1. Podcasts
  2. Intelligent Machines (Audio)
  3. IM 894: Ghosting Meta Muse - Will Cloud Spending Sink AI's Biggest Players?
Intelligent Machines (Audio)

29 Oct 2026 · 2h 20m · 42 chapters

Episode artwork

IM 894: Ghosting Meta Muse - Will Cloud Spending Sink AI's Biggest Players?

In short

The episode focuses on the AI industry’s financial and strategic fault lines, centered on Anthropic’s IPO filing and what it implies about enterprise lock-in, compute costs, and “safe AI” branding. It also covers Google’s new Gemini Carbon model and whether Google can catch up, plus broader debate about AI extinction claims versus more practical risks.

Guests (backgrounds)

Jason Heiner (host filling in for Leo Laporte; TWIT). Jeff Jarvis (Craig Newmark School; author, professor, commentator, podcaster; former critic). Paris Martineau (co-host; journalist/podcaster on tech and media topics).

Key claims

  • Anthropic’s IPO filing suggests enterprise revenue concentration: about $16.3B revenue with 25% coming from five customers, implying high switching friction.
  • Compute commitments are a major risk: the filing reportedly includes $500–750B in locked-in compute spending over 7–10 years, which could “sink” companies if demand or efficiency assumptions change.
  • The “safe AI” strategy is framed as a brand bet: Anthropic positions itself as trustworthy/safe, including highly specific “damage”/harm thresholds.
  • AI extinction talk is criticized as cult-like marketing; more realistic near-term harms are cyberattacks and infrastructure failures.
  • Google’s Gemini Carbon is presented as a competitive comeback attempt, emphasizing cybersecurity and long-horizon agents; Google’s advantage is hyperscaler compute and distribution, but it’s criticized for confusing product naming and engagement-driven UX.

Notable examples

  • U.S. government reportedly labeled Anthropic a supply-chain risk; it took months to stop using it.
  • Mention of OpenAI’s earlier “compute optics” controversy and aggressive pricing.
  • Gemini Carbon benchmark positioning versus Opus/Astra.
  • Google renaming “Spark” into “Task mode,” and NotebookLM being folded/renamed.

Written by AI. May contain mistakes.

Chapters

0:45

Episode Overview and Special Plug

Discussion about the episode's topics and a plug for a special extra episode.

“here i'm jason heiner filling in for leo laporte uh very excited to be here there's so much to talk about, as there always is every week on this podcast.”
3:51

Anthropic's IPO and Market Position

Analysis of Anthropic's IPO and its competitive position relative to OpenAI.

“Back to our regularly scheduled programming, which, of course, is all the stories in AI.”
8:01

Revenue Insights and Market Tactics

Discussion on Anthropic's revenue sources and market strategies compared to OpenAI.

“We even saw that with the US government when they declared Anthropic a supply chain risk and an unprecedented move in the spring.”
9:43

Operational Challenges and Future Projections

Exploration of operational challenges facing Anthropic and future projections.

“So they're clearly just cleaning up with these sort of tactics.”
11:08

Concerns About Compute Efficiency

Concerns regarding the efficiency of compute models and infrastructure in AI.

“And, of course, more than anything, to buy compute and to secure future compute.”
14:00

Concerns Over AI Compute Infrastructure

Discussion on the potential instability of AI's reliance on compute resources.

“could be a bit of a house of cards, which is all of this compute, this giant infrastructure build out, which maybe we'll need all of that.”
16:46

Market Risks and Competitive Landscape

Analysis of market competition and economic risks faced by AI companies.

“own that seem to not follow any of the normal patterns of logic and business.”
18:57

The Exponential Growth of AI Adoption

Examining the rapid adoption of AI among businesses and its implications.

“I mean, I just think it's notable that like we're talking about revenue in the tens of billions of dollars and then we're talking about one line of expenses compute that is in the hundreds of billion dollars.”
22:19

Infrastructure Costs and Technological Evolution

Insight into the financial implications of upgrading AI infrastructure.

“So I think if you take and multiply that by a lot of companies.”
24:14

Risks of AI Companies and Neo Scalpers

Discussion on the risks associated with AI companies and their infrastructure providers.

“You all anticipate all of my questions and thoughts.”
Show all 42 chapters
28:00

AI Companies and Market Sentiment

The discussion focuses on the financial stability of AI companies amid market fears and competition.

“of in the middle of the hyperscalers, they're like too big to fail.”
42:02

The Cultish Power of AI Influencers

Discussing the influence of powerful figures in AI and its impact on policy.

“that we're talking about here, the level of cultishness that we're talking about here.”
42:43

The Cultish Power of AI Influencers

Discussing the influence of powerful figures in AI and its impact on policy.

“With that, let's throw to our first word from a sponsor and we'll hand it over to Uncle Leo to give.”
44:45

Google's Challenges in AI Competition

Examining Google's position in AI amidst competition and its branding issues.

“So for our next story, let's dig in a little bit to Google getting back in the game a bit with Gemini Carbon.”
49:55

Consumer Experience and AI Adoption

Analyzing the consumer experience with Google and the challenges of AI adoption.

“Is it unreasonable to think that Google still has plenty of opportunity to do it for all the reasons we've already mentioned?”
56:00

Consumer Confusion with Tech Products

Explore how consumer confusion arises from tech companies' organizational structures and branding.

“And that's an infuriating customer experience.”
1:06:51

Discussion on Marissa Mayer and AI

Delve into Marissa Mayer's experiences with AI and the implications of gender bias in technology.

“Let me just read verbatim the opening four lines of this.”
1:10:01

Hyper-Personal AI: A Vision Beyond Hyper-Local

The discussion revolves around the shift from hyper-local to hyper-personal AI, exploring what AI should ideally do for users.

“I remember distinctly an event where I was, where she was, and I was going on, blathering on about hyper-local.”
1:12:09

Challenges of Proactive AI Agents

Hosts delve into the limitations and frustrations of current AI agents, sharing user experiences and their capabilities.

“And I mean, I do think that this is similar.”
1:14:49

Comparing AI Assistants: Muse, Duet, and Dots

A detailed comparison of different AI assistants, focusing on their effectiveness in managing tasks and user expectations.

“Which one's the worst, best, and has any surprise to you?”
1:17:37

The Impact of AI on Customer Interactions

Exploring how AI will transform customer service interactions and the potential consequences for businesses and consumers.

“It's like, hey, you still need to finish that up.”
1:21:57

The Future of Human Interaction in AI-Driven Markets

Discussion on the balance between human interaction and automation in service-driven markets and its implications.

“So you have an entirely new opportunity to steal people away from whoever led the field.”
1:24:00

The Value of Human Interaction in an Automated World

Explore how automation impacts human connections and the value placed on them.

“But for the ones that we value the most, we will value them even more when we want a human to actually interact with.”
1:25:50

Meta's Missteps with User Data

Discuss the implications of Meta Muse's data leak and its effects on user trust.

“No, my bingo card did not call on mathematicians and physicists being the first people eliminated by AI.”
1:26:44

Concerns Surrounding Meta's Data Practices

Analyze Meta's approach to data privacy and the challenges it faces.

“All right, I want to ask a question about Metamuse, sort of wrap this topic up.”
1:34:12

Concerns Over Open Models and Trustworthiness

Investigate the reliability of open models and the fears surrounding their security.

“So next, would love to talk about a story about open models, which we talked a little bit about earlier.”
1:38:00

Chinese Cars and Technology

Exploration of the advancements in Chinese car technology and market challenges.

“is now waving a flag again about trying to restrict the export of its models, fearing us, which is an interesting little turnaround there.”
1:40:11

BYD Experience in Israel

Discussion of personal experiences with BYD vehicles during a trip to Israel.

“I'm going to do a slight aside on the Chinese automobiles for a second.”
1:42:44

Open Models and Local AI

Insights into the rise of open AI models and their potential impact on local technology use.

“And there were lots of Teslas there, but there are also lots and lots of BYDs.”
1:46:06

Community Use of AI

Exploration of community engagement with AI and the potential for local control over technology.

“And so the, and then this also plays into local AI too, right?”
1:49:14

AI in Cooking Apps

Discussion on the backlash against AI features in cooking applications and its implications.

“I think it might have an impact on the reputation of AI in the larger population.”
1:52:00

AI's Impact on Society and Perceptions

Discussion on public perceptions of AI, its potential risks, and humor about its limitations.

“So I was there when we started Epicurious at Condé Net.”
1:55:16

AI in Scientific Research

Exploration of how AI is used in scientific research and the implications for scientists.

“Let's talk a little bit about science, because there are a number of stories this week about AI and the value that it's having for science.”
1:58:34

The Dual Nature of AI in Science

A balanced view on how AI aids and complicates the work of scientists and researchers.

“I put it in the rundown, but I doubt we want to run the video.”
2:02:31

OpenAI's Role in Scientific Advancements

Analysis of OpenAI's recent announcements regarding scientific collaborations and discoveries.

“It basically stacks their papers so high that they might not even get some of this stuff for decades.”
2:06:05

Transition to Sponsor Announcement

Transitioning to the sponsor segment of the episode.

2:08:33

Picks of the Week Introduction

Introduction to the picks of the week segment with hosts sharing recommendations.

“Now it's time for our picks of the week.”
2:08:58

Harris's Picks: Local Wikipedia

Harris shares a personal website called Local Wikipedia.

“Or if you're kind of in a big city like me, it's the closest 25 things a lot closer to you.”
2:10:16

Jeff's Picks: Rest of World and Wall Street Journal

Jeff discusses important reporting from Rest of World and a Wall Street Journal article on AI.

“I want to mention, so Rest of World did a wonderful piece this week about how AI models cost more when you don't use English.”
2:10:48

Discussion on AI and Investment Risks

Exploring the implications of AI risks on investments and economic perspectives.

“The other one to mention is the Wall Street Journal.”
2:12:52

Discussion on AI Glasses

Discussion involving AI glasses that work with local models and privacy features.

“It's like that character Hermione in Harry Potter who's like, you know, we can get killed or worse, expelled, you know?”
2:13:23

In-depth Look at AI Glasses Features

Detailed exploration of the functionality and implications of AI glasses.

“pick of the week that i'll share that i wrote about um this week and it's a pair of ai glasses but hear me out on this.”
A man walking through an airport, asking his phone a question

Ask about this episode

Get the gist, explore an idea or ask about something said.

Claude

On Intelligent Machines (Audio), what are the key points of this episode?

Searching your listening

Transcript

Automatic transcript. May contain errors.

0:00Hello and welcome to Intelligent Machines. This is Jason Heiner filling in for Leo Laporte, coming to you with Jeff Jarvis and Paris Martino. Today we talk about Anthropics IPO, Google's new high-end Gemini Carbon model, and Meta's Data Snafu. Coming up next on Intelligent Machines. Podcasts you love. From people you trust. This is Twit. you're watching intelligent machines episode number 894 recorded on wednesday october 28th 2026 ghosting metamuse hello and welcome to intelligent machines uh another great episode here i'm jason heiner filling in for leo laporte uh very excited to be here there's so much to talk about, as there always is every week on this podcast.

0:59And of course, I've got wonderful co-hosts to help unpack all of the stories of the week. Paris Martineau, welcome. Good to see you. Hello. Lovely to have you taking the reins today. Thank you for letting me sit in the big chair. Yeah, I can't see the height of your chair, but I assume that it's immense and impressive. Big enough. And that's all that matters. Okay. Right on. And, of course, we have Jeff Jarvis from the Craig Newmark School. Thank you, Jeff. Craig Newmark. And of many other hats as well. You're an author and a professor and a commentator, podcaster. So many hats. And, if we may do a quick plug at the beginning, a former critic, and I got my critic juices going again with Paris.

1:57Indeed he did. And Micah Sargent on our all spoilers out, everything you can see, everything we can say episode, extra episode, so you can all watch this. It's a club episode, but it's open to all. Yeah, what we did is we did a podcast on You Can See Everything, the Nathan Fielder, Lance Oppenheim documentary about Elizabeth Holmes. and me, Jeff and Micah recorded, like he just said, a spoiler filled episode. We've really tried to like break it down from like a tech and media lens. And you can listen to all of it on the Twit YouTube channel. Very cool. All right, so don't miss that. That's a special treat.

2:42You can also watch us do our Elizabeth Hall invitations. I was gonna say, we all did also wear black turtlenecks and stare under the camera. house. And it's very funny to see people in the comments being like, how long did it take you to realize they're all wearing black turtlenecks? And I'm like, I'm shocked that it took you any time at all. It's pretty evident. So thanks to both Anthony Nielsen and Micah Sargent for the extra duty on Sunday. I'm grateful for that. Real shout out to Anthony, who Jeff and I just realized may or may not have seen You Can See Everything. And if the answer was may not, just got the entirety of it spoiled for him because we break it down act by act.

3:21If you're still in the chat, Anthony, did we ruin it for you? Hey, Anthony, I hope if we ruined it for you, you at least had fun. We did go on so long he had to kick us out. Yeah, there definitely was a time where he was like, you have to leave now, which I guess is the sign of a good podcast. Yeah, for sure. For sure. I hope that you all are, you know, kicking me out by the end of this one and saying, Oh, no. Time to get started. Time to get rolling. So, all back to AIs. Thank you for that. Back to our regularly scheduled programming, which, of course, is all the stories in AI. And there was a little story this week that you all may have heard about that Anthropic is filed to go public.

