← Back to search

Alex Finn: Anthropic has the best model, Grok has the best harness | TWiAI 39

This Week in AI · 2026-11-12 · 70 min
relevance 45 11382 words Episode page ↗ Audio ↗
Show full episode description
Thanks to our partner, @Paypal! Pay zero processing fees on your first $100K in eligible PayPal payment volume. Learn more at http://paypal.launch.co/ Today’s show: Anthropic shipped Claude Fable 5.2. Panelist Alex Finn calls it a breakthrough on intelligence, speed, and cost at once, while noting that it’s still best to actually build things in Grok Bot, “by far the best harness on Planet Earth.” He actually asked Grok Bot, Claude Duet and ChatGPT Dots to send a $1,000 PayPal. Only Grok Bot got it done. TWiAI 39’s all star panel also includes Sara Hooker (co-founder & CEO of Adaption Labs) and Manu Sharma (co-founder & CEO of Labelbox). They’re discussing whether the model or harness matters more, and Sara observes that the entire Fable 5.2 roll-out is “almost entirely a story about cost.” Anthropic wants to own both the model and the harness so switching feels more painful. Manu explains why enterprises tend to remain locked in once IT wires up all their integrations… but Jason still feels his team drifting from Claude over to Grok Bot. So who wins in the end? PLUS: Anthropic sets its IPO range and drops new Claude office apps, Nvidia’s still investigating whether its CEO unethically sent extra chips to China, and are shadowy overseas interests behind the anti-data center push? The panel has some theories… Guests Sara Hooker on X: https://x.com/sarahookr Adaption Labs: https://adaptionlabs.ai/ Manu Sharma on X: https://x.com/manuaero Labelbox: https://labelbox.com/ Alex Finn on X: https://x.com/AlexFinn Henry Intelligent Machines: https://meethenry.ai/ Relevant Links Anthropic releases Claude Fable 5.2 and Claude Mythos 5.2 ahead of IPO (CNBC): https://www.cnbc.com/2026/11/09/anthropic-claude-fable-5-2-mythos-5-2.html Claude's launch post: https://x.com/claudeai/status/2119867114450268322 OpenAI launches GPT-6 Cyber with a higher API price than Anthropic and Google (VentureBeat): https://venturebeat.com/technology/openai-launches-gpt-6-cyber-with-stronger-exploit-research-and-higher-api-price-than-anthropic-and-google Grok Bot turns agents into persistent digital coworkers (VentureBeat): https://venturebeat.com/ai/spacexais-grok-bot-turns-agents-into-persistent-digital-coworkers-that-can-operate-your-apps-for-120-per-month/ Hermes Agent (Nous Research): https://github.com/NousResearch/hermes-agent Anthropic sets IPO price range of $70 to $78 a share (TechCrunch): https://techcrunch.com/2026/11/10/anthropic-sets-ipo-price-range-of-70-to-78-per-share-to-reach-an-implied-valuation-of-1-404-trillion/ Nvidia will investigate Jensen Huang's relationship with a possible Chinese spy (TechCrunch): https://techcrunch.com/2026/11/08/nvidia-says-its-investigating-ceo-jensen-huangs-relationship-with-possible-chinese-spy/ Report: Alice Huang helped move $7.5B in Nvidia chips into China (Tom's Hardware): https://www.tomshardware.com/tech-industry/policy/alice-huang-reportedly-instrumental-in-securing-over-usd7-5-billion-of-nvidia-ai-gpu-imports-during-three-year-relationship-with-jensen-huang-helped-move-240-000-banned-chips-into-china-mostly-h100-and-h200 An unexpected activist emerged to help block a $5 billion data center in Maryland (NPR): https://www.npr.org/2026/11/07/nx-s1-6007770/data-centers-activist-moratorium </
✨ Episode Outline — click any point to jump to it in the episode
Problem solved
Whether Anthropic's new Claude Fable 5.2 justifies its premium when GrokBot's harness delivers more day-to-day business value, and why Anthropic refuses to co-optimize with third-party harnesses like Hermes.
Benefits
  • Claude Fable 5.2 reportedly cheaper, faster and smarter than Opus 5.5 at once
  • GrokBot harness called the best for pure knowledge work
  • Smaller Sonnet/Haiku-class models reliably handle recurring daily tasks
  • Frontier models unlock long-horizon, multi-app and science/engineering work
  • GPT-6 Cyber offers security-tuned defenses for a hard cybersecurity stretch
Use cases
  • Alex Finn had Fable 5.2 build a complex video game for two weeks with no input, 'AAA level'
  • Jason's GrokBot manages his sleep schedule via reminders alongside his Whoop
  • Jason considering moving his 20-25 person team from Claude (~$20K/yr with overages) to GrokBot ($40/mo, ~$10K/yr)
  • Labelbox offloads recurring tasks to Sonnet/Haiku-class models, keeps frontier models for science and engineering
  • Fable 5.2 manipulates robotic arms out of the box on robotics benchmarks
KPIs / results
  • Fable 5.2 pricing: $8 per million input tokens, $40 per million output
  • GPT-6 Cyber stayed in scope 99.93% of the time, up from 94.49% for GPT-5.6
  • Anthropic filed for a $1.3 trillion IPO
  • GrokBot $40/month vs ~$20,000/year Claude spend for Jason's team
Tools / build
0:00 / 0:00
Hey everybody, welcome back, This Week in AI. We do this show every week to talk about AI. It probably should be today in AI because there is so much going on, Lon, that in the seven days between each episode, it's kind of like seven weeks worth of news. So we're gonna basically try to hit five, maybe six topics today with an all-star panel. Please, Lon. I mean, this is incredible. These are all builders building cool stuff in the world. Let's give them their shine. Shine, please. Thanks to our friends at PayPal, the exclusive sponsor of This Week in AI. Pay zero processing fees on your first hundred thousand ineligible PayPal payment volume. Find out more at paypal.launch.co. We got all three returning panelists this week. We've got all stars joining us. First up, she's the co-founder and CEO of Adaption Labs. They build AI systems that keep learning after deployment so companies get stronger results from smaller teams. Cheaper models. Sarah Hooker is here with us. Thanks, Sarah, for joining us. Up next, old friend of the show. He's been on twice this week at AI 16 and 23. He's the founder and CEO of Henry Intelligent Machines and creator of Creator Buddy. Give it up for our old pal, Alex Finn. Good to be here. Good to see everybody. You too, thanks. Have you been sleeping, Alex? Have you gotten any sleep? I know Fable dropped. I see your grok botting out. I mean, what are you sleeping? Two hours a night? I'm saying no more than five hours a night, the average sleep. My grok bot has been reminding me when to fall asleep and when to wake up, so it's been managing my sleep schedule. So a little bit better. Sleep bot is working with you and your whoop. Okay, great. Also from this week at AI episode 22, he's the co-founder and CEO of Labelbox. They're building the reinforcement learning environments and evals that Frontier Labs train their models in. Give it up for Manu Sherma. Manu, thanks so much for being here. Thank you for having me. All right. Lon, let's get to work. What's the top story here? Within five days, two leading AI labs both shipped new Frontier models while keeping their most dangerous cyber capabilities locked up for vetted users only. Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, two months after the 5.1 version. Anthropic says they're the same underlying model. Fable is available to everyone. It's got cyber and biology safeguards. Mythos goes out only to vetted cybersecurity and life sciences organizations. Last Thursday, also OpenAI released GPT-6 Cyber, their security-tuned model. They rate it critical for cybersecurity under their preparedness framework. That is the highest risk tier. Both labs are pointing to real bug hunts. And OpenAI's pitch reliability without safety prompts. GPT-6 Cyber stayed in scope 99.93% of the time, up from 94.49% for GPT-56. Fable 5.2 costs $8 million, $8 per million input tokens and $40 per million output. So lots of information about both of these. But I want to go to Alex. You called Fable 5.2 the greatest AI model ever just yesterday. What are you basing that on? And does it matter to you that you don't have access to Mythos in full? Well, I've been blessed enough to have been testing Fable 5.2 for a little bit now. I've been basically spending my entire days using it and building things and experimenting. And I was blown away from first prompt with it. I absolutely love it. But what really blows me away is Anthropic clearly has had a breakthrough over the last month. Every model they've released from Opus 5.5 to Sonnet 5.5 to even like Haiku and now Fable has been a massive jump over the previous version in basically every single way. From intelligence to speed to cost. Usually you can only have two of those three when you release a new model. That's the way every other company has worked. But Anthropic over the last month with these new models, each one has gotten cheaper, has gotten faster, and has gotten significantly more intelligent. There's really no weaknesses to these models. It's incredible what it's able to do. I had it build a really complex video game for two weeks straight without me doing any sort of input whatsoever. And it came out like AAA level. So it's unbelievable. They're about to IPO and they've clearly had some major breakthrough with AI that no other company in the space has been able to figure out. You agree, Sarah? Is this as good as Alex is making it up to me? Yeah, I mean- Better, cheaper, faster? That was a rave review. I think it's a bit of a simpler story. So you don't create a new technique in a month. This is almost entirely a story about cost. Anthropic has massive pressure. The last few months have really told the story of 70% of all traffic moving to open rates. They have to rationalize costs to a large degree. I think GPT Cyber is different. There's almost two separate stories with this release. Anthropic, it's about cost pressures and essentially having to justify the premium, especially when, Alex, I think your video use case is great, but a lot of companies and everyday users basically want the right model for what they're doing and may not be swayed to move to 5.2 unless they see it making sense for not token maxing and driving up the API bills. The GPT release is different, but also super interesting. It's really tapping into what is going to be a really horrible few months for cybersecurity. And essentially, every company out there right now wants to know, what can I do? And these are almost being treated as like a conversation about, how do I get the latest defenses or how do I have more control over my AI and automate bringing it back under control so I'm not vulnerable to some of these dependencies? Both are very interesting stories and super relevant. Manu, what do you got? You think this is a function, Sarah is saying, of the IPO and the pressures of open source and business users? Because I popped the first thing I did when I saw this is I gave it some jobs I was giving to GrokBot and it didn't do as good a job as GrokBot. And I was like, wait a second. I don't get it. Why are my jobs not doing a better job? And I'm guessing that's the harness of GrokBot. I guess it's the connectivity I have to other services. But I'm, as a business user, thinking about kicking out Claude and just putting the whole team on GrokBot. This week, I said to everybody, GrokBot is but $40 a month, $480 a year, $500. That's $10,000 a year. I'm paying a similar, I'm probably paying $20,000 a year to Claude with the overages. And I'm like, my people are getting more out of GrokBot. So I think I'm kicking this thing out. I'm going to kick Claude out. I think anthropic models have been phenomenal for almost a year and they've been added. I think there is not necessarily one novelty or some new kind of a technique per se, but it's a culmination of massive amount of investments that I've gone through in scaling reinforcement learning across the board. What I see is like when incredible tier one lab that is building a model, they're not just scaling one technique. They are actually scaling in multiple access. So there is, of course, reinforcement learning where models are learning to just be more capable. If you look at 5.2, it has a step change difference in its capabilities in multimodal and scientific knowledge. I don't know if you guys looked at one of the metrics where Fable 5.2 is performing incredibly well for robotic tasks. So out of the box, it can manipulate the robotic arms to do things that a ton of startups have been focused on achieving the similar results that are only focused on robotics. And so I think the team cooked across those dimensions. Not just reinforcement learning is one axis. There is another axis, which is around how are these models used day to day across their applications. And so the preference and the tuning that happens in making these models likable and kind of, you know, from all these axes, like haste and behavior and things like that. So some of these best companies are scaling across all those axes. And that's, I think, the result we are seeing in the market where some of these best models are, you know, orders of magnitude better in using longer and longer horizon tasks across day to day things. So, Manu, is that why I, as a venture capitalist and podcast and event producer, am not seeing the value in this release and previous releases of my entire team? It's like, let's downgrade to SANA. Let's use this. Let's use the last model to save money. And everybody in the company is just drifting to GrokBot because it seems to solve the problems faster. And maybe these models are maybe helping robotics, video games, you know, things that maybe the average business user doesn't get leverage from. Well, I think we're seeing that trend across my company also where a ton of things that we're doing every day to day recurring tasks are now we're offloading to, let's say, smaller models. Kind of like, think of like Sonnet or Haiku class models because they are doing it reliably. But at the same time, there are emergence of, again, reimagining many of the things we do in our company that only the frontier models can do because they are longer horizon. They're doing multi-hops across multiple applications and so forth. Or, you know, in our case, like we're doing a lot of science and engineering work in advancing our products or technology and things like that. For those cases, you do need the frontier models and they have this massive unlock that you see. But you're right, like an everyday business user is probably getting more value from kind of these smaller, faster, lower latency models, as you said, J.Kow. I love Grok model. It is sort of a workforce model for a lot of things I do every day. And so is like Sonnet 5.5. I think also, Jason, I think you're doing a little bit of comparing apples to oranges as well. Yes. Sonnet or Opus and Fable, those are the models. GrokBot is the harness. GrokBot is by far the best harness on planet Earth for just getting pure knowledge work done. Nothing comes close to GrokBot. Why? Why is that the case? Why would I thought Anthropic was the harness? What you're going to see increasingly is two things. One is that the difference between the harness and the model matter less than like, are you optimizing both together? Anthropic has made the decision not to play with outside third-party harnesses. They're not co-optimizing for it. This was the whole Hermes kind of separation. It's unlikely GrokBot is- Explain the Hermes separation for everybody who does it. No, Hermes obviously is another harness. It's used by geeks. Alex fell in love with it for 17 days and he just tossed her to the curb just like he tossed Open Claw to the curb. He's brutal that way. He just goes to whoever the new pretty girl or guy is. Literally up on my phone as we speak. Yeah. That's not your main. That's your side. That's your side harness. That's not your primer. It is my side piece, right? It is my Chinese spy, right? It's what you do when you're 2,000 miles from home, but not what you do onshore. We understand. Sarah, this is the truth, isn't it? This is where we're at. Explain the Hermes thing. Having heard what Alex has been saying, I suspect it's the truth. Here's the thing. Harness really matters as long as it's co-optimized with the model. Anthropic has made a series of decisions which basically shows what their roadmap is. They believe that they should own both the harness and the model, and it should feel sticky if you try and switch it or layer it, right? It should feel a bit painful. What you're describing of like and not playing well