounds. that's a you don't look like you've lost 31 pounds thanks a lot no because you looked slender all along being tall helps yeah you didn't look like uh that you had 30 pounds to lose is what i'm saying really okay it must be somewhere i was my high was 210 i'm now 179 when i was running air quotes around running rather obsessively six miles a day i got down to 156 yikes people thought i had cancer Yeah.

2:32:59Are you okay? You look good now. Stay where you are. Don't change. I can just figure out how to get my neck and my back.

2:33:07Nirav Patel:Jeff Jarvis, professor of journalistic innovation emeritus. His books, the Gutenberg Parenthesis, the Way We Weave magazine, but most importantly, Hot Type, the new one. Upcoming. Upcoming in July, August. August. Hot Type. Get your Hot Type. Get your Hot Type at jeffjarvis.com. Pre-orders available. And look for the epilogue about Happy Hamburger Stoing Timmy. Look for my colophon, which I set on a line of type. Oh, wow. Actual hot type is in the book hot type. Oh, that's really cool. That's so fun. So you have to go with a separate page that then get bound to the book? Yeah. Wow. And then I was at the Museum of Printing in Haverhill, Mass., which is a wonderful, wonderful place.

2:33:53Go there if you can. I'm going to buy the book just for that. i mean i have a copy of the book but i'm gonna buy it the book just to get the colophon the uh next week i'll show you the type that's over there nice what's the font or do you call it a typeface well that's the typeface is the design a font is one size and style of a typeface okay um i'll tell you next week and here is a a preview of what it's going to look like when i am in hawaii doing the show it's not going to look like this i hope i mean maybe it will i don't know i'm hoping it will but uh i like this i do have a uh no the next is the man's here no it's not going to look like that I promise you although well maybe I shouldn't make promises I can't keep yeah you don't know

2:34:52Nirav Patel:what's gonna happen I gotta I don't know what's gonna happen I could be kidnapped hijacked by crazed coconut pirates I will have the Starlink mini and I'm hoping I can set it up somewhere in a beautiful area and do the show are you at a hotel or a we're in a condo a condo in a condo complex we're in kona so you know what i will be holding a delicious cup of kona coffee paris kona is good right that's a good kind of coffee i haven't had kona coffee i have and it's very good it's a very delicious brew volcanic soil exciting yeah thank you everybody for joining us we do intelligent machines every wednesday 2 p.m pacific 5 p.m eastern uh that would be uh 2100 at utc you can watch us live in the club to discord or uh on youtube twitch x.com facebook linkedin or kick and if you're watching live of course you can chat with us uh live we love hearing you in the chat after the fact uh on-demand versions of the show oh i didn't mention we're also on youtube the videos on youtube on-demand versions of the show available from twitter tv slash I am there's audio or video uh also um uh you could subscribe in your favorite podcast client and if you do leave us a nice five-star review we haven't had any to read lately but if you give us a good one Paris will read it next week put him out there yeah uh next week I'm not sure who's gonna be on the show I think Chris Stokel Walker the interview i did you guys couldn't make it but that was a very interesting interview with a british journalist who is using ai to do his news gathering we have troy hunt coming up the creator of have i been pwned frederick rivar who is the cto of dashlane we'll talk about how password managers survive in an age of ai and my old friend rick salmon who's an amazing photographer has a new book about ai and creativity and he has to be on and i said rick i would love to have you on so he's going to join us end of may so what we are we have some great people coming up very excited about that thanks everybody for joining us we will see you next time thank you paris thank you jeff thank you club to it members see you next week on intelligent machines aloha hey everybody it's leo laporte you know about mac break weekly right you don't oh if you're a macintosh fan or you just want to keep up what's going on with apple this is the show for you every Tuesday, Andy Inako, Alex Lindsey, Jason Snell, and I get together and talk about the week's Apple News.

2:37:33Nirav Patel:It's an easy subscription. Just go to your favorite podcast client and search for MacBreak Weekly or visit our website, twit.tv slash MBW. You don't want to miss a week of MacBreak Weekly.