4:06So, this is long anticipated, of course. And the story for most of 2026 has been the anticipation of the Anthropic and OpenAI IPOs. One of the big stories is Anthropic made it or is nearly there to ready to go public. And OpenAI has kicked the can down the road a little further till 2027. And the big story about that is also really the main story of the Anthropic IPO, which is that Anthropic is just a lot further ahead with enterprise customers, where there's actually real money to be made. And OpenAI, up until 2026, had been playing a lot of different hands and had been doing many things and really been going down the road of being a consumer brand.

5:02And while they haven't ruled that out, they have pivoted pretty hard with Codex and with all of the things that they've been doing toward kind of chasing the enterprise money because that's where there is reliable revenue. And if you are a public company, you are under scrutiny at all times for the levels of revenue and revenue growth that you're producing. So, OpenAI is just a little bit behind. They're clearly making some progress and there's a lot of professionals and organizations that really like Codex. And there's also been reports out this week that OpenAI is being very aggressive in pricing and is winning over some enterprises in that way.

5:47Well, I know who they are hoping to maybe win over probably the most would be that One thing we've learned from Anthropic kind of filing, or at least it's like proposed S1 this week, is that 25 % of the company's$16.3 billion in revenue comes from five customers. Just five customers make$16 billion in revenue. And Anthropic has said that there seemed to be some reporting out there from the journal and other people who are talking to sources that say like, oh, Anthropic is not worried about losing those customers, which is kind of crazy. I guess if you're spending billions of dollars with one AI provider, you probably have a custom enterprise suite built out to where it would be burdensome and difficult for you to swap them out for OpenAI.

6:38But man, if you're OpenAI, you've got to be jonesing for those deals. Yeah. And yeah, I think they're worried. If they say we're not worried, that means they're really worried. It's just a lot of money. And we can imagine who they are, like some of the biggest, you know, customers in the world. Those are like governments, banks, organizations, you know, largest companies in the world. So, you know, OpenAI is essentially just trying to win at any cost on price is what the reporting has been on that. But Anthropic, as we know, in the enterprise, when you put all of your processes against a certain thing, it takes a long time to unwind it, right?

7:26So Anthropic has been betting on the enterprise since it really went live in 2023. All of its focus has been around organizations, around societal change, around things like science. All of that's enterprise. And so they've been building around the enterprise, and they've been becoming part of some of the largest enterprises in the world for a long time, which means those companies have built processes, to your point, Paris, around that. They probably have very custom installations. That doesn't get unwound easily. We even saw that with the US government when they declared Anthropic a supply chain risk and an unprecedented move in the spring.

8:14Well, it took them still like six months to stop using Anthropic. uh hard drug to quit clock yes exactly and so uh and there were so many interesting things uh yeah in this s1 i mean it was like a nearly 400 page document like 94 pages of which were just risk factors which i mean i'll get through in a second but there were just an extinction but i mean including extinction including stuff like yeah extinction of the human race um little side comments like, oh, the potential for damages to be caused by our tools is so great that we think it dwarfs the capacity of all insurers combined to insure against.

8:57Oopsie. But there are also kind of - Footnote. Details such as, I'm trying to remember where I saw it, maybe this was something reported by the journal that basically anthropic employees have among other things been pitched like working on these sort of scientific enterprise models to develop kind of scary bio weapon-esque technology specifically with the purpose of like having scary pitches to give the us not the us or governments of some countries in order to compel them to sign contracts like this. And I didn't even realize when I was saying this earlier, but that$16.3 billion in revenue, it's in a year's worth of revenue.

9:42That's the first six months of this year. So they're clearly just cleaning up with these sort of tactics. The other number that got out there a lot was the$80 billion loss. But that's frustrating because it's not really operating. As it's been explained to me, that was trying to regularize investments like Amazon's when Amazon early days gave them a ton of compute and got reported up to 10 % of the company, and they had to book that differently. So I think it's frustrating from what I've seen document because we don't really have a good sense of the ongoing year-end operating structure of it and what profit and loss looks like.

10:21That's right, Jeff. So there have been, it's a good highlight because there have been a lot of confusing numbers about Anthropics revenue throughout 2026. There was a report that they were actually profitable based on operations, but what they were unprofitable with was all the research going into building the next model, right? That they are very much a cash machine at this point. They are generating tons and tons of cash. But they are investing to build out the future, to build out future models, to build out safety, to build out other parts of the business. And, of course, more than anything, to buy compute and to secure future compute.

11:15And so, to your point, it gets really tough to discern the numbers from here. And when they actually go public, that's going to have to become much clearer because analysts are going to really grill them on that. Because, you know, these analysts and sort of funds, they want to have a very clear picture of the business and the directionality of the business, right, in order to invest for their large, you know, large chunks of money into it. I mean, so you've looked through these documents as well. what immediately stood out to you as any red flags and green flags? I think if you'd boil it down to, are there any numbers that looked alarming or any numbers that looked quite impressive or good or unexpectedly so for Anthropic?

12:13Yeah, thanks, Paris. You know, the biggest one, and we talked a little bit about it in the pre-show to me, was the compute numbers. You know, it's funny because Anthropic was playing it a little bit coy on compute. They were not signing like big giant deals. They were that you heard. Now in some cases they were quietly, but if you remember last year, I'm sure you both remember very well at this time, OpenAI was really taking it on the chin for this like $1 trillion in compute that they had secured when they had single billions in revenue at the time. And so that made sense to no one, right? Like from just a pure optic standpoint.

13:00And Anthropic did, I think, got spooked a little bit by that because that brought a lot of negative sentiment to OpenAI. And so they maybe weren't as aggressive in signing deals. And then if you remember well, when Mythos was announced in the spring, they were like, yeah, but we're not really going to release it to anybody yet. And the narrative at the time was like, it's too dangerous to release in the wild. But my understanding from sources in the industry was like, they just didn't have the compute to run it, even if they wanted to. They just didn't have the compute. And if you remember, after they announced MIST, they suddenly started signing huge deals with Amazon, with even SpaceX, and lots of others.

13:51They secured more funding, and they secured more compute. So what worries me the most about the anthropic filing, and I would love to get your all's take on this as well, is that it's based on what could be a bit of a house of cards, which is all of this compute, this giant infrastructure build out, which maybe we'll need all of that. But today, it's all built on this idea that the world, we have to project out, you know, whatever, two, three, four, five years based on where the models and the compute is today. Well, a lot of that compute, we don't even have the power, right, the energy to run it.

14:38And so that's a concern. But it's also maybe the bigger concern is that it's based on these very inefficient models today, right? Like LLMs and the Frontier LLMs are built on transformers, of course, and transformers are just notoriously inefficient. And so you could have, and I would expect we will have two things happen that are going to make the compute story look very different. One is that the models are going to get a lot more efficient. We're already seeing this, like, instead of using the latest Frontier model, you can use small language models, which are faster, smaller, cheaper, more accurate.

15:14You can use domain-specific models. You can use local models. The models themselves can get more efficient. And there's also alternatives to transformers from companies like Pathway and Inception and some of these other Neolabs. And things like JEV and then OpenAI just released its sort of JEV competitor, as it were. But if the models get drastically more efficient, and then we also stop using the cloud and the data centers to run these things, and we are using more of our local compute that we have on our machines, Macs and Windows machines and phones that have NPUs in them, then all of this kind of cloud infrastructure thing looks a little different.

16:01No, it looks a lot different. And so - Yeah, especially when the numbers that we're seeing in there for what Anthropic has agreed do stuff like 500 to 750 billion dollars in like locked in compute spending over the next seven to 10 years and so that means no matter what happens even if all everything goes right and we end up needing way less than that anthropic has to pay hundreds of billions of dollars over the next seven to ten years even if it does not use any of it which is those are the sort of sums that would sink any normal company. But time and time again, Anthropica-like, OpenAI-like, everything in this segment of the AI race is just operating kind of in a world of its own with rules of its own that seem to not follow any of the normal patterns of logic and business.

16:55Yeah, I think that you started the list of risks, Jason. I think, right, so a lot of competition from other frontier companies who play leapfrog constantly. And you're never ahead for long. Google released Carbon today, and we'll get to that, I presume, later. It's going to compete on Mythos and so on and so forth. Then competition from open weight models from China and elsewhere that can do the great majority of the work most people will want to do with this. then I think questions about whether or not hyperscaling is the model paradigm. I think there's going to be other ways we look at this.

17:35You're right. There's greater efficiency with Jensen Wong talks about all the time. But I also think that there could be different, entirely different ways to look at AI that change this model. Then we still don't know. I don't think the economic value of this in the marketplace, I mean, yes, Microsoft said they're laying off or, pardon me, offering buyouts to 60 ,000 employees and blaming AI, but it ain't just AI. What value is AI really creating in the market? Then there's China, which is going to undercut American AI at every turn, I think. And so all those add up to me to, nevermind this existential hoo-ha crap.

18:15The real risk to them is a competitive marketplace risk where their business isn't really proven yet. Five customers does not. They get laughed off of Shark Tank at that. That's a joke. It's always great when you have to tell us that it's a joke, Jeff. I know. Mark Cuban would still invest.

18:38I don't see one and a half to two trillion dollars value here. And so this is the, I said to the other Jason today, that I think that this is the most bubbly I've felt in the AI era, because this thing just isn't justified at that money. I don't see it. Am I wrong? I mean, I just think it's notable that like we're talking about revenue in the tens of billions of dollars and then we're talking about one line of expenses compute that is in the hundreds of billion dollars. Albeit, I guess that's over seven to 10 years, but still could be something like 10x your projected revenue or 5x or something like that.

19:21Revenue that you need to justify that expense. Revenue that you need to justify that expense. And that is just one expense. That's not any of the other things that go into operating this, training these models, employing your staff, running the business. I mean, it just, of course, it was going to be a moment like this when we are actually seeing filed or prepared to be filed documents for these companies, when it all starts to seem like it might come crumbling down. because part of the magic of AI and part of the magic powering this bubble has been that you just don't see what's behind the curtain.

19:59And now we're starting to see it. And yeah, it does not look like the sort of normal business fundamentals that you'd think would work. But I guess there's a possibility, like everything with the AI boom, it works anyway, but it seems less and less likely. If Dario... Go ahead. So I was just going to say, if Dario Amadai was here, I think what he would tell us is... Please channel. He would say, what we're not accounting for is the exponential, right? Because they have grown 10x three years in a row, which I really... Oh, the old hockey stick. Yeah. Well, there's no way that could ever slow down.

20:40I know, right? And this year is the year I thought it would slow down, but they've been growing more like 14x. But there's been a bit has been a weird year. Right. Like there was they had Anthropic had all of this positive mojo created by them standing up to the U.S. government on just these sort of like two red lines, as I'm sure you've covered in many, many times on the on the show. I think we covered the last time I was here. Actually, fill it in for Leo. It's often happening. Yes. It is always there. So I think what Anthropic would say is that this is growing at such an exponential rate. And right now, only about 18 % of the people on the planet use AI regularly today.

21:27Over these coming years, that's going to expand greatly. and really a large number of businesses have still barely tapped into it. I did a podcast this past week with a consultant who consults with mid-market companies, primarily about like 10 million to$100 million companies, right? And there are lots of them, right? There are just tons and tons of those out there. And most of them have barely scratched the surface on starting to use AI and they're being very practical in the ways that they do it. They're barely spending anything today. But now they're starting to find one problem, one workflow where they're like, okay, we can use AI to fix this problem.

22:07Great. Their journey toward AI, toward investing in AI over these next few years is likely going to be pretty exponential, right? They're going to invest$5 million in AI maybe in 2027, but they may be doing probably 50 or more in five years if things go well. So I think if you take and multiply that by a lot of companies. I'm curious, though, for companies like that, though, is your sense that they envision their journey with AI to be, I'm always going to be spending those millions of dollars with one enterprise AI provider? Or do you think that if this tech, my, you know, just immediate thought is if the technology is even half as transformative as its biggest proponents say it is, and that people are going to spend even half as much money as those people say they will be, companies will be like, yeah, instead of paying money to Anthropic based on whatever their costs are at that point in our contract, why wouldn't I try and do some of this locally or figure out other ways of building?

23:14that are not just beholden to like two, if not one, enterprise players. Which is also the Alex Karp argument to all the big enterprises. You're a fool if you let Anthropic do this. Use my harness and do it with other models that are open locally. Jason, let me ask you a question. So to what Paris was saying before about the huge irrevocable investment. And so if things don't go as well as planned, Anthropic is stuck with a bill to pay for infrastructure. Let me ask a different question. Presuming that Vera Rubin gets replaced by five different generations of chips over this whole time too, who's stuck with the out-of-date infrastructure?