is probably the universal truth of a decision they've made. They want to lock you in a little bit. I think the question of if that's a good strategic decision is playing out real time, like all the shift towards open weights. The realization that people can customize much more rapidly, especially with automated self-improvement, has made it way less sticky to feel like you can only depend on a handful of APIs. And so that narrative is going to play out over the next year. But yes, for Hermes, they purposefully not integrated. I would be surprised if GrokBot is seriously integrating in any way with anthropic models, mainly because they've made it purposefully difficult to do so. So, Manu, what's your take here on the economics of how the industry is playing out? Obviously, we're going to talk next about the IPO. Anthropic has filed for a $1.3 trillion IPO. If you could tee up just some bullet points there, Lon. The reality here is they're making at Anthropic, obviously, they're making super economic decisions as they move from a private company to a public company. And is this ultimately the long-term mistake? And does it open them up for attacks from the side where business users like me, who only have 20, 25 people on the payroll, we're quite promiscuous. We'll move models. I mean, we're not as promiscuous as Alex, but we're not having like multiple, we're not in a polycube with our models and harnesses like Alex is. But we'll swap and go right to Grok if it's better and solves our problems. Well, I think Anthropic remains in an extremely strong position for a foreseeable future. I think there is a distinction between a smaller team that is able to hook up a lot of integrations in the applications and so forth, you know, with their personal credentials and so forth, versus a larger business and a larger enterprise where you might not be able to do that. Like you might not be able to hook up GitHub integration and a whole bunch of other applications. Because the IT department, the CTO have made this edict. Hey, security is paramount for us. We're not letting people freelance. Yes. Priority among most businesses today. And so I see incredible lock-in when you connect, let's say, Quad Cowork with all the integrations that is provisioned by the IT team in the large enterprise, you know, and you're using it more and more. It has more memory and context and things like that, that, you know, it's just darn that much difficult to make that hop over to the next application. And I'm using the word application because people are not necessarily choosing a model necessarily in this knowledge work context. They're actually picking an application like a Codex or, you know, GrokBot or Cloud app. And that actually, I see, I'm seeing tremendous lock-in. Once you hook up all these integrations that they are managed by some other team, you don't want to, you know, rock the boat and get something else installed and things like that. Yeah, this is why big companies are going to lose, Manu, because what they really should be taking at this stage of the game is an approach that each department can try two or three products. They're responsible for their security. Put a forward-deployed engineer in there and let a thousand flowers bloom and move back and forth. I have people making code with Lovable. I have people using Cloud Code. I have people on Grok. I got people on perplexity computer. And I'm just like, you know what? It's a small amount of risk right now. And I'm willing to take a modest amount of risk or have a little bit of a bake-off, Lon, to do this. Let's do two things real quick. Sure. Tee up the S1, please. And then you may have seen my tweet today earlier, Lon, about the Office Suite and how they just launched three more products into Cloud talking about stickiness. And we'll bring that up in a second. But let's just give us the quick S1 update. And then, Alex, we'll go to you. Okay. Let's do the S1 update first. The traffic set a price range of $70 to $78 a share for its IPO in an updated filing. On Tuesday, they're offering 825 million shares. At the top of the range, it would raise up to $64.35 billion and be worth up to $1.405 trillion on a fully diluted basis, up from the $965 billion valuation from the May funding round. CNBC reports citing, you know, people familiar, that Anthropoc plans to price on November 18th and start trading on NASDAQ November 19th under the ticker A&T. About 20 of the shares will go to retail investors versus the usual 5% to 10%. Separately, AMD, Broadcom, Google, Micron, Microsoft, NVIDIA, and TSMC have agreed to buy $26 billion of stock in a private placement at the IPO price. Biggest tech IPO since SpaceX raised a record $85.7 billion about five months ago. Jason, there's your... So all of their partners, all these major partners like Microsoft, are agreeing to buy more of the company, the IPO, and prop up the IPO price and build their war chest. They have over $100 billion in cash, I think, and they're getting close to $200 billion at the IPO. That buys a lot of data centers, Alex, and we're going to talk about data centers in a moment. Yeah, it's... Anthropoc is a great company. The products they make are incredible. My gut says, hey, I want to invest this. The issue is, is I just can't envision how does this go from $1 trillion to $10 trillion? I think intelligence is obviously always going to be in demand. There's always going to be more of it. But how do they build such an incredible strong moat where they can 10x from here? When I look at SpaceX, I can see a thousand different ways they build moats that no other companies can attack, right? They're the only ones that can bring infrastructure to space. They're the only ones building this incredible satellite internet network that can give the entire world internet. Anthropoc, they have the best models right now. But what happens tomorrow if ChadGBT has a better model? What happens the next day if Grok becomes a better model? And all of a sudden, their entire moat just goes poof up in smokes, right? To me, it feels like the only way these frontier labs whose only product is intelligence, it feels like the only way they could 10x from here is if they focus less on the consumer side, which it feels like open source is just going to eat alive constantly and try to go just for the grand slam breakthroughs, right? solve cancer, solve diseases, solve abundance, and just go for these grand slam plays. Because I don't know how, you know, when- It makes sense, Alex, that, yeah. And Sarah, like the, in order to move the needle on 80 billion in revenue, which is their annualized revenue run rate, AR, not ARR, annual recurring revenue, but pirate speak, ARR, annual reoccurring run rate, which is take the last month times it by 12, take the last quarter times it by four, which people are very confused about these two numbers. I keep having to educate people. But Sarah, you know, one thesis I have is I think Claude is trying to commodify the entire SaaS stack and make it free. They have Claude code, they have Claude dashboards, they have Claude design. And then today I saw they took out, they put out Claude whiteboards, Claude sheets, which is a shot across the Microsoft Excel bow and Claude data, which I guess maybe is going after Databricks or, you know, some of those large data plays. What's your take here, Sarah? What is the business of Anthropik going to be? And are you long the stock? I think Anthropik is going to do very well on the IPO. I think it's going to be super successful. Actually, I think it's a remarkable example of the impact and reach of AI. Probably the more interesting question is what happens in the year after the IPO. I think so. Anthropik can't compete on tokens. That game is just too dangerous. There's too many open rates models that are viable for many companies out there. It has to go up vertical. Alex is kind of hinting at this, but I suspect we already see signs of that, right? The Accenture partnership, the move into legal, tax, healthcare, finance. These are kind of the juicy industries. It means that timing is super important because it currently serves those enterprises. What you're talking about is almost what are the tools that you need to service many of those