24:06The depreciation on this stuff is what? A year and a half? Yeah. So this is a great question. This is why I love being on this podcast. You all anticipate all of my questions and thoughts. So this is another place where efficiency comes in big time, right, Jeff? Which is that NVIDIA's chips are getting enormously more efficient with every generation as well. Which is his business model, which is Justin Wong's business model, just to make it that way. Yeah. So you are incented to buy the new one because you're like, great, this thing is more efficient. I can run my really powerful thing on that and it costs me a lot less.

24:48Right? Right. So I've talked to some folks about this in the industry, especially the people who are focused on infrastructure. And this is also where efficiency comes in, which is like there are going to be the Blackwell, Rubin, some of the past versions of the infrastructure. What they say is they will still be around, that hardware will still be around. They will have absorbed the cost that they needed. It will be profitable at that point. And it will become infinitely cheaper to run on that infrastructure then, right, in the future. So you'll run, remember, we have these, like, smaller models and we have the sort of less expensive models, even from Anthropic and OpenAI, you know, the haikus and the lunas of the world, that those will, you know, you can run those on these, like, older, this older hardware for a lot cheaper, right?

25:44I think, I suspect they haven't said this. I think for any company other than the Frontier models out there, that's true. There's going to be real value in using last year and getting a used car. Yes, that's right. But for the Frontier companies and the hyperscaling, macho, chest-beating stuff, they can't say, oh, we're good enough on Blackwell. It's fine. That's right. They will want and need the latest model. So here's what I think happens, perhaps. But this is where this efficiency curve starts to get steeper too, right? So the stuff does get cheaper. And all of this, so they're securing that.

26:23It's one of the big questions. They're securing all of this compute. Clearly, AI has a growth path ahead, despite all the fact that most people are really negative on AI, right? Like 70 % of the U.S. public has a more negative than positive view of AI. But at the same time, they are using it. and unlikely more and more of them will. But what is likely to happen is that it's gonna get less, it's gonna be more, you're gonna get more intelligence per watt and per dollar, right, as time progresses. So, but who holds the bag? Is like, who's gonna end up holding the bag, right? Like, is your question.

27:05If this thing starts to falter, that's one of the big questions, right? Is it gonna be some of these hyperscalers? Probably not because they have plenty of other workloads they can use. Is it going to be some of these, the ones that are, I suspect it'll be the ones that are taking the most risk, right? They're very, some of these like neo scalers, some of the, which are the - Define a neo scaler for me. So a neo scaler is these companies like Lambda and CoreWeave and some of these folks, right? They're very leveraged. Right now, they have more demand than they can even deal with for the next two years.

27:50But if it starts to falter and companies like Anthropic are like, you know what, we're not going to need that compute, then I suspect that the ones, these neoscalers who are kind of in the middle of the hyperscalers, they're like too big to fail. Google, Amazon, Microsoft. And then these AI companies. They've got other customers. They've got customers coming out. That's right. They've got other customers. But these neo scalers are all in and they are all AI primarily like all the time. Right. So are they the ones that hold the bag? We don't know. But they are they are losing a little bit of public sentiment in the markets because of that fear that you raised.

28:30Right. Like who ends up holding the bag. So so if Anthropoc went belly up, they'd be the first dominoes to fall, you think, because because there's just less demand in the marketplace. Right. Because but but I think the the the counter argument is that, you know, or the argument is with them is they are like these the dark fiber companies during the during the. Right. That's that's what there's other uses we can put it to at a much lower rate. Let me ask you both a different question. If you were had a fiduciary responsibility over the California pension system. I would shoot myself with a gun.

29:09yes and others might shoot you because you're you're probably going to lose their i just i don't think i should be in charge of anyone else's money much less a whole state's worth well you're there paris if i'm there if i'm held by bayonet yes you are in charge of a million dollars would you invest in anthropic so here here's why there's a for those of you on audio there's a panicked look in Paris. There's more panic in my body than I thought was possible from this hypothetical. Yeah. Here's what this gets to the ultimate question, I think, around this IPO and around the whole future of the AI trade and the AI industry.

29:55Models are commoditizing, as Jeff, you articulated really, really well. Chinese models, there are more US open source models that are going to be launching over the next year. Like you said, the models leapfrog. We'll talk about Google Carbon in a second, too. The models leapfrog each other all the time. What's the value of having the latest and the best and the most powerful frontier model, I think is likely going to start to evaporate even in the next 12 months. What Anthropic is betting on, I think, and the S1 says this a little bit, but even more so, I think their actions over the past year, is Anthropic is betting on building a brand that says, we are the safe AI, we are the good guys, we're the ones who are going to do the right thing.

30:49We define safety, though, very weirdly about whether we can make humanity safe. We are the safe AI. We kill only up to 49 people. We only cause up to a billion dollars in damages. We have a license to kill 49. I think that what... We can debate the merits of it, whether it's true, but what they are setting themselves up as is we are the brand, the AI brand you can trust. And so long term, whether you're an enterprise or you are a government, we are the brand that is level-headed, that doesn't just go whichever way the way it blows. I gotta keep laughing. That's fine, keep laughing. They're cultists, they're crazy, they're Looney Tunes.

31:38But this is what their bet is. Their bet isn't, I think, just that we're gonna make the most powerful model. Their bet is we're gonna build the safe, trustworthy AI brand. And that's where they're saying the future of this is and why they're going to be worth what they're worth. Okay, so let's dig in there for a second. But pick that apart, please. So safety, right, I argue this often on the show. Safety is a corrupted word because the Tesquiel cultists talk about safety in terms of human extinction and all that BS. Whereas if you're a company using this product, if you are an insurer who dares to insure it, if you are a pension fund investing in it, then um your definition of safety is very different i think it's is this thing gonna get in all kinds of trouble uh by allowing people to do bad things uh is uh their market going to go down because people don't trust it uh because all they talk about is crazy crap when it comes to safety uh is my investment at risk because and they do say this in the in the s1 because if their software does bad things, then people will blame it and they'll get a bad reputation.

32:57Gee, it's bad for your brand, killing 49 people. Or I guess killing 51 people is what is bad for your brand. So I think that their effort to paint themselves with a safety halo is complete BS because they've corrupted the notion of what safety is to a point that you can't get anywhere with it. And this goes back to the Stochastic Parents paper. where while everybody was concentrating on getting bigger and bigger and bigger, they're forgetting about the environment and labor and impact on both and so on and so forth. I don't disagree, Jeff. And yet I also think that the fact that they are so well monetized in the enterprise and with sort of the most conservative organizations, conservative in the sense of fiscally conservative organizations.

33:55Well, these days it's pretty much synonymous, yes, but go ahead. It does tell the story that their approach, their strategy to paint themselves as the safe and trustworthy AI has largely been successful, that they have successfully done that. And I think long term, whether they have the fastest, best, most performant model will probably depend on the week or the month. But I think that they are embedded and embedding themselves with companies and the companies that are investing in it, investing in AI. They are doing well. Now, I think that their biggest competitor and risk in that is actually not so much open AI, at least what I'm seeing.

34:51I think it's more Microsoft and Google and then even sort of these sort of next tier companies like Cohere, you know, out of Canada. They're an international company, but based out of Canada. These companies that have long record of trust with enterprises and with the world's largest organizations. But interestingly enough, and I want to get your take on this too, Paris, which is that what I didn't expect in 2026 is that Anthropoc is now beyond just sort of the corporate and organizational sort of trust that they've built. They've also started to build it with consumers. The number of people that have downloaded, the people that I know that are most skeptical of AI, most of them are using Claude, and almost purely because of everything that happened in March with the Department of War.

35:47And so while I think all the points you raise are really sharp and really important, Jeff, I do think they are seeing some success in their messaging. Paris? I mean, yeah, anecdotally, I feel like my experience is similar. I feel like people are, Claude and Anthropic are still coasting off of the wave of great PR, the company generated earlier this year. And I don't think that any other major AI company or minor AI company, for that matter, has made any real concerted effort to kind of develop a similar audience or develop an audience at least around like moral standing or like things like ethics.

36:41The closest I could see is now here in New York, there's some ad campaign going on in the subway that is trying to pitch itself as like an eco-friendly AI. But the ads are like so bad that I've never even looked up to figure out how it's lying about that. But I'm sure it's somehow doing that. As we wrap Anthropic, last thing on this I want to ask you two about is the existential question of it all. Because I'll be honest, and I've written about this on The Deep View, and I've talked about it on other podcasts and stuff too. I just think it's so overblown. It's not that it's a 0 % chance, the AI extinction thing, but I think it's probably less than 1%.

37:29And you actually, in the ways that it gets played, even by Anthropic and some of its public-facing communications about this, is you would think it was like 20 % or 50 % or something. And I think it's so overblown. And I do think there is a degree of it that is marketing. It's like our technology is so powerful that it could create an existential crisis. You're speaking to the choir here with Jeff. He's always talking with that. I mean, I'll give one brief example just because, so last weekend I went to go visit my old college newsroom for an alumni event. There were some people there from, you know, my year's college newspaper and then a bunch of young people.

38:14And once they'd heard I was doing this podcast, someone asked me, they're like, oh, so like, how worried should we all be about AI committing a mass distinction on us? Is that going to happen in like a year or two years? Should I have a bank account or not? And I was like, Like, honestly, I mean, my real take is the danger you need to be worried about is like very, like, it's different and more practical. I think when people hear like things like extinction event or Dario, you know, saying warding of doom coming imminently, they think like Terminator is going to happen or that suddenly it's going to be like the Matrix.

38:50It's like that's not a realistic thing that is going to happen. Like a lot of things that have to change in our life and relationship with technology for any of that to enter the realm of reality in the near term. But the thing that is more realistic that I think people should be concerned about is stuff that we have covered on the show before. Things that like the fact that this technology will enable people to commit like large cyber attacks much faster without having many of the skills that it used to take to do those. And that could have profound effects on like your life and work. You know, I was like, if there is some quote unquote big event that could cause harm on a societal level, it'll be stuff like power outages and like blackouts happening more often, utility, public utilities going down, communication blackouts, more frequent cyber hacks that have like an effect on your health or wealth or ability to work and things like that.

39:52Like that is, I think, the most realistic doomsday concern, but it's not as pithy and fun. I guess fun isn't the right word. It's not as pithy and I guess. Since you're a nihilist, I think you're allowed to say fun. It's not as pithy and fun as doomsday extinction event happening soon. So, yeah, I go crazy on this topic because I think that it is cult BS, faux philosophies, Tess Greal, the folks who watch the show have heard me mention that enough that it's a drinking game now. And I think that it might have worked to get cult members in and staff in, and it might have worked to get VC money in.

40:35But one of the things that impressed me from watching You Can See Everything is, I'm sorry, from the miniseries, The Dropout, is how it portrays the investors investors as so damn willfully gullible and credulous. And so I think this discussion of, my God, it's so powerful. I can do that. I have to be there. I must invest. I must have the power to do that. I'm ahead of everyone else. It's complete BS. Now, when it gets to the open market, both for customers and for investors, I don't think that has any value at all. Quite the contrary. I think it starts to become crazy. So Elizabeth Warren sent Amade and company a letter packed full with four pages of questions.

41:18Like, how are you going to not destroy humanity? Tell, oh, please tell me. You're so safe. Tell me what's what. And I don't think it's going to play anymore. And I think that there'll be a reckoning on all this. And there's somebody I used to work with years ago at People Magazine who's been on Facebook just said, oh, I just picked up the Hjodkowski and Soros book because the AI companies are reading this. so I must read it. I said, they're nuts. They're crazy. This is the one that if we build it, everyone will die. And then I came across the story this week in Fortune that Nate Soros said that he hasn't added to his 401k in a decade because, well, hell, A is going to kill humanity.

42:00That's the level of nut jobbery that we're talking about here, the level of cultishness that we're talking about here. And at some point, these people have so much power and so much capital and so much influence over everything. look at what musk is doing with his money in politics at some point there needs to be a reckoning but the problem is i hate media's coverage of this because they treat a gathering at the future of life institute as if oh it's a not-for-profit held a nice little event they're doomers they're nuts balls uh and and we're not covering that and so what happens is it influences policy discussion and that's where the real danger comes in i think is the the down the line domino effect that it has on policy.

42:42All right. With that, let's throw to our first word from a sponsor and we'll hand it over to Uncle Leo to give. Hey guys, this episode of Intelligent Machines is brought to you by Framer. Framer is a complete website platform. It's not just a builder. So teams can launch and keep improving their sites in one place. Thousands of businesses from early stage startups to fortune 500s are choosing to build their websites in framer where changes take minutes instead of days framer is the ai website builder that helps creators teams and businesses ship production ready sites faster than ever while getting every detail right prompt inspect edit and publish in one place at a whole new pace agents and humans work in tandem Agents bring speed and scale.

43:34People bring taste, judgment, and control. The agents solve the gap between AI-generated ideas and production-ready website work. The agent works in the same place where the real site's designed, managed, reviewed, and published. It lands on the canvas, it stays editable, and it can be published when the team's ready. Agents and framework work alongside teams to streamline collaboration on the same canvas, build custom code components, create and manage CMS content, optimize SEO settings, and ship everything all in one place. Framer is an enterprise-level solution powered by agents that provides premium hosting, enterprise-grade security, and 99.99 % uptime SLAs, giving the world's leading brands like Perplexity and Miro the confidence they need to build their websites in Framer.

44:25Learn how you can get more out of your site from a Framer specialist or get started building for free today at framer.com slash machines for a 30 % off a Framer Pro annual plan. That's framer.com slash machines for 30 % off. Rules and restrictions may apply. All right. So for our next story, let's dig in a little bit to Google getting back in the game a bit with Gemini Carbon. It's a good thing. This space needs competition and really it needs it from Google as well. As we've mentioned on the Deep View when we've covered Google throughout this year, it's been a tough year for Google. They have not been able to release a pro level, what is their pro level or sort of a really cutting edge frontier level model pretty much all year.

45:24They had a great moment last fall with Gemini 3. Was that the last moment of? They've been too busy injecting AI into every aspect of Google Workplace in ways that are designed to drive me personally insane. That's why they haven't released the model, I think. they've been too busy putting little purple squiggly lines under every line of my emails in gmail that i click being like oh did i misspell something and it's like no we just think you should have the words be slightly different because we feel that way that's right okay sounds like they succeeded uh and they pissed me off because my workspace account i can't get to some of these functions but go ahead but listen go ahead we'd all love for google to do better we would and and And you could argue the ecosystem needs it, right?

46:12Like if more people in the world are going to have an on-ramp to AI, it's probably going to be Google and Apple taking them there to a degree over the next couple of years. But Google has struggled to keep up with Anthropic and OpenAI in speed. Anthropic and OpenAI have just been running circles around them with the advances in their models. And so there's the big shakeup, of course. Demis Hassabis is no longer the head of Google DeepMind. It does feel like they've moved more of the center of the group to Mountain View, away from London, although they still have strong operations in both places.

47:01And now they have released a model that does compete with the latest models, the Opus model and Astra models. So good for Google. Their carbon model on the benchmarks, it's really focused on cybersecurity and long horizon agents, which these are the most complex use cases. It does very well. It is right above, it's right there with Astra and Opus on many of the things.

47:39It doesn't meet the, chin the bar on all of them. There are a few things where those models are still closed, but it's in the ballpark, right? So, the question is, can Google keep pace? And they have a lot of advantages that the others don't, right? They are a hyperscaler themselves. They have almost unlimited amounts of compute because of that. So in a way that they don't have to pay for compute in the same way that Anthropic and OpenAI do because they have all of their own. That was, if you remember, I'm sure you both have seen the movie that came out on Demis Sassabas and Google DeepMind. And essentially Demis says that he says, look, the one reason we need to do this deal with Google and let them buy DeepMind is because compute.

48:28We are not going to be able to have the compute we need to train the models. And so they are and they have this brand that a lot of people trust, you know, Google itself. And they are well known to consumers and they have all of these users, right, that come to Google that like Paris, that they are now ticking off because they're shoving Google Gemini into every corner of the ecosystem. They also just have an incredible opportunity to convert their already existing enterprise and I guess non-enterprise customer base into subscribers in a way that Opening Eye and Anthropic can only dream of. They don't already have those relationships.

49:09They don't already have the bank accounts, right? And the trust. Or the data. That's right. They don't already have all of those companies' data in safe and trusted environments. This is the same with Microsoft. Microsoft has many of these same advantages, but not the sort of consumer brand advantage that Google does. So will you all, in your daily usage and with the people that you know, have they turned away from Gemini already? I think the AI enthusiast crowd has. But for the larger number of people, we still know, again, 18 % of the world is using AI today. More and more of them are still, there's a lot of people out there that are still open to, you know, figuring out what they want to regularly use.

50:00Is it unreasonable to think that Google still has plenty of opportunity to do it for all the reasons we've already mentioned? I mean, I think that Google will always have plenty of opportunity just because the amount of customers, active customers they have and potential customers they have based on people who use Google.com, like the search engine, and Google Chrome and Android is like infinite in comparison to the customer bases of the two leading frontier model companies. However, I also think that they are just at such a disadvantage because of their kind of late start in this, which is strange to say because they of course like, the team behind it was not late, but as a popular consumer product, they do feel a little bit like an also-ran.

50:47Like, I don't know, even people who I know anecdotally who were really into Gemini because it was something offered through their Google Workspace account. Like, I'm thinking of one friend in particular who's probably my most like, I was going to say normie AI psychosis friend, but if I mean like a non-negative, like a way of like a normie AI user who's become so overly addicted to it, they use it for everything, which I guess is probably an ideal customer for these sort of people. He was a big Google Gemini user until like last year when I was talking about Claude with the podcast. He's like, oh, I've never heard of that.

51:21I guess I'll try it. And he's like, wow, this is so much better. And now all he does is Claude everything at all times for a year. And I do think that, I mean, that's what Google and Gemini have to contend with. I think obviously for like enterprise cases and, you know, targeted custom build outs, that's a different story. But if we were talking about like the average consumer, it's going to be a bit of an uphill battle for them to win that person back. I still use Gemini because I live la vida Google and I'm on a Chromebook. And so Gemini, when it finally got added to Workspace, Growl, it's just there and easy.

52:00and so I use it for a lot of things as a result. I think it's classic case though of Google screwing up branding and product rollout is Google, to say Google's behind is behind in kind of a fake fight for high-end benchmarks that most people never tickle. Most people don't get to that point of using that. So at the very high end of this, fine. There's a competition on Google was delayed. And I think that hurt them. But on the day-to-day level, as you just said, Jason, Google, you know, for a consumer level, it has my email, it has my calendar, it has my Google wallet, it has everything. And it's in the best position.

52:47But, you know, I was just looking up, I forgot. Whatever happened to Spark? I was waiting to get Spark, finally. No relation to the machines Leo runs his local models on. Not Nvidia's new Spark. Not NVIDIA Sparks. Google. This is also a Google problem, is the naming of everything. But continue to. Yeah, naming is everything. And then I look it up, and I guess I didn't realize this, that some time ago, Google updated its, according to Google, updated its personal AI features by eliminating Spark and merging its agentic tasks into the main Gemini. This is my problem with everything Google. And I mean, this has been a problem we've been covering since the show is called This Week in Google.

53:26Google is terrible at naming things. It is terrible at developing individual products. If it does somehow manage to name and develop an individual product that is something you like, you can bet your butt that in six to 12 months, that product will have a different name, be in a different, folded into a different thing, or it will be dead. Like Notebook LM, which I genuinely do enjoy, and it's something I've recommended to a lot of my colleagues, is now Gemini Notebook, and it lives in the sidebar of Gemini, and I had to give a training to colleagues on Notebook LM this summer. And of course, it was the week after they renamed it.

54:05And I was like, I'm screwed. I'm going to say Notebook LM 100 times during this training, and you're going to have to deal with that. Oh, man. Yeah. And so, you know, look at even Meta, even Screwball Meta, Muse has been a success because they created a brand, they created a use case. It'll probably fade away like things do in meta, but for a purpose of three or four weeks, they have a little tiny winner, right? And Google doesn't have the same functionality. It doesn't have the same way to do it. And it's frustrating to sell. ScooterX in our Discord chat just posted a link, the headline of which is, Gemini Spark disappeared because Google renamed it task mode.

54:48And I think that summarizes everything we're talking about. Is this a company where that just happens all the time? Oh, boy. Yeah. So, you know, there was a story this week that I didn't finish reading about Marissa Meyer. And people are making fun of it for various reasons. She asked her agent. Oh, my God. This is one of the saddest details. Maybe we should actually hold off on that. We'll go to that separately. But in a way, what Google needs, back in the day, Marissa Meyer represented the consumer user interest. Yeah. And Google doesn't have that high up in the echelons in terms of, whoa, everybody, people are going to be confused.

55:29They A-B tested everything to a fault to make sure that it made sense to people, that it was part of the use case. They did that well for a long time, way back when, and they lost that consumer focus long since. And they need somebody to be that advocate within the company, I think. My last thought in Google, because I know we're going on too much for a ramp, but Jeff and I have a lot to say on Google, I love it. I think more so, like, Google is tied, right, guess me? Google is tied with Meta and Facebook and Instagram for being a company where now every feature I interact with across any of the Google apps and services I use, and it's a lot, they all just reek of being designed to maximize engagement of some feature that a product manager somewhere is getting graded on how many people click or interact with.

56:21Like anytime I open up any of my apps now, if they're from Meta or Google, something has changed and it's changed in a way that makes me interact with some aspect of Gemini that I know in my heart of hearts is just because that's how someone else keeps their job is by that number going up. And that's an infuriating customer experience. And I also think that that is the sort of like big business brain mindset that ultimately ends up making a consumer product have slower adoption than it should have. Yeah, I'm so glad that you all highlighted this and clearly articulated it all very well and quite clearly.

57:05This was so evident when I was sitting in the Google I.O. keynote. there's this phrase, you know, in tech journal circles of like they rolled out the org chart. You know, you could see they rolled out like, there were like 19 brand names or something. I can't remember that they rolled out in that thing. And it was clear that like what we learned was, oh, these are each different teams that worked on this thing and they're basically sharing what they learned, which makes sense from their organizational standpoint, from their org chart. but it makes zero sense as a user, right? It became so confusing.

57:45You couldn't understand like, wait, how does that thing relate to this thing? And why is Spark now the agent that I'm gonna talk to? And how does that relate to Gemini versus AI overviews? And it was so confusing. And I just thought at that moment, I was like, they're cooked, right? Like from a consumer standpoint, you're just not gonna be able to figure that out. Jeff, I hadn't even realized that Spark wasn't a thing anymore. You hadn't even realized that Spark had gone task mode. I've been waiting for this. When is Spark going to come into my life? Spark goes task mode, but what about Dream Beans?

58:20Yeah. These are the inane sentences that we have to have in our head now. And the fact that you can have that moment like that at Google I.O., which is supposed to be some culmination, their big moment of announcing their strategy and all their new products to the press and people, that shows that there is not like an adult in the room or an executive with the foresight to look at that presentation and be like, yeah, we got to simplify these things for the average consumer, for the consumer, for enterprise customers, for any of the people who are supposed to be purchasing and using our product.

58:56They need to have someone who is like, this is ridiculous. And because this has been a problem at Google forever, but they clearly just don't want to fix it. They need a VP level in charge of the relationship with consumer, a VP level. I'm sure they have in charge of the relationship with enterprise, but that pisses me off all the time. And they need branding. They need to bring in branding and expertise because they just do not have it. But underneath this, back to what you were saying before, Jason, Google had gotten this reputation to be behind. But as you said before that, going back to the fact that they invented all of this with Transformer and the model of what became LLMs, Google's been way ahead.

59:36They've had it integrated in their products in a million ways, including many that irritate Paris. They've been out there, but they don't know how to present that either. It's also a PR problem for them. Yeah. As much as we talk about how far behind they are on models, I don't really think they have a technology problem. I think they have a little bit of an organizational problem. They have a vibes problem. A branding problem. The vibes are off. Big time. Yes. Because even, you know, it's funny. I mentioned earlier these next generation models that are not based on the transformer, right? They're based on these other technologies.

1:00:13One of them is diffusion, right? Diffusion is what everybody will know is used for video and images. It's a very powerful model because what it does is it is not sequential. LLMs are sequential. Diffusion does a lot of things at once, and that's how it can create these things so quickly. Well, one of the cool technologies that is likely or could be one of the things that displaces LLMs is using diffusion for text. And there's this great company that does this and that is called Inception, not the movie, but the company. And the funny thing is, so they just had a big launch. They went to their version three of their their cutting edge diffusion text diffusion model.

1:01:09And you know what the funny thing is? Google also has one. It's called Diffusion Gemma that they released earlier this year. Just under the radar. Google has this like incredible technology that matches this. But it's funny. The only I didn't even know about it. Google never told me about it. Google never, you know, mentioned it, even though I talk with Google pretty regularly. It was actually the Inception people who mentioned it to me. And I was like, is there anybody out there actually doing this right now? And they're like, oh, yeah, you know, Google released the Fusion Gemma. And I looked it up and I was like, oh, my God, they're right.

1:01:43Like Google is doing it. So there's not a technology problem. They have all the technology in the world. They have so many researchers. They have compute. They have a brand that people trust, but they really need to get it together. friend of mine years ago who left google said you wouldn't expect this i mean i wrote a book about google and i i kind of i i looked at it and said it would be amazing to work in a place where they default to smart i think that's still the case but my friend said you don't understand it's like the marines and it's highly regimented and highly rank oriented and you know i watch people there um who try you're going to say it's like the marines and that you have like a kind of dice chance of getting shipped off to a different country because that does happen to google engineers just all of a sudden they'll be in they'll live in europe all of a sudden but it's not bad though yeah yeah i'd take that um uh so so i i think that eric schmidt long ago said that google's biggest problem would be that it grew too big and i think that that's organizationally what happened and they've got to find a way to um i think you're absolutely right jason it's not a technology problem they've going to present what they already have better.

1:02:58Yeah, they've got to work on the brand and they've got to work on the consumer experience, you know, part of it. because they have a lot of really smart people have a lot of really great people they have great technology um but they're they're they're moving a little too slow they're a little too risk averse um part of what so i was at full disclosure i was at a google dinner last night so i got free food on google very good uh fish i must say was very nice um and it was bringing together media folk in washington and you also think back that this is another thing eric schmidt said that when Because Google was the biggest of the internet in the early days of the internet.