companies. But I suspect more clearly over the next year, it actually has to maybe compete directly in some of those verticals. And I think that's reflected in the anxiety of a lot of these companies to preserve the edge and start to do risk some of their dependency. We see that legal and probably tax the most at the forefront of that, feeling that pain point. But it's going to be a very common conversation just because, you know, to get at those figures, you really do have to be not just an enterprise customer, but you almost have to move up the stack much more aggressively. Exactly. Manu, what's your—OK, Lon, you can come in here too, yeah. I was just saying one thing that I thought was interesting from all these anthropic stories over the last few days. They now say 90% of the training in code written for Claude is written by Claude. So I think, does that compound over time for the Frontier Labs? They're the only ones that have access to their state-of-the-art, this is the very best model on the planet right now before it's released. How much of an advantage does that give them that they can use that internally before anyone else gets their grubby little hands on it? Well, it's definitely a strong advantage. I mean, I think like at a macro view, if you look at the business, anthropic is raking in about, I think, over 80% gross margins. By the way, this is a phenomenal metric. I mean, I think just about a year ago, we thought that it was just not possible for a model company. And so the way I think of anthropic is that it's a pure frontier AI lab company. Like they are—they don't have a cap expense. They are buying compute from other companies. They're primarily focused on building these amazing models and applications above them. And so for investors who are looking at making a bet on just that pure play intelligence game, I think Anthropic is the company for it. Now, of course, Lon, as you mentioned, where that high gross margins hasn't just come up, you know, come about just, you know, with like walking a park. They are using models and like anybody else, like any of the frontier AI labs, to make the compute more efficient. So I think I saw somewhere in their documents that they have much higher compute multiplier over the last year, meaning that same GPUs, they're able to serve more tokens more efficiently, and that ultimately helps the economics of the business. I think that if Anthropic were to reduce its prices by 2x, the usage would go more than 2 to 3x easily. I think more and more people will just default using their models if they continue to reduce their prices. Now, that's not a play they have to do it right now. It's a play that they can do at a later time if and when they see much more intensive pricing competition from similar models, similar class of models. One of the other things I've noticed is, I'm actually curious if you guys also have noticed, open source in the last four months or so hasn't really advanced as much. Like I am not seeing kind of a similar jump in capabilities with open source-based models as fast as we are seeing in Gemini, like in a handful of these companies. And I wonder if that is actually a trend or it's such a small blip for now. Well, this definitely speaks to the situation with China because the top models are coming out of China. The top models require the best hardware. And apparently, China got caught with their hand in the cookie jar. They got their hand in the cookie jar. And apparently, a spy who loved me, or loved Jensen, was diverting top chips and top servers from Singapore, Lon, to China. Which might explain why DeepSeek and these ones, Alibaba and everybody else, is doing so well. In a memo to employees, NVIDIA CEO Jensen Wang admitted to an affair with a former executive at Megaspeed that is a Singapore-based NVIDIA customer. And the U.S. government suspects that this woman, Alice Wang, no relation to Jensen, was recruited by a Chinese spy agency. The latest that we're hearing today, I'll add, the board of NVIDIA says it has no reason to believe that either national security or NVIDIA's secrets were compromised. An independent investigation with outside counsel is underway. What's new today, the information report citing two people that know her, that the woman, Alice Wang, used her NVIDIA connection to help get at least 7.5 billion in NVIDIA chips. That would be roughly 240,000 chips. Two Chinese companies, Reuters, looking into this. So far, they can't independently verify the information's numbers. So that seems to be a little bit more of the context here. You know, Sarah, we'll go to you. You know, I mean, what do you think this means for the industry? What do you think this means for China? Have they been relying on shortcuts to get around these export controls? And does that explain the power of some of these Chinese models, the Kimi K3s and beyond that we've been seeing? So it's interesting. I think probably most relevant for this story is why is it playing out with this magnitude? And I think it speaks much more to how we are thinking about compute as sovereign and critical power, right? So the fact that this story is dominating, I think Jensen independently, given the board stance and his myths within his own company, that's a separate question. I think he's likely to survive this. I think that it's more about, well, what do the dynamics, why is this the major story of export controls? Export controls have been notoriously hard to enforce. So this spy component has brought it to the headline that everyone sees. But frankly, export controls haven't really worked that well, independent of this issue. What are the main issues at play? One is Malaysia, Vietnam, every adjacent country to China has built up data centers. And so even if you can't import GPUs into China, there are many adjacent data centers that when you go and visit that region, it's pretty clear support many Chinese labs and factories. Then there's the dynamics within China. So yes, there has clearly been some flow of GPU machines that has just been really tough to regulate. In parallel, however, China is pretty adamant about weaning itself off GPU dependency. So they're pretty aggressively supporting their own in-house chip development. And what the Chinese labs are particularly well recognized for is efficiency. So DeepSeq is probably the best marker of this. So you have this dual issue. One, export controls haven't been that good. But two, China recognizes that it doesn't like this dependency, and it's trying as quickly as possible to ramp up alternatives. So the U.S. has to ask, if chip controls are not working, independent of the spy issue, what do you really do to limit access? And I think that's a valid question because this has been an unsuccessful policy. I'm not buying this story, by the way. I think there are more questions about this situation that I'm asking right now. Why did this news came just around midterm and after midterm? And why is it such a big deal where apparently in the news leaks, I've seen somewhere where our deep state or intelligence agencies knew about this for over a year. Don't you think that Mr. President would have been briefed about this a year ago or sometime leading to it? He was. And do you think if it was actually a serious concern that Mr. President would just call, hey, bro, I know about this situation. You've got to really put it under control. This is a serious matter for us. So I think there's something more strange about this situation currently right now. I think we have to wait it out. But we've got to ask some kind of basic questions like why is it such a big deal now? So why now? Interesting question. Yeah. And then, you know, Alex. Why not six months ago where this could have not been a bigger problem? Well, six months ago, it could have made the data center trade, the AI trade and the midterms even more challenging. It doesn't matter. The Republicans, Alex, got just dragged. But Alex, let me ask you a really just cut and dry question. Should NVIDIA fire Jensen? No, I think that would be one of the stupidest things we can possibly do. I think there's several companies in this world. They're, you know, very, very limited where the CEO and founder is the company. They are the visionary. They