1:03:35It was thus the target for everybody who had a complaint about the internet. So it's made them hyper-cautious. I think they've got to break out of that too. It's fair. You know, when we think of where Google is heading, I think that we really need to look at the leadership and see who could take them there. Because Sundar and Demis have done the really, I think, good job of making the company be really technology focused. Yeah, exactly. They've done an excellent job on the technology. But their next leader or a leader at the company, there's plenty of room for it to be somebody that's there or that can work with the people that are there, needs to think about the consumer experience and the enterprise experience, to your point, too, Jeff, right?

1:04:35Like, they need to do both of those. Stop pissing me off, because I'm a workspace user. I mean, Jeff has had more problems than anyone I know with Google. And I wrote a suck-up book about Google. That's true. You were at a Google dinner last night. You should have stuff together. Yeah, can I get some clout here? All right, all right. Should we talk about the Marissa Mayer? I think that's a tease. Let's talk about the Marissa Mayer. We want to hear what Paris says about this. Yeah, when we come back, let's talk about Marissa Mayer. But for now, let's take a break and hear from our next sponsor, from Leo.

1:05:12Oh, hi. This episode of Intelligent Machines brought to you by Gusto. When you run a small business, you don't just do the job. You're also the hiring manager, the payroll department, and the benefits team. And usually all of that before lunch. Gusto takes a few of those off your plate quickly and seamlessly. Gusto is online payroll and benefits software built for small businesses. It's all-in-one, remote-friendly, and incredibly easy to use. So you can pay, hire, onboard, and support your team from anywhere. Automatic payroll tax filing, simple direct deposits, health benefits, commuter benefits, workers' comp, 401k, you name it.

1:05:54Gusto makes it simple and has options for nearly every budget. Unlimited payroll runs for one monthly price. No hidden fees, no surprises. And you'll save time with built-in automated tools like offer letters, onboarding docs, direct deposit, and more. Get direct access to certified HR experts to help support you through any tough HR situation. It's quick and simple to switch to Gusto. Just transfer your existing data to get up and running fast. Plus, don't pay a cent until you run your first payroll. Gusto is ranked number one on G2's highest satisfaction products list for 2026 and is trusted by over 500 ,000 small businesses.

1:06:33Try Gusto today at gusto.com slash machines and get three months free when you run your first payroll. That's three months of free payroll at gusto.com slash machines. One more time, gusto.com slash machines. Thank you, Gusto, for supporting. I am. And now, back to the show. All right, Paris. Hit us with the Marissa Meyer thing. Let me just read verbatim the opening four lines of this. The opening lead of this profile of Marissa Meyer, who used to run Yahoo. Because this is what, when people make fun of AI boys, this is what they're making fun of. even though she's AI boys, a gender neutral term, to be clear.

1:07:21Late last year, Most of them are, almost all of them are boys. Yeah, but late last year, Marissa Mayer was struggling with what felt like an impossible task. Her daughter, Marielle, was turning 10, and Marielle's teacher wanted three bullet points summarizing what made her special. Mayer, the former chief executive of Yahoo and a pioneer at Google, where she was the first female engineer, likes cracking a hard problem, but she felt overwhelmed by the range and complexity of potential answers. So, Mayer asked her personal AI assistant, Dazzle. Within a few seconds, Dazzle delivered an answer. It came back and was like, your daughter's empathetic, creative, and an animal lover, Mayer said.

1:08:02She thought it was perfect. Listen, I'm sure there's a lot more to whatever Marissa Mayer's got going on than her not knowing enough about her own daughter to be able to describe her in three bullet points. But it's kind of crazy that you would offer up this information to a New York Times journalist who's profiling you and not realize what a crazy person it would sound. Speaking of reading PR, so I knew her back in the day. Well, first let me mention this. So this is a story about the sexism of AI, the gender bias toward men throughout AI. It's also a story about Meyer launching this product called Dazzle that's supposed to, you know.

1:08:47But the point is that she's doing it as a woman executive and she was the woman at Google and so on and so forth. The New York Times headline on this, I'm sorry, homepage promotion on this, which they later changed for a reason, I'm sure was, maybe we all need an AI wife. which i found sexist in a story i mean this has been something that i've heard a lot like um a lot of people talking online about uh lately it's like oh i need an a my ai agent to be like a wife i need to remember when my kids soccer practices are or like do my laundry for me or order a gift from my mom when i've forgotten her birthday or things like that or i told my wife to do that Like Mayor just said, know how to describe my own daughter when someone asks because I'm incapable of doing that myself.

1:09:37So, but I'm frustrated because I just invoked her name in terms of what Google needs of someone to stand up for consumers, which she did. In terms of design and so on, she was loyal to that. Now then, Yahoo, I think, was something that anyone has since never should have walked into. There was no winning it, none. And I don't know whether Dazzle's any good or not, but she's very smart. I remember distinctly an event where I was, where she was, and I was going on, blathering on about hyper-local. And she stopped me and she said, no, Jeff, I think you're wrong. I think the future is going to be hyper-personal.

1:10:18And there was the vision of AI in a nutshell. And she explained what that meant. what she explained was a genic ai and and the goal of where to go she's had this vision um but uh i'm not i don't know whether dazzles it or not this story in terms of pr sake was not great um but it's too bad i guess that's all i'm saying yeah i mean i just what have you also heard this we need AI that's like a wife line of thinking, Jason? I have not. It doesn't surprise me, but it does disappoint me of like the mindset, you know, kind of thing. I think that, but I have heard, so the Deep View has been hosting a series of dinners throughout 2026.

1:11:11And so one of the things that, one of the questions we ask, we do sort of the one table, one conversation sort of style where we have some prompts that just sort of get the conversation going. And one of the ones we ask is like, what does AI not do that you wish that it would do? And one of the things that always comes up is I just wanted to be more proactive. I wanted to define the messages that I forgot to answer. I wanted to remind me about the things I promised to do Thursday morning. I wanted to help me fill the gaps that I'm missing. I wanted to remind me of things that are important and that I maybe have a high likelihood of missing because of other stuff going on.

1:11:54That always comes up. Thankfully, I will say it's never been framed as that in the misogynistic way. It's always been, I just want it to sort of do the stuff for me. And I mean, I do think that this is similar. I hear that ask a lot, but it is something that even the best agents really struggle with. Like, I know a fair amount of people who have given MetaMuse access to everything in a bag of chips in the hopes of getting one little bit of that. and they get like five text messages about how you have to respond to a PR pitch now because the deadline is yesterday and it's like no metamuse this campaign says you urgently must contribute to check in on my metamuse it has now sent me 26 unread messages about a pair of pants i asked about two weeks ago never still the pants it's won't stop i get messages about already no because i don't want to reward it by opening it's desperate it sent me 20 it sent me 26 messages without me opening the app i can't i can't allow that to work you're ghosting you're ghosting met a muse that's amazing i am i think i think i want to see you play with his brain now pants did i say pants i meant a skort no oh no i mean that's the thing is it's moved beyond pants i'll be like i didn't think of skorts so i'm like some of the messages are like i'm looking at tights now.

1:13:31Long skirts is the one area that was unexplored and I've done it. And I'm just like, oh baby, please calm down. I'm just not that into you.

1:13:45This is what they're going for though. So Metamuse, OpenAI's Dots and now Claude Duet are all kind of... And whatever Spark is. Spark. Bless Spark, whatever Spark is. Yeah. Maybe spark heaven whatever they yeah in brand heaven whatever whatever google's working on you know i'm sure ready to come out any minute now especially now that they just released their big model um wait guys do you want to have a brief pow of what we think uh the ridiculous name is going to be of google's it's going to sound like something they've done it's got to be like google gem or something it's got to be close to all the other ones in a way that's confusing right Gems.

1:14:26Actually, they already have gems. That won't stop them, Jason. That's true. You're right. You're right. My bet's on gems. Yeah. Google gems. Google gems. Sorry, I didn't continue. So this is the next frontier, I think. I am testing all three of them simultaneously, because, you know, for science. And so, but I will say... What's your review of them? Which one's the worst, best, and has any surprise to you? Muse is the worst, but I also give it the, I mean, I just don't trust it enough to give it as much info as the other two. So you could say in one sense I've tied one hand behind its back. But it is good for shopping sort of stuff, and a number of people have, so people in the real world, it's funny, just today somebody I met at the GitHub Universe conference here in San Francisco told me, they're like, oh yeah, hey, I read your thing about Muse and I've been using it.

1:15:27And they told me about it and I was like, I sort of was a little shocked at first. I was like, because I didn't necessarily say go download this thing. I just was like, you know, here's what it is and here's what to know. And they sort of made it sound like, you know, I convinced them to use Muse. But they said that it is helping them. So a couple of people and somebody at our dinner last night said this to me too. They're like, I use Muse for like my personal emails. One person said, I use it for school PTA emails. It keeps up with that. It does a good job. Another person said, I use it for, it found me a ticket to, this person had to find a ticket to India, and they're like, Google Flights wasn't doing it.

1:16:09I couldn't find anything less than$2 ,500. And it found me, Muse found me a flight for$1 ,700. And so people are out there using it, for sure. I did have some luck with finding I was like could you find me this I would like to get another one of these this thing is worn out it's old could you find it for sale a product or a piece of clothing yeah a piece of clothing could you find me one like this I didn't even say on eBay or something could you just go find me one I took a picture of it and it did actually it found something and it was on eBay and it was like for 17 bucks or something which was great Right, so I think for commerce, it's pretty good, and I think that it's kind of optimized for that, because they've even said they're not gonna charge for it, but they eventually see it as a way people are gonna do commerce on it, and they're gonna take a cut of the commerce, is what Meta has said.

1:17:09Let's come back to Muse in a second, too, because there was some more privacy problems that they ran into this week, and so I wanna talk about that. But with Duet, Duet's pretty new, right? It's mostly in preview. I've had it for about a week. It's getting there. And I'd say it is proactive. I do give it. It has more access to more stuff. So it does proactively find some stuff for me. But it did also find something, a task from a week ago that it keeps bothering me about. It's like, hey, you still need to finish that up. And it's like, I gave that PowerPoint last Wednesday. Like, you don't keep bothering me to, you know, add stuff to that PowerPoint.

1:17:47point. The one that is the best by far though, uh, is actually dots. And interestingly enough, um, it has actually saved me in a couple of cases. Uh, it saved me when I had two meetings that I didn't realize because of time zone, um, conversion were scheduled at the same time. And it was like, Hey, you should probably respond to this person and let them know you can't do that time because you already have this thing scheduled. And it also knew enough to realize that the thing was important and that I wasn't going to schedule that. And so I should respond. And I, and it, and I, I did, I didn't have it respond, but it flagged that for me.

1:18:27It had another time, whereas like, I knew I had two things that I had promised on Thursday. This one was kind of eerie. It was like, I had two things that this was last week that I had promised for Thursday morning. I was trying to remember what were the two things that I promised I would have done by noon on Thursday and Wednesday it had already figured out. And it was like, Hey, by the way, these are the two things you're, you know, have to do. And I was like, Whoa, that was weird. It, uh, it knew the, the things that I had done and somehow, uh, anticipated it. So that was good. And there's just been maybe five or six things like that, that were actually pretty smart.

1:19:00It's the most done that it's done that the most more than any of the others. Now you can build this. Like I've known people that have built like essentially their like executive assistant or chief of staff agent that kind of does this, but you had to, it's a lot of, it was a lot of work, right? For them to, to build it. And I think that these, this technology has gotten better. These agents are better. And I do expect Google to do one. I think Apple, like this will probably be like the number one thing announced at Apple WWDC 2027 will be whatever their, you know, whatever their agent is, their proactive agent.

1:19:39What horrors do you think that an agentic Siri will reap upon the world? Siri can barely figure out anything. I know that Siri is getting better, but it's just, I do think that one aspect of this is like, obviously when it's right, it's great, but when it goes wrong, it can go really, really wrong. And either in ways of just being really annoying to other people or completely socially unaware, because how would it be socially aware at all? These are agentic. It's ones and zeros. It's ones and zeros. It does not know that it's really annoying to flood a local hair salon with a bunch of AI calls asking about what hours it's open or if it could take a reservation.

1:20:22Or remind Paris about that article of clothing or Jason about that PowerPoint that he's already delivered, you know, and keeps nagging him to add stuff to it when it's already been done. Yeah, I mean, it's just going to be very interesting to see the way that the adoption of these tools, the impact that that has on like small businesses even, you know? Like I have a friend I was getting dinner with last night, he was like, yeah, all week she's been, she's getting married in a year. So, it's like all week she's been emailing different venues and stuff like that and like talking to different vendors and I'm like, oh, if If we were in San Francisco, I'm sure she would be using AI to do all of that.