are the ones driving everything. And if you cut off the head of that snake, then the entire snake dies. Elon with all his companies, I think without Elon, any of his companies immediately go to zero. And I think Jensen is one of those founders as well. I think there's others out there like Dario, but like, and this is no disrespect to anyone. You know, you could change. This is going to come off as extremely disrespectful, but I'm going to say anyway. You could change the Google CEO and Google's not going to zero, right? Like that's, it's very much in a business. You're talking about a difference between a founder and a hired CEO. Right. And that's a fair statement. That's, I think most people would agree. There's no dig to any hired CEOs, but the founders are unique. Right. And NVIDIA is one of those companies. So anything outside of Jensen purposely going out of his way to sabotage American security, right? If he had some, I'm not saying this will happen, but if he had some sort of deal with the spies, like, I want to take down America here, have all the secrets. Then yes, we should probably get rid of Jensen. But outside of that, it would be absurd to get rid of him. Uh, it would hurt America on a level we can't even fathom if NVIDIA went to zero. And so no Jensen, although it was obviously a very bad thing to do, he shouldn't have done that morally speaking. Um, he should not be removed. That would be crazy. That would hurt all Americans pretty dramatically. Yeah. I'm in strong agreement with you. I think we have to remember, this is like a really important moment in time. We have this country that is battling itself and there are dark forces, uh, the CCP, Russia, other nation states. They are all trying to trip us up. What is the best way to trip up American, uh, exceptionalism, get us to fight against each other, compromise individuals. And that's what they're doing. They want us to slow down. They want to get rid of this one or two year lead we have. China's working on a hundred year timeline, Lon. They're working on a 200 year timeline. For them, winning super intelligence, just like winning, uh, manufacturing and creating that middle class over the last 20 years and winning the globe when it comes to, uh, manufacturing, that was their priority, you know, 50 years ago. The next priority, uh, was, uh, winning energy and winning or their next priorities, winning energy. Thus their belt and road nuclear program where they're putting nuclear power plants everywhere, you know, from Africa and back, uh, crossing India, Pakistan and everywhere else. Uh, and then finally winning super intelligence. That is their massive priority. How do you win super intelligence? You'd knock out a chess piece and let's call it what it is. Jensen's the king. It's the number one company in America. They just, they're trying to take the king off the board. And then you add to this, I am going to guarantee and predict right now, we're going to do a lot of investigations and find out the anti-data center movement is backed by useful idiots who have been funded by dark forces like this, uh, Chinese spy. And a lot of it is going to wind up being traced back to Iran, Russia, and most of all China. They're trying to establish, which means Americans need to, and I'm just, I'll get off my Ted talk in, but 30 seconds. America needs to decide. America needs to decide. Do we want a communist country that controls their people down to the footstep? America needs to decide. Controls religion and puts Uyghurs in literal slave camps. Camps that people, uh, Jewish people have gone on the record and called concentration camps. When a Jewish person says what's happening to this other group of people is like a concentration camp. You must take that seriously. This is not the group of people who should lead humanity. And America has to make a decision. Are we all in as a group together going to win the race to super intelligence? Or are we going to fight like idiots because of another group of idiots are being funded under the table by these dark forces? And if you think I'm a conspiracy theorist, look at what happened to the NRA and look at what happened, uh, to right-wing podcasters. These two right-wing organizations were infiltrated by the Russians. The Russians put little seed funds out to compromise those useful idiots. That's their favorite thing to do. And then the Chinese are getting the most sophisticated non-idiots as well. This is a huge blaring alarm for America. Thanks for coming to my TED Talk. Can I, can we just game theory one thing out though? Like here's my, here's my question. Please, please poke holes in my theory. I don't know if it's a full hole, but I'm curious. Well, cause NVIDIA, I think has been the leading voice in America for, we should continue sending some chips to China. Like we shouldn't cut them off entirely. We should sell them this. Like Jensen's been very vocal and public about that, even aside from what was going on behind the scenes. Now that he's been discredited in this way, do you think that hurts that? Like you said this yourself the other day, like, is there a chance that we're going to now become much more hawkish about keeping chips away from China? Because the big defender of that policy has now fallen in this way? A hundred percent correct. The actual want of the majority of people in the Trump White House, according to what I heard on back channels and which you can find on the public internets, was a lot of those people wanted to send zero chips to China to suffocate them and let them go create their own chips. Then we said, well, wait, maybe we could win by having them use the last generation as a chip, use our lead. Obviously that didn't work because they were able to figure out how to get the chips. So I think it's going to restart the argument. I don't know what's better to sell them the old chips and make sure they don't create an NVIDIA competitor because apparently they're creating an NVIDIA competitor. It doesn't matter. They're going to do it anyway. Again, we're in a two and four year cycle. They're in a 100, 200 year cycle. Manu, what do you think here about the global race to super intelligence? I mean, being hyperbolic or- You don't have to call it super intelligence. You can call it- I like calling it super intelligence because it is distinctly different than- Yeah. No, no. I'm calling it super intelligence to make a distinction. Thank you. It's not just for the Trump thing. Super intelligence being where we're headed to, AGI, where we are now. So I'm actually not using it for Trump reasons. I'm using it. Thanks for your TDS. The reason we're doing it, I'm using that term Manu, is because we're not at super intelligence and we're at AGI. Yeah, it's very tough for me to accurately answer this one. But let me start with this. I don't think Chinese businesses or labs, if you will, have to have chips in China. They can just do an API call to a data center somewhere else. Adjacent countries, yeah. And so I think the whole premise that the chips blockage is somehow thwarting them, I think these guys are incredibly smart people. They have the highest number of researchers in AI in a single country is in China. So I think if it is their stated priority that they want to build the super intelligence, I think there are enormous ways that doesn't require chips to be in China mainland to still achieve those goals. I think the question really then comes down to is what kind of viable businesses they can actually go create to go fund access to the chips on demand or data centers and so forth. And I think that's really fundamentally where my macro question is, where most of the revenues that are being generated by super intelligence systems are in, I would say, Western world. Knowledge work, knowledge work, S&P 2000s, like these enterprises and so forth, and their applications being built. When you go into China, you don't really have this rich ecosystem of enterprises and business applications and things like that, which is why historically America and then maybe Israel is the second to America has been the powerhouse of making amazing enterprise software and applications and things like that. So anyway, I think, you know, if I go with this chain of