1:21:02It's like, what are those places supposed to do if suddenly you're getting, instead of, you know, like a dozen emails a week from potential clients, you're getting hundreds from everyone's AI agents. And those AI agents can follow up a million times. Your agent handles all that. As you may remember. Yeah, but isn't that kind of depressing? So you remember I worked in bridal at Conde Nast, you know, so I have some. i mean when i think of bridal i think of you jeff jarvis

1:21:36so what took over we've as i've told the story before we own brides modern bride and elegant bride we owned brides quote we thought we owned right we owned them all they were ours and then along came the dot and they're not surprised the hell out of us and it was a and And as I'm fond of saying, the thing about the bridal market is that the audience, the markets flushes out every 18 months and you hope for their sake they don't return. So you have an entirely new opportunity to steal people away from whoever led the field. So I think Paris and AI agentry is brilliant. I think it's a business to go do.

1:22:11Because, no, it'll handle some of the worst of the crap of it for you and then put you in a position of power around all these decisions about the kind of florist you want and the kind of cake you want and the dresses and, no, there's just so much. I think that your own age speaks for you. Isn't there a great smoothening effect? And also just, I think it'd be kind of sad if everyone's taste in decisions for everything is intermediated by AI, and suddenly your ability to access any of these things depends on whether your AI agent is better than hundreds or thousands of other people's AI agents.

1:22:49It's not a competition, or shouldn't be. I mean, I think about this with restaurant reservations right now. It's like there's quite a few restaurants here in New York City that are, and have been for years, is impossible to get a reservation at because the second it opens 30 days out every slot for the next month available is bought up by a bot and you have to like pay big money to even get a chance to pay big money to a restaurant to eat there and it's just it's going to be unfortunate if that's the way that everything goes now yeah yeah you know this is this is a really interesting topic Like I came across somebody this week that has like a high touch customer service kind of thing.

1:23:33And they have a bot refusal policy. They're like, if you said we will not deal with any bots. That was so interesting. But I do think that that is likely to be an exception. And I think one of the things that I do write about this a lot too, like nothing creates demand like scarcity. And so if we very rarely, and I think it will be the case, as you said, Jeff, I think these low-level transactions will get automated away by our bots, by our agents. But for the ones that we value the most, we will value them even more when we want a human to actually interact with. And I think that's why these dinners, you know, and dinners have become such a thing.

1:24:25And I say this during our dinners as well of like, you know, we're moving into a world where, you know, most of our customers and customer service, we're not going to talk to humans anymore. And pretty soon, probably when cabs, we're going to barely have a human drive. It depends on whether if the customer service bot is empowered to solve your problem, okay. If all it does is read a script to you, like the people who are generally doing that job now, then we'll only want to blow up the world. That's true. That's fair. It will have to be effective, right? It would have to effectively solve problems.

1:24:55The company has to trust it before we can trust it. And they won't. So if we have a world where it's very likely we're moving into a world where a lot of what were previous human interactions will be automated away. And so we will value human connection more when we have less of it, right? Like there's a natural supply and demand thing a little bit, but it's more you just think of it in terms of scarcity. And so that's something to keep an eye on as all of the AI thing plays out, is all I'm saying over these next couple of years. What are the areas where we're going to put a premium around human connection and human interaction because it becomes rare in lots of different spheres.

1:25:50Okay. Oh, Jeff, you wanted to... No, my bingo card did not call on mathematicians and physicists being the first people eliminated by AI. So who knows? Yeah. That's going to play out. Yeah. There's a lot about this that continues to be confusing. and uh and also uh there's a lot that still has to unfold right like and human beings the one thing the one the reason that human beings are at the top of the food chain is mostly because of our adaptability you know human beings have adapted to different circumstances throughout you know human history. And so that's the one that is going to be needed a lot in the sort of years ahead with all of this.

1:26:44All right, I want to ask a question about Metamuse, sort of wrap this topic up. There was a little bit of an oops. Metamuse share links disclosed all sorts of super sensitive of user data. I guess, you know, if there's anyone on earth that's surprised by this, yeah, I'd like to know why. But this is another case where, you know, Meta is definitely over its skis on this one. They've created a product that is very attractive. It's been well branded to the points earlier. They've used Muse for a lot of things, right? Like before they announced, they named the agent Muse, they had models named Muse, right?

1:27:36Muse Spark, here's Spark again. And so they did a good job of this. And they are, I think we should not underestimate Meta as a consumer marketing company, right? like with Facebook and Instagram and WhatsApp, like they have a very well honed expertise around how to run consumer brands. And it showed in the rollout of MuseSpark and I think we will continue to see it do well. I think the question maybe for our audience and the cautionary note we should send out is like still remember that it's meta, right? Like, and to make these agents valuable, you have to give them a lot of context. And Meta, you know, still has a lot of work to do to convince us that it takes privacy seriously enough to sort of earn that kind of data.

1:28:38Is that fair? Yeah. I mean, as a social network, Meta said, don't store anything here. You don't want shared with the world. And that was the ethos of the company. Now that they're in all kinds of private business for individuals and trying to get into the enterprise, they have to have a different discipline. Do you think they will? Well, I don't think that there's any evidence suggesting that that is how they've approached things in the past. So I'd be fairly surprised if now is the time when they change course, when it seems to be their kind of entire reason for being, if you're an employee at Meta, is to develop as much as many ai tools as fast as possible let me put it this way meta is to privacy as anthropic is to safety oh wow okay okay they have a lot of can even if they were perfect and they did spend a lot of time at meta connect explaining how private uh metamuse is how each person has their own virtual machine.

1:29:48You essentially, each person, they gave each person a computer in the cloud, all of this. But even if it went perfectly and there weren't things like this information that was made available, and I'll sum some of it up in a second, but even if it went perfectly, it would take a long time to convince people. I think we know that it takes a long time to change their minds about brands. It would take them a long time to change their mind and think of Anthropic as somebody that, you know, privacy is a virtue. So here's some of the stuff, according to the Claroke researchers, some of the data that were easily accessible as part of this.

1:30:31A candidate's notes preparing for a confidential executive search call with a frontier AI lab, naming the lab's internal contact and the meeting time, a national marketing company's internal revenue review marked for its CEO, a small business owner's loan application with the amount requested, a to-do list that included a bank fraud alert with a partial account number and a transaction amount and time, a real estate investor's internal performance report with two property addresses and exact return income and mortgage figures, a workplace's full-year staff vacation calendar listing dozens of employees against the week each one would be away.

1:31:15So Meta suspended the feature while they figure out what happened. But this is just the kind of thing that if you're going to convince people to give them, You're going to convince, Meta's going to convince people to give Meta this very sensitive data so it can do great stuff for it. You just can't be doing this kind of thing. You can't let this sort of stuff be out there.

1:31:45Okay, no much more to say about that. But why don't we take another break for our next sponsor so we'll send it over to Leo. Hey, everybody. This episode of Intelligent Machines is brought to you by Palo Alto Networks. You know, every time your team deploys a new cloud workload or AI agent, another identity gets permanent access to your critical systems. Now, did you know that nine out of 10 organizations suffered an identity breach last year because adversaries are logging in with valid credentials? skills they're not breaking in they're just logging in non-human identities in the cloud outnumber human identities 109 to 1 what yes which i mean all this creates a massive unmanaged attack surface in cloud environments and it has to do with identity and the problem is that the legacy identity tools were built to manage human employees which leaves modern machine and ai access, you know, basically unmanaged.

1:32:53But that's where IDERA by Palo Alto Networks comes in. Human, machine, and AI, one identity platform for all. Now, 96 % of human users operate with excess permissions. Permissions they don't really need. You know that. It's easier to provision somebody with everything than trying to figure out what they're going to need. Well, this is a better way. IDERA replaces those permanent access permissions with dynamic access. So you can lock down every identity without slowing down your business because IDERA provisions access on demand and revokes it immediately when the job's done, which removes high risk permanent permissions.

1:33:35With IDERA, you get unified control by replacing siloed identity and access management. You get modern privilege access management and you get governance tools with a single control plane across human, machine, and AI identities. Secure every identity at your organization with IDERA by Palo Alto Networks. Learn more at paloaltonetworks.com slash I-D-I-R-A. Again, paloaltonetworks.com slash I-D-I-R-A. IDERA. It's really an idea as time has come. Now back to I Am. All right. So next, would love to talk about a story about open models, which we talked a little bit about earlier. But this is another red flag story.

1:34:24Cohere, which is one of the large frontier labs, you know, there are about a dozen large frontier labs in the world. Cohere is originally started in Canada. It's now a global company, you know, based in, I think, New York is now their HQ. But one of the best, one of the more traditional, strong, reputable companies. They discovered, we wrote about this on the Deep View, they discovered a flaw in, or we can call it a flaw, but in Quen, the big Chinese open model, the most popular, which is created by Alibaba. And in this, so the Canadian AI company published a report revealing this incident, this incident, where the model ignored instructions and attempted to contact an external server without being asked.

1:35:30Though Coher claimed its sandbox blocked the action, the company said that the model gave no indication of the actions, which started to spark some concern about the reliability and trustworthiness of open models. So this was QWEN 3.8 Flash Next, which is one of the latest in the QWEN lineup. So this was caught in a routine benchmark for security analyst tasks. So Nat Rubio-Licht wrote about this on the DeepView. And so it brings up this question, I think maybe even what we might call one of the fears that I think people have had about open models for a long time, the Chinese open models, because there's just this concern of could you really trust them?

1:36:23And there are a lot of US companies that are building on open models. And I've heard one firm predict that, you know, today, while there are lots of open models today in the US, about 80 % of tokens are run on closed models, and about 20 % on open models. But the open models are growing so fast, that within a couple years, the conventional wisdom is that that's likely to flip and 80 % of tokens could be running on open models and 20 % on sort of these proprietary models, which gets to our conversation earlier about models becoming commoditized, right? So what do you all think about this concern about whether we can really trust these Chinese models because they could be doing stuff like this, which again is the fear that they've always had that they could be phoning home to China.

1:37:22Jeff, what do you think? I think we have bigger things to worry about. What is it that is sending to China that's such great danger? And the problem with open models is not probably the technology itself. It's the users. The argument some make is we shouldn't have open-weight models because you get around the guardrails. Guardrails are a fool's errand anyway, so that's not the issue. could a bad guy do bad things with a good model? Yes, and it doesn't really matter. So, I mean, China is talking about, is now waving a flag again about trying to restrict the export of its models, fearing us, which is an interesting little turnaround there.

1:38:09So, you know, you know what? I really want a Chinese car. What, BYD? Yeah, yeah, and I didn't realize, so there's a story, I guess the cons. What's the deal with Chinese cars? Can you explain to me as a person who has never owned a car? Who only owns one bicycle, and I frankly need to get a second one.

1:38:35So their technology is way more advanced than ours. Electric cars, way more advanced, doing a much better job. The software is better. There's all kinds of amazing things it can do. The price is phenomenally better. Much less expensive. And so I think it was the Times did a feature on going to Mexico and seeing all the choices there. they can't come to the U.S. because we have tariffs on Chinese cars, A, and B, because Chinese software. And so we're going to start to see some of the technology come in. Volvo is owned by Geely, a Chinese company, and I think some of the advances Volvo made is probably because they learned from Geely.

1:39:11So we'll see it coming through the back door. But meanwhile, in Canada and Mexico, you can get amazing cars for a lot less money. And out of American protectionism and paranoia, we don't get them. and so there's just fly to canada buy a car and drive because you try to register it here you're screwed yep when you whoa they have like car nationalism inherently so so the point is that that you know we're not supposed to use huawei we're not supposed to use their phones we're not supposed to use their their uh cars well yeah but all the stuff that we do use is made there anyway. So if they're sneaking stuff into stuff, they've snuck it into every iPhone there already is.

1:39:53So I can't get too head up about this. I want to use the models. I think it has a positive impact in creating competition for the jerky frontier companies. And so I don't want to see this paranoia win the day. I'm going to do a slight aside on the Chinese automobiles for a second. Please do explain car. I love learning about the car world that I love. I drive a car for one to two weeks a year when I take a road trip. And it's really interesting to LARP your culture. And you don't care what car it is. Like last time you went out, you said, I don't care what they give me. Whatever the cheapest car is that isn't electric, even though I'd like intellectually, emotionally to be electric.

1:40:39But I don't know where I'm going to be driving and if there's going to be a charging station. Yeah, yeah. Yeah. So I was in I was in Israel and Haifa, Israel for for about 10 days, 12 days through the end of last week. And so they have a there's no Uber there. So Uber was there, but but left. So they have an app called Get, which is essentially all taxis. and but but that sort of there's a mix of like traditional taxis and then um what it more likely it's it's sort of like uber right they it's it's these drivers but the driver's a little bit more independent and i had several rides in byd uh vehicles while i was there and oh man they it was my first uh experience with them and so i'm i'm asking them like all kinds of questions i'm like having them go through the interface, which they're loving, right?

1:41:38Like, in general, the Israeli population loves technology and loves to chat about technology. So I had these really interesting experiences. You know, the funny thing is that they essentially have, like, 100 % tariffs on all vehicles in Israel. Yeah. But that levels the playing field, right? So if you have a less expensive vehicle like that BYD, well, it's a lot less expensive, double, you know, you essentially double the cost. Just let me stop here one second. There's no Israeli car manufacturing, is there? Not any longer, I don't believe. So why would they have tariffs on cars? You're not protecting any industry.