thought, I don't think they need chips in the mainland. I think it may actually, yes, it helps more if the chips are in mainland for wide range of businesses, but the most competent businesses in China are probably getting by just all right. And when it comes to kind of being selling chips to China and so forth, I go back and forth, right? I think I see good arguments for thwarting access to chips in mainland, but I also see the argument, which is, you know, American businesses want to do business. They want to compete globally and so forth. And why not let the market forces decide rather than geopolitics decide the market opportunities for businesses? And so, yeah. So it sounds like you're a pro letting the free market work it out, even though China is not a free market and they pick their winners. That would be the counter to your position, right? They pick winners, they price dump, they're right now in the process with BYD and their car companies, Xiaomi, of price dumping. Illegally selling product under the cost of making it in order to kill competitors. They're not a fair competitor. Well, that's right. But I think like, look at Elon. I think he's got the best selling EV model Y in China, which is kind of the Olympics of the sports, if you will. And perhaps your product has to be just so much better that in order if you can win China, you can ruin every other market and so forth. You're totally right, Jason. So I think... It's a complex issue, I think. It's a very complex issue, right? The most complex issue, right? There are precedents for like a rocket and missile technology, like ITAR restrictions that happens in most national security technology, which totally makes sense. That's not just for China. That's for basically every other country that the government has deemed to not be an ally. So you could apply that lens onto the chips. I think I buy that argument. But I also buy the argument that, you know, this technology is different. This is a ethereal technology. Like the diffusion of AI happens so much faster than any technology in the history of mankind. And because it's computer science at the end of the day, every country that has sufficient kind of technology advantage will get to it. Sarah, I want to go to you. Breaking news. Breaking news. What happened? Model Y is no longer the best-selling car in China. It's number two. Number one is the Gili Zhang Wan EX2. And there you see it there. The front looks exactly like a Model 3. They literally chopped off the Model 3 front. I mean, fuchsia. Care to guess the price, Alex, on this? I hope you can't see it. Well, I saw it and I can't believe it. $9,000. Basically, the wheels on the Model... Wheels on Alex's Model X are more expensive. But Alex is a baller. Let's call it what it is. Wait, that's from May 30th. I thought like Elon just did an interview where he said about Model Y being the best-selling. Okay, let me translate this because I have the source. Top 10 retail sales October 26th. All right, there you go. That's in the world? Top 10 sales January to October. Yeah, there it is. Model Y. I haven't heard of any of these cars other than the Model Y. You probably have heard of BYD for sure. Yeah, but I've never heard of a Xing Wan. No, you wouldn't. They have so many cars there. It's like literally going to Akihabara and looking at headphones. And there's a thousand pairs of headphones. And it's like all the same maker. Sarah, what's your take here geopolitically on the situation? I think we can stay out of the personal issues here. But obviously, we want to keep Jensen in the seat. I think you would agree. And what is your thought on this competition? And am I being hyperbolic about the race to superintelligence? Or is this an acute issue for humanity? You're not being hyperbolic about the race for superintelligence. But there's two separate issues. What Mana is putting out is that even if we wanted to do export controls, it's failed as a policy. And it's not because of a rogue spy. It's because fundamentally, Chinese companies are using data centers in Malaysia, Vietnam, name a country in Southeast Asia. And they're probably using that compute and circumventing controls. There's a separate issue, which you're pointing out, which is we're also kind of hobbling ourselves in the US by having things like absurd conversations about data centers that aren't really rooted in logic or any reason about having precision about the topic that's very emotional. And I think that is a separate topic and certainly one in which we need to just bring just more serious as to the conversation because it's getting absurd. So data centers are traditionally like one of the most boring utilities. No windows. Really nothing much goes on. There's no pollutants, really. It's not like oil. It's not like a manufacturing where you have like massive amounts of refuse that you have to deal with at the end of the day. This is core infrastructure similar to utilities. But it's become, and I think this is more about how people in the US are feeling about AI. It's become just this symbol of maybe anxiety and how people feel about technology in general. And if you want to talk about what will actually defeat us, it's not export controls. It's the fact that so much of the general public feels like this is actually a symbol that is relevant about how they feel about AI. And I think that's what has to be tackled because with a general public that feels adverse or best and different to progress in AI, it will hurt in all these ways. And data centers are just a symbol of it. But frankly, it impacts everything. Yeah. Now, there's a really interesting piece of legislation, Alex, that people are not tracking. But it turns out that Chip Security Act, HR 3447, bipartisan, and the provisions in HR 344 include location verification mandate, gives the Department of Commerce 180 days after the enactment to require hardware, firmware, or software-enabled location verification mechanisms on covered integrated circuit products prior to export, re-export, or in-country transfer. In other words, we're going to track these GPUs. And this is going to sail right through after this situation. This is going to be built in, Alex. And it's fait accompli now after a Chinese spy compromised the largest company in the world to whatever extent she did, allegedly. Disclaimer, disclaimer, disclaimer. Alex, what are your thoughts here on the... And to be clear, no back doors here. It's not like you can turn off these GPUs. It's not like you can see what they're doing. But they are going to build location into them, which means you'd have to disable it. And if you did divert it, as this person allegedly did with this company, it would become immediately known under this circumstance because there would be triangulation, IP, and perhaps even GPS in these things, although you could block that with a roof. Your thoughts, Alex? Well, I want to make two separate points, if you don't mind. One on this one, just dovetailing on what Sarah said. Number one, I think these chip export controls are much like the war on drugs. It's just not going to work. You're going to fight as hard as you can. And it's noble. People shouldn't be doing drugs. Drugs are bad. China getting chips is bad. That's a bad thing to happen. The issue is, is if people want enough, druggies' lives being dependent on drugs, China's life being dependent on chips, they're just going to find a way to get it, no matter how hard they try. No matter how hard we try, no matter how many controls we put in place, they want it more. They will be more aggressive than we will be when it comes to stopping it. And so I think it's kind of pointless. And they're going to find, not that it's not a noble fight to fight. It is a noble fight to fight. They're still going to figure out a way to get it. Number two on the data center thing, I think that we are in a psyop war at the moment with China and many other countries. And I think we are losing horribly in a lot of different ways. I think you saw during the midterms, a lot of candidates that ran on, I'm stopping all data centers. They ran on that. They won when there's no real good faith reason to run on stopping data centers. They're a totem. They're a symbol of wealth