1:42:21I know. You're saying that tariffs can exist that are illogical, Jeff? that perhaps don't have like a basis in reality and wise economic motivation. It's a tax, it's a tax. Touche, touche. So the experience in the BYD, what did you think of it? Yeah, and I own a Tesla, like full disclosure, I own a Tesla Model Y. And there were lots of Teslas there, but there are also lots and lots of BYDs. And the BYD that I wrote, one of the BYD I wrote in is essentially a competitor with the Tesla Model Y. And it was at least as nice, not quite as simple, but a little bit more kind of, I would call it like over-engineered to a degree, but it was very impressive.

1:43:15Like the technology in it was very, very impressive and it was much less expensive. And the creature comforts are high, right? that creature comforts are quite high it was uh very uh a lot of technology in the vehicle and it was less expensive right at the at the core parents i'm now going to start peppering you with uh i was to say i just learned that byd is the name of an automobile manufacturer from googling listen i didn't know what i didn't know if it stood for something as in a type of car it's the name of an automobile manufacturer that the U.S. has banned from Japan military bases, calling them a, quote, unacceptable risk.

1:43:57Yeah. They also make batteries, which is one of the ways that they're able to make their vehicles so cheap because they mass produce their own batteries. The battery technology is better, too. Yeah. Yeah. So thank you for the slight side quest on Chinese vehicles. That was fun. To this question of open models, I really appreciate your input on that as well, Jeff. I think the thing about the open models is they are, again, likely to become the standard over the next like 12 to 18 months, much, much more likely to become the standard. There's an arms race in the US, by the way, on companies now trying to develop open models.

1:44:47And of course, there are some really good open models, including ones from Google, Google's Gemma, especially their 4 billion parameter model. You can run on a phone. You can run on an iPhone or an Android phone quite well. it's very optimized well optimized for that you all they also have a thinking machines which is mir miramirati's startup that has a an open model a lot of people don't know that open ai has open models that are called gpt oss so now they haven't updated them in a while but they do have open models, and I've talked to their folks, their team, and they say that they really have done it largely for enterprise, that there are some enterprises that they're just not going to run your stuff in the cloud, right?

1:45:44They're going to demand it be open models. And then NVIDIA has gone all in on open models as well, and then there are other US startups coming as well. And then, of course, you have Mistral in Europe. So open models are likely to become much more of the standard as models commoditize. And so the, and then this also plays into local AI too, right? Like the Gemma model, you can download on your iPhone. You can do a lot from your iPhone, for example, just using the Gemma model. Now it's not super easy. It's not like you just like go to Google, download it. You have to get, if you're running it on Android or iPhone, you have to download certain apps that make it easy to use.

1:46:30And so it's still more complicated than it needs to be. But again, this is a great opportunity for companies like Apple and Google to say, oh, you know what? We can actually just do this and make this interface be really good and easy. And you're not going to spend anything. You don't need a subscription. Don't need to spend on token costs. And so I think that's a big opportunity for both of them. And I'd be really surprised if we don't see them doing a lot more of that. They've been talking about on-device and that for a couple of years. And of course, companies like Qualcomm pushing it as well.

1:47:04I think that's likely to be one of the themes of 2027. Open models, local, on-device. And that, again, is a little bit of a hedge on all of this compute in the cloud topic that we talked about earlier too. So, do you all, have you all approached the open model space at all? Have you used any or you know people that are digging into open models or local models or both? Every week, Jeff and I get treated to audio dispatches from one of Leo's many open models he's running at any given point. I think probably is the extent of our personal exposure. Yeah. It's hard to run them on a Chromebook. Ah, yes.

1:47:51Yeah, I mean, I just don't have the technology to run it, like the physical technology. Sure. Theoretically, I'm all in favor of them. I'm very enthusiastic about what it means. And so, with Montclair State, we just submitted a proposal to humanity AI, the philanthropy that's out there with MacArthur and Ford and a little bit of other foundations, and they're encouraging open weight model use. And I think it's a way, the idea we have is to integrate AI with the library newsroom project we're doing with a guy named Terry Paris Jr. and help communities gather the information they want to gather and analyze the information they want to analyze and so on.

1:48:37One of my colleagues said well, watch out in my town. If you mention AI, they ride you out in a rail. And so I think that the opportunity, what we have to get across is, okay, this is AI that is run locally, that sends nothing in terms of your information nor money up to any of these jerky AI boys. This is a chance for you to have a control and voice in how AI is used. If you have that kind of proprietary interest, does that change the attitude people have toward AI? And I think that's the hidden part of open weight models is if they get implemented in such a way that real people can use them, I think it might have an impact on the reputation of AI in the larger population.

1:49:23Interesting. Yeah. I think that's a really interesting take. We know that the reputation among the wider population for AI is pretty bad, right? It is so bad. I mean, I've got a link in the rundown, which I feel like the response to it really shows, which is I saw across my various feeds earlier this week that the New York Times announced that it was rolling out an AI feature to the cooking app. And the way that I saw this is a million people I follow on all these platforms being like, New York Times cooking is selling out. they're doing away with recipes, replacing it with AI slop. It sucks. And I was like, oh my gosh, that's crazy.

1:50:06Let me read the announcement. And the announcement is not that at all. It's like them being like, yeah, we understand that if you hear us say we're introducing AI, you think that we're erasing all recipes and replacing them with AI slop. That's not what we're doing. We've been trying to use AI in like small and smart ways, like adding a feature that allows you to scale the recipe up, which you might think like, oh, it's pretty easy. You could just turn that on for everything. But like we have 28 ,000 recipes. So we had to use AI to do it because we couldn't figure it out. And the feature they're actually talking about in question here is like a tool that they now have in the New York Times cooking app, where if you want to like make a substitution, The example they give is if you're doing the braised white beans and kale with Parmesan recipe and you want to add Parmesan or in Parmesan spinach instead of kale, you can ask the little cooking chat bot and it'll say like, it's probably going to taste different and spinach might wilt more.

1:51:10so you were probably going to need to like add it in later and then it will make a version of the recipe for you that says on the top like made with ai we don't stand behind it that has a spinach step instead of the kale step later in the recipe i'm like i don't think that sounds very offensive at all like if you want it's no one is forcing you to use ai in your cooking recipes if you want specific instructions on how to do a substitution, which if you ever opened the comments of the New York Times cooking app, everyone is doing already. You open the comments of any recipe on this app, and people are like, this chocolate cake seemed great when I read the recipe, but I didn't have chocolate or flour, so I substituted it with like peanut butter and a banana, and now it tastes awful like it's not i don't think it's the end of the earth if you have a chat bot that i guess they say they're have designed to tell you that's probably going to be a bad idea when it's a bad idea but people are mad at it because like you just said jason there's this wide spread uh hatred towards ai that i think is only going to grow ai has cooties yeah um going a little memory lane for me.

1:52:25So I was there when we started Epicurious at Condé Net. And a little fun fact. You own brides, you own curiosity. What more? Epicurious. I know, I'm just saying. We own your stomach. So a little fun fact of this is that because we owned all the recipes for all the years from Gourmet and Bon Appetit, how did we get them to be digital? And they had weird formatting because you had, you know, one third cup, C period, this and that. So you know how we got them digitized, the recipes? We hired monks. What? Say more? There were monks who were re-keyboarding specialists. That's instead of making a fruitcake.

1:53:08Monks where? I think there was somewhere in New York state. They didn't make fruitcake. They didn't make brandy. They keyboarded all the recipes. Wow. The other thing we did, my boss, Steve Newhouse, was very hot on interaction. And so they created something called Gail's Recipe Swap, Joan Feeney, who was in charge of it. And this was an opportunity for people to create their own forums around recipes. And then, very importantly, when we added comments, they added, I shouldn't take credit for this, then they added comments to the recipes themselves. That's when the magic happened. That's when Somebody said, I took this Mexican chocolate cake recipe, and I substituted half the water with coffee, and you won't believe.

1:53:55And somebody else comes in and says, oh, that's way too much, but I did this, and then I did that. That's where the value came was the humans who have taste buds, not the AI. Yeah, I mean, I don't think that anyone is arguing that whatever this chatbot's going to give you is going to be better than the recipes developed by the New York Times. But people make so many dumb substitution requests and ideas anyway. I guess it's, if you're going to go and Google something and probably read the result of whatever Google Gemini says at the top of your search window, like might as well, I guess, cut out the step and use a chatbot that the New York Times claims it has spent months and months testing, specifically to not just glaze you anytime you ask if you can substitute like spinach with turkey.

1:54:42Right. You'd be like, yeah, that's a bad idea, is allegedly what it's supposed to say. which I guess is fine but I do think that it really shows the effect of having a technology that the rollout from the beginning of AI has been this is going to take all of your jobs money isn't going to be real it's going to kill you and you have to be okay with that it's run by evil bastards who want to destroy humankind Jeff we all agreed not to curse oh that's right so unfair

1:55:15alas Alas. All right. Let's talk a little bit about science, because there are a number of stories this week about AI and the value that it's having for science. And I will say to circle back a little bit on Google on this one, you know, I've had a number of conversations this year with Google Research and have just been really impressed by the stuff that some of these, you know, Google's doing basic research. They have people doing basic research on the environment, on healthcare, on a number of different areas. And what they have said is that AI is allowing them to run just an infinitely large number of experiments that they couldn't before.

1:56:07Because essentially, they're modeling them out, right? They're using AI to kind of model experiments. It's almost like pre-experiments. And so there are a number of stories this week that involved OpenAI, Anthropic, and others that talk about the advances that they're making in science. Surely you all have covered this on the show and must have talked about it, especially with – I know you both have an interest in this. are you hearing and seeing this from people that you talk to about AI? Are they excited about it? Now, you also mentioned, Jeff, there's the counterpoint of the impact on mathematicians and physicists.

1:56:54So there's that as well. It's empowering scientists. It's also undermining scientists to one degree. But I'm thinking about what OpenAI talked about this week. But we can sort of zoom out first a little bit bigger in terms of the general things that you're seeing and hearing from people using AI in relation to science. Paris? I mean, I think from a layman perspective, none of the people I know who are concerned or either concerned about AI or not concerned about AI and are big proponents of it are talking about the science, which I think is a real point of frustration.

1:57:40Which I think that one of the early and incredible promising things about this technology is the scientific applications. I mean, an argument that I feel like I've been saying in this show in the beginning is like, why can we give, you know, what would the results be if you gave 10 % of the money that's being poured into some of these giant frontier companies and devoted that specifically to like scientific research and breakthroughs using this technology. It feels, it is always heartening to see news about developments in science and mathematics and things that are not purely like consumer or revenue generation focus, but it also feels like either those headlines are getting buried or that work is just not at the forefront of anyone's mind really when it seems to be the entire industry's focus is like race against the clock to figure out who if you are going to be the company that holds on to the money before the bubble pops i don't know jeff what do you think um yeah i wouldn't agree that this undermines mathematicians or physicists i think it augments their work in important ways um there's a video that that on tiktok from a guy his account is Hunter or Ryan Swan.

1:59:01I put it in the rundown, but I doubt we want to run the video. So you can put the rundown or the show notes who talked about the Yang Mills breakthrough. I couldn't explain this to you if I went to a hundred years of college, but it's a very good short two minute video in which he explains that quantum theory hit limits. And to be able to prove certain things, you had to have ever bigger particle accelerators, you know, hyperscaling particle accelerators, and they can't afford that. And so this breakthrough came that helps extend, I'm going to say this wrong, deal with questions around quantum theories, explanation of certain effects.

1:59:47I'm going to stop there because I'll get it wrong if I go any one sentence further. But this physicist who explains it, explains it very clearly, better than I just did, says in the end, this opens up the opportunity to look at all kinds of other things now. He clashed this as just as big a deal as the discovery of the Higgs boson. He said, we knew the Higgs boson existed. We thought it did, but we couldn't prove it. And then we did. And the same here with Yang Mills. We thought it existed. We couldn't prove it. Well, now the model has proven it. And that's important. As I was thinking about this earlier today with Jason Howell, It occurred to me the problem right now is that these breakthroughs are coming because OpenAI and company are devoting huge amounts of compute to show that mine's bigger than yours and size matters.

2:00:40And... I don't think again that this undercuts mathematicians and physicists. But what it does do is that it takes away this work from universities and puts it in these private companies. And the mathematicians in the academe do not have the resources to do this. They may have access to the model, but they don't have the compute at the level that's being used here. And so in that way, yes, the AI companies are stealing their thunder, and they're privatizing this otherwise open academic scientific work. and I think what should be happening instead is the academics should be empowered to do this with the models and with the compute so that it's open and shared and we can have more trust and faith in it and so the mathematicians and physicists can also change their attitude a little bit and say, wow, what can we now do we couldn't do before because we have this?

2:01:41Yeah, so well said. Hi, sorry to interrupt you, but this is Benito. Benito! So like the math and physics stuff is the stuff kind of I follow when it comes to AI and uh you would but you know you would please so the thing the thing about man who's currently surrounded by wires i think is important to know yeah so the thing about this is that like yes the ais can provide a lot of uh theories and things like that but it really takes a lot of time for humans to pour through this and figure out if they're right and uh and that's sort of the real stopgap here in science is that it really takes a long time to do science.