disparity, Alex. That's the reason here. It's like, if you're pissed off at the Trump administration, if you're pissed off at inflation, you're pissed off at the war, and you're pissed off at rich people who are at the White House every three days, apparently, being live streamed like it's a poker game. It's like, this is the revenge you can get. Stop a data center. Speaking of stopping data centers, Lon. As we always are. As we always are. On this program. It's a bizarre story of a mall that has been raised and closed for over 20 years. The land over mall. Yes. An activist who is stopping it because of water concerns, which is a resolved issue. They use less water than any golf course. And this blight here, this disgusting parking lot, where a mall used to be, is in a downtrodden area where they could use the tax revenue. And this NPR story. And this NPR story, when I talked earlier about dark forces at work, this to me rang a bell. I know I'm turning into conspiracy Calacanis all of a sudden. But Lon, cue up how this story was framed by NPR. So a $5 billion data center campus. It was planned for the site on the old Landover Mall in Prince George's County, Maryland. It's on hold. The mall closed in 2002 and was demolished. Left around 90 acres of empty land. County planners approved a data center in 2024. Set to break ground later this year or early 2027. That woman you see pictured, she's local resident Taylor Frazier McCollum. She started a petition that got over 25,000 signatures, leading to protests, multiple local moratoriums on data centers. The article that you're describing described, I mean, she's an activist. And they wrote about how she's a civil rights activist. The article quoted her family members comparing her to some of the great, you know, the Medgar Evers, the Martin Luther Kings, some of the heroes of the civil rights movement. But now when the project is paused, NPR says some of the worry about declining tax revenue. Interesting. I want to bring in one more perspective, Jason, before we take your thoughts on this. Sam Lyman of the Bitcoin Policy Institute. Now, he's saying that a China-linked socialist party helped to organize this fight and many others like it. Ding, ding, ding. NPR story does not actually mention this at all. But Lyman connects it to the Party for Socialism and Liberation, which is a Chinese-aligned organization. It's run by Shanghai-based Neville Roy Singham, who has bankrolled a lot of pro-CCP, anti-American nonprofits. And he's a big voice in the anti-data center movement. So you're not the only one having some conspiracy theories. And I guess unpack this for us, what your conspiracy theory really is here. You know, listen, if you start splashing money around to the civil rights movement, to the NRA, the guns right and move, any of these organizations, you can get them so easily. They're underfunded, right? Sure. They take donations. So all you have to do is start sending donations from LLCs, front corporations. They probably don't even know that they're being manipulated. There was a Russian, Russia, Russia, Russia, who infiltrated the leadership, again, through certain tactics that these foreign adversaries are willing to do and compromise the NRA itself with Russian dollars. Why? In order to ferment the gun issue affecting politics and maybe getting a certain political candidate into office who was the preferred one. So what we're going to find out, I'm just putting a prediction here on the show, Sarah. I think what we're going to find out in a couple of months or into next year is that much of the data center movement is covert money being sent to people who, you know, through LLCs, who I don't think know they're working for the Russians or the Chinese or any other adversary, and that they're actually what the CIA would call useful idiots. They're being manipulated. We have one more example here before we, Sarah, definitely want to hear what your thoughts are here. One more example. This was also tweeted out from Bobby Mars. Apparently a different data center in Western New York was also brought down by a local activist and a lot of the conspiracy folks online are suggesting this is likely also funded by similar groups. So we're seeing this. It's not just in this one Maryland County. We're apparently seeing this in several locations. And Sarah, they're also paying tons of influencers on X and Instagram to suddenly be anti data center. They're getting paid to post. Tell Lorenz, we'll pull up and find a tweet of it. Tell Lorenz has been reporting on how she's been asked to be anti data center in her thing. So what's your thought here? Is Conspiracy Calacanis losing it? Do I have? Yeah, I mean, I don't know. Do I have all in psychosis? Do I have a high psychosis? I'm willing to have it. Yeah, I don't know much about the Bitcoin Research Institute, which I think law and you quoted first. But I think there's two things, right? I think the amount of money influx on it doesn't matter as much as the fact that it's just ripped through public opinion. And you've got to ask, like, why was it so, you know, why was it such a receptive audience? I think you spoke to that. I think people see this as a symbol of how they feel about AI. But frankly, there's another component, which is worth talking about. If you think about American exceptionalism, data centers are very handy as a focal point for anti-AI because it is very easy to block infrastructure being built in the US. Like, frankly, one of the easiest things you can do is derail a train line being built or derail another road that is needed for a community. And that's because a few citizens can have exceptionate influence on the process. So if we talk about where this goes back to, like, what are the core advantages that America needs to build up? One of them almost certainly is how do you make sure that core infrastructure that is needed is able to be executed on when it's so easy for citizens in a handful of places to exert this amount of influence. And I think that what is valid about this conversation is that often this influence is exerted with untrue claims. So discussions about the impact, you know, data centers, if you think about it relative to other like, you know, energy, energy utility or infrastructure for oil or gas, data centers are probably as boring as it gets. You know, you have very predictable inputs, outputs. Water is loomed large in the public imagination, but really is not at all different from a growth course, if that. And then you have very little actual output, especially compared to agriculture of energy. So I think these are the things that have really derailed the conversation. And I think there's a valid conversation there about why is it being, how does America deal with needing to prioritize this infrastructure? I don't know if you have thoughts there, but it says more about how easy it is to derail as well core infrastructure projects. Yeah. I mean, I think it's clearly the feeds have taken over ration, rationale and logic among a lot of these groups in America. Right. So I think, you know, I don't think there's a one single issue like water or something else. I think it's purely a vibe that they feel when they are going through the feed, where the algorithms are feeding very different stories about anti-data center. I think, look, our industry has not gotten the message right. And I think we, our industry, the data centers, the hyperscalers who are building these things, they need to do a better job in communicating with actual metrics on how good these things are for their local communities. Now, it's not just, it just does not stop at the, with the private companies. I think the local government and the state government has an important role to play. If they believe that these things are actually great for their local communities, then they have to campaign and communicate and meet where, meet their, you know, their citizens, where they are at in forms of communication. They prefer to understand the benefits of these things. And I do think, I do believe that there are amazing benefits, you know, bringing revenue and so things like that and so forth. But it has to be told in a way that average citizens understand. And in some ways, you know, shift the vibe. It's not a single issue. What would you do, Lon? If you were running a data center company, not a big jump, you use a lot of tokens. If you're running one of these and you go into this community and you say, hey, listen, the mall's gone. Let's say you have some amount of budget. What do you do? Do you bring the, the, the Greek, I'm not going to say Chinese menu. It's a little bit of rough week for the Chinese. I'm going to say the Greek diner menu. Because I can say that because I'm Greek. You bring the Greek diner menu unfolded and be like, hey, we got everything. Your 24 hour Greek diner. What's on the top of your list? How do you solve this problem? If you watch mobster movies, which is, I, so much of my life approach is, is based on what, what happens in crime movies. But they, they're, they're bringing things that the community doesn't want. You know, they're running numbers. They've got illegal stuff going on. It's a protection racket, whatever. But they win over the community because they do things for everybody. They become a pillar of the community. So the, the, the, the people in the neighborhood don't know how they could live without them. But so, did you ever see that, that, that famous clip after John Gotti died where there were guys in the street, like upset, like, like they were, they were celebrating his life. He was a great guy. What about the murder? What murder? What murder? That's the, that's the vibe. You got to give turkeys on Thanksgiving. That's the vibe. These data centers, they got to be Scrooge on, on Christmas morning. Exactly. Like just, just throw money around, make it. I, I, we were talking a few weeks ago about those meta ads that I still see constantly everywhere in Texas, like we're, we're rebuilding El Paso. And it's like, that's great. But like, I need specifics. It can't just be like a guy in a hard hat or like a guy brushing, you know, sweeping up. Like, tell me exactly, like build something great that everybody in El Paso is like, I couldn't live without the, the meta concert hall, the meta hospital, like whatever. Like as if you win people over, they will love data centers. You, you, you evoke John Gotti. As I always, he, it was when I, lived in Brooklyn. He was the guy. Sure. But he was, he was Queens. He did the most amazing fireworks, 4th of July block parties in Ozone Park. That was his thing. And by the way, Al Capone during the depression was running soup kitchens. Sure. Now I don't think, I'm not saying that data centers are criminals, but I think, Alex, you just give free electricity. You say anybody who lives in this community gets free electricity. Anybody who lives in this community, my big one, you, let me run this up the flagpole, Alex. So I see you don't like the free electricity. What if you said put, make the data center 25% smaller and put a campus for retraining for electricians, construction workers like Matt is doing. What if you just said, Hey, anybody in the community gets a free trade thing? Okay. Yeah. That's a bandaid. Let me give you, let me give you an example of how to fix this. My parents, I think are your average boomer in America, the average boomers. They have zero hobbies. They spend all day watching CNN. And if you say the word Trump, they have a seizure. Like that's their, that's their life. Right. It's pretty close. Similar to Alonium. Um, these, uh, typically I'd say 99% of people fall into this persona hate data centers. You ask them why they can't really tell you, Oh, they're loud in the water has cancer in it. Well, that doesn't really make any sense at all, but here's the thing. My parents love data centers. They love it. Why is that? Because Thanksgiving a year ago, uh, the door broke and they called up their, uh, carpenter guy or whatever, whoever fixes doors. And I said, okay, I'll be there. It's going to cost a thousand dollars. I take my dad's phone. I install Chad GPT. I take a picture of the door and I say, how do we fix this? And it fixed the door with things they already had in their house for free. Ever since then they call me, Oh my God, you won't believe what I did with Chad GPT. Oh my God, you won't believe what I did today with it. This is the most amazing thing ever. And they love AI. They are very pro data centers. So I don't think the responses to the fixes is just giving away free things. I think that's always just kind of how you get small temporary wins. I don't think it's how you get long-term wins. We need programs, specifically educational ones in this country that get people to use this unbelievable life-changing technology and improve their lives. If we're going to give away free things, give them free rock accounts, give them free Chad GPT accounts, and then incentivize them to use that technology to better their lives. And I think if you do that, they'll all of a sudden look at these data center debates and go, what the hell are these people talking about? Fireworks AI, they host, they fine-tune open source models for companies, including Uber, Shopify, DoorDash, the usual suspects. They said Monday that hackers access customer data sets and models over the past month. For Forbes, it started with unauthorized use of internal credentials in early October, followed by a separate attack October 24th on computers storing R&D data. Exposed data included billing and account information, payment history, and parts of unfinished customer AI models. Fireworks blames LLM jacking, where attackers steal access to AI models and resell it, leaving the owner the bill. CEO Lin Zhao said on X that attackers' behavior suggested AI agents, Jason, were involved. I think this is probably the next few months. We now are in the middle of a cyber. They were probably more vulnerable because Fireworks is an aggregator of compute, right? So they're sourcing a ton of compute from different places, which makes them more vulnerable to anticipating all the different things. The result, again, companies want more control and predictability, you know, especially when they're sending their data. They want to have probably much more self-improvement and fine-tuning that is within their control and premises. And so we'll see much more anchoring of that. But I suspect this is part of the larger trend. We've seen a bunch of cyber attacks. It is going to be rough for the next six months until cyber is an adversarial game. So right now, attackers are on, have an advantage, and defense is going to have to come up to speed. But on that note, this was super fun. Thank you, Sarah, for your time. And give a plug for Sarah's company, Lon. Oh, Sarah's company? Please, plugs. Sarah, you're welcome to give the plug yourself. Adaption Labs, adaptionlabs.ai. Throw it on the screen. Sarah, if you're hiring or you're looking for customers, please, please take this time to be selfish and talk about yourself. Please. You gave so much tonight. We make it much easier for people to own their own AI. So fine-tuning, probably most companies have nightmare memories of when they tried it a few years ago. We automate all the choices and we self-improve it based on your context. So it stays up to date with your business. The moment is super exciting because it's the moment that this is on the mind of every single company out there. So with that, I'll leave you be. Okay. This is fun. Thank you again. Lon, what's the domain name there? Just so we put it up there. Adaptionlabs.ai, folks. Yes. And you use the promo code J-CAL. Sarah will give you 10% off your first day. That's not part of it. I'm making it up. I'm making it up. All right. Let's give Manu a shout out. Sure. Thank you so much to Sarah. And thank you, Manu. Thank you, Manu. Labelbox.com, folks, if you want to go check out what Manu is doing. Manu, just explain one more time what you do and how people can engage you. Are you looking for a certain type of employee right now? Are you looking for a certain type of customer or partner? Yeah. So Labelbox has a couple of products, but I think one of the things I want to highlight today is Recursion. Recursion is an agent platform for larger enterprises or businesses that they want to automate recurring, highly technical, complex tasks in the background. And it's incredible the level of accuracy and performance you can