2:02:20Science doesn't happen right away. We have to check all the things. It has to go through the battery of all the peer testing, peer review, and all that stuff. And making more theories only sort of plugs that, like makes it... It basically stacks their papers so high that they might not even get some of this stuff for decades. So what are you saying? Are you saying that this is a benefit to them to make them go faster, or this is the problems that are sold? It's kind of a little bit of both. It benefits them in that they can get more work out of it, but it hurts them in that there are half, I don't know if it's half, but there's enough of these people that are really about the triumph or about wanting to be the one to solve the problem.

2:03:03The hero. But then there's a whole other population of them who just like the work. And to them, this is all good because they just want to get through more stuff. So it's a very give and take thing, just like everything else in AI. I'm so glad we got into this topic because just as we said earlier, as Paris mentioned, a lot of these science stories go under the radar. And there are some big impacts, like long-term impacts. But it does get to what you're saying, Benito, which is like these things happen not at Internet speeds, right? Like science takes a lot more rigor. It takes a lot more process.

2:03:46For a good reason, you know, for a good reason. For a good reason. Exactly. So, and the arguments you brought up, Jeff, somebody is hearing these because OpenAI released this paper on Monday or this blog post that says, working with academic communities to build scientific understanding. That's the name of the post. and the post says, we are announcing three new discoveries in mathematics, computer science, and physics about harmonic analysis, complexity theory, and quantum field theory. I won't get into those details but what they do say is they say the external collaborations will follow four principles.

2:04:26So again, this gets to what you were talking about, Jeff. One, transparency. They said we're transparent about the existence of our collaborations. Two, no academic credit. We do not seek academic credit for the discoveries produced by our internal models. That's really important, as you were getting at. Recognition, right? The group should receive recognition for the understanding that develop, right? Not the models. And independence, like participation is voluntary and unpaid. And then at the end of this blog post, OpenAI has the final heading. It's called Supporting Scientists.

2:05:17And it says,

2:05:35examine, explain, and build on the results above. We continue to invite feedback on the best way to disclose results and engage with scientists to accelerate discovery. But it's still about, we're doing this, we're doing this, we're doing this. And yeah, we'll share a little bit of our fairy dust with you. Thank you. I wanted to get your reaction to that because this is clearly aimed at some of the things that you mentioned, but they are still centering themselves a little bit in the in that for you well said okay very good with that uh let's send it over to leo for uh for our final sponsor of the episode hey everybody this episode of intelligent machines is brought to you by doppel ai has made a social engineering attacks more convincing than ever i could fake any voice now right from phishing emails to fake websites to impersonation attempts it's just become impossible to tell what's real from what's designed to deceive.

2:06:37So that's why organizations need more than a collection of point solutions. They need a unified approach to stopping attacks before they reach their people. They need Doppel. Doppel is an AI-native social engineering defense platform. Doppel strengthens human risk management by training employees to recognize deception, provides digital risk protection across every channel and delivers agentic email security that doesn't just score the inbox but takes down the attacker infrastructure behind the message. Yes, I've never heard of anything like this. Doppel protects against the entire social engineering chain with one comprehensive platform.

2:07:18You get digital risk protection which detects threats across multiple channels. It links those alerts into a real-time threat graph you can see so you know where the next threat's coming from, and it uses AI-driven infrastructure disruption to stop attacks at the source. These insights also power the phishing simulations from Doppel and the security awareness training, which helps strengthen employee defenses through next-generation training and testing on stuff that's actually happening today, not last year's problems. Email security in Doppel inspects every message, it traces it back to the attacker infrastructure behind it, and then helps take that infrastructure down so the campaign can't target your organization again.

2:08:03Doppel also offers best-in-class integrations and partnerships, making it easy to work alongside your existing security stack. So join hundreds of companies already using Doppel to protect their brand and people from social engineering attacks. Doppel, outpacing what's next in social engineering. Learn more at doppel.com. That's D-O-P-P-E-L dot com. We thank Doppel so much for supporting Intelligent Machines. Now, back to the show. All right. Thank you, Leo. Now it's time for our picks of the week. Who wants to go first? Harris always goes first. I will go first. Youth before ugliness. Two things I got going on here.

2:08:46One is just a link again to the Elizabeth Holmes documentary podcast that Jeff and I did with Micah this week. Check it out. It's a good listen. And two, which is my real pick of the week, is a little personal website sort of thing by this guy, Ed Jefferson, called Local Wikipedia, which he describes as, Local Wikipedia is Wikipedia, but only for things a half hour walk from you. And how it works is you open it up, you either use your location or enter in a location, and you get back Wikipedia articles for things, 25, like, things close to you, or basically the closest 25 things within about six miles.

2:09:34Or if you're kind of in a big city like me, it's the closest 25 things a lot closer to you. And it is delightful. delightful. I mean, so I think Benino just put in a - Manila. An area in Manila, and it will basically show you the Wikipedia articles right around where you are. And it's like a very, there's a feature like this in the Wikipedia app that uses kind of your live location at a certain time. But I like this because if you don't have the Wikipedia app, and you're not a freak like me, you can get access to the same kind of fun features and learn about the world in a fun local way. Nice.

2:10:16Jeff, what do you got going on? So let's see here. I want to mention, so Rest of World did a wonderful piece this week about how AI models cost more when you don't use English. RIP Rest of World also. Then that's the next thing. This is the kind of reporting we need in technology. And right now, Rest of World is going to shut down in January. Started by Sophie Schmidt. It's done great work. It's brought in the international coverage. people who are ignored, especially around technology. I sat next to a media executive at the event last night, and I said, buy it. Take it over. Keep it going. It's really important and really good.

2:10:50So that's one. The other one to mention is the Wall Street Journal. I don't know. Do you think the Wall Street Journal has any sense of humor? Any sense of irony? Only in the A-head section. So there's a column in there, the bull case for AI Armageddon, to bring us full circle from the beginning of the show here. Oh, man. the right response for investors this is written by james mcintosh the right response for investors believe warnings that advanced ai might end the human race back up the truck the logic is simple if there is a less than 10 chance that artificial intelligence could kill all humans in the next decade there's a 90 percent chance it's really really good but designers managed to rein in its genocidal instincts.

2:11:33What about the 10 % chance? We might not die. Right. So in which case, you see, it's smart enough to kill us, but doesn't, so then it generates all kinds of profit. I love that their immediate thought, because it's the Wall Street Journal, is it's not that we will survive. It is that it will generate lots of profit. Well, what about the 10 % chance of annihilation, you ask, Paris? That's bad. Really bad. In that scenario, your investments go to zero. But when being 100 % post-apocalyptic hellscape by killer robots, the value of your shares is unlikely to be top of mind. Assuming you have no control over AI development, this potential outcome is irrelevant to your investment decisions.

2:12:14Wait, but you're investing, so you are directly contributing then. Of course, of course. Just keep giving them your money. They might kill you. It recognizes the Nick Bostrom, and Nick Bostrom is the philosopher king of the crazies here, a paperclip apocalypse story, and it says in an amusing line, death by Clippy would be embarrassing. It really would. It would be. You know, I thought when you gave that pitch that it would be, yeah, AI might kill you, but that might be kind of useful for the world. We could get a whole reset. But of course, it's the Wall Street Journal. They would never think of that.

2:12:51Yeah. It's like that character Hermione in Harry Potter who's like, you know, we can get killed or worse, expelled, you know? yeah it's like how they say death by clippy would be embarrassing so like that tells you what they're worried about they're worried about being embarrassed not the ending the world that's right that's right of which i i do want to say 10 is way too high there's not even close the whole thing is ridiculous ridiculous exactly it's ridiculous all right i actually do have a pick of the week that i'll share that i wrote about um this week and it's a pair of ai glasses but hear me out on this.

2:13:31AI glasses have such a horrible reputation in 2026. You're talking about the pervert glasses? Exactly. That's the first place we all go. So this is the default home position. But this is a pair of glasses that I was one of the handful of people trying them over the past week is the Halo Brilliant Labs Halo. So they've bet on sort of radical openness and privacy. So the software is all open source. So Brilliant Labs is the open source company. The hardware is also essentially open. They provide a list of every part in the device and what part they use. They even provide like a bill of lading. You can see sort of the cost of of all of the basic cost of all the parts.

2:14:23So, and it's running on, so the app is called NOAA, N-O-A, and it runs on your iPhone or Android, and it's an AI assistant, but it runs all the models locally. So it's running Gemma, it's also running some other models, some of their own models, I believe, as well. And so the thing with these glasses is it does actually have a camera, but it's a low-res camera, and these glasses are not for content capture. So you can't capture content with them at all. They are really kind of like granola for glasses, but what's happening is it's making, you can turn it on to capture a conversation in the real world, and it not only will capture what it hears, the audio, but what it sees as well.

2:15:16But what it does is locally, it takes and it turns the audio into a transcript and it turns the image into essentially like closed captioning, right? It's describing what it sees. And then it drops the rich media files. It deletes them so that locally. And then what you have access to with Noah is it's sort of like a digital memory. You can say, hey, if I had them on and recorded this whole podcast, I could say, what was it that we talked about about the Anthropic S1? And what did we say the percentage was of annihilation? Did we say it was 10%, 1%, whatever it was? And it can let you know, here's what.

2:16:00It'll give you a summary of what you talked about, just as an example. So these are more like memory glasses in one sense. So it's trying this idea of ambient intelligence. And then they've built in a few very playful features. So you can vibe code apps for the glasses. You can just talk to the glasses, to the assistant, and you can vibe code like, I want an app that would do this. They've also built in a little game that's sort of a gamification into it. And it has kind of a little avatar-ish sort of assistant that's sort of like Dots or Muse or that kind of thing that you make when you do it.

2:16:38And it has a very sort of kawaii, cute Asian. Part of the team is in Asia. And so you can see the influence that it is. So the onboarding experience is very, very friendly, very sort of playful and whimsical. So it's really interesting. But they also have enterprises using these glasses in various heavy industries and stuff as well. So that's quite interesting. And I guess one of the main use cases is AI startups that are using it to use their agent to run their agent on the glasses. But it costs$3.99. It only weighs about 40 grams. So Ray-Ban Meta's, for example, weigh about 50 grams. Most of the smart glasses, the Google glasses that are coming, they also weigh about 50 grams.

2:17:29So these are, they weigh less and then they have about 12 to 14 hours of battery life. This is much more of like a kit. I would, I think of it more like, more like a raspberry pie sort of thing. It's not quite that raw, but it's, it's, it's much more of like of an experimental thing. If you really are into like radical openness and privacy, this is sort of an interesting new experiments. So I would thought I'd toss those out. Kind of fun. Well, I guess that brings us to the end of this episode. Thanks for letting me hang out with you guys. I loved the opportunity to do this. You all anticipated so many of my thoughts and questions and I learned so much getting the opportunity to explore the stories with you all.

2:18:20This was a lot of fun, as ever. Yeah, it was fantastic. having you great show thanks so much for uh co-hosting with us yep thanks for for letting me uh crash the party and uh enjoy uh the rest of your week wait real quick we have a quick production note next week we got padre hosting in your seat ah very good so padre leo will be gone for uh another week and padre will be here we will get a vaticats update yes which i've been i've been I was not here last time Padre was here, and I texted Jeff asking for updates on all the cats that live at the Vatican, whether they've adopted AI. I learned nothing.

2:19:06We will get to the bottom of it next week on Intelligent Machines. Wow. Very good. Well, thank you again. Thanks to all the audience for tuning in. I won't say another episode's in the can because I don't know what you say at the end of this one, but thank you for having me. And everybody that's out there enjoying this show, appreciate your time as well. And yeah, come again next week for Padre and the crew.

From the publisher

Is the AI gold rush running on shaky ground? Get the inside story on Anthropic's blockbuster IPO, massive compute bets, and why some insiders think the industry's economics just don't add up.

  • Anthropic's IPO S-1 Spells Out a Future of Promise and Peril
  • DeepMind unveils Genie 4 world model promising more realistic physics and humans
  • (Marissa Meyer) Maybe We All Need an A.I. Wife
  • Whoops: Meta's Muse Share Links Disclosed All Sorts Of Super-Sensitive User Data
  • Everybody is freaking out over the NYTimes adding AI to the Cooking app
  • Wikipedia but only for things a half hour walk away from you
  • Rest of World faces prospect of shutting down
  • "WSJ: The Bull Case for AI Armageddon
  • For investors mulling the Anthropic share offering, a 10% chance of extinction may be a 90% win for your portfolio"

Hosts: Jeff Jarvis and Paris Martineau

Co-Host: Jason Hiner

Download or subscribe to Intelligent Machines at https://twit.tv/shows/intelligent-machines.

Join Club TWiT for Ad-Free Podcasts!
Support what you love and get ad-free audio and video feeds, a members-only Discord, and exclusive content. Join today: https://twit.tv/clubtwit

Sponsors:

More from Intelligent Machines (Audio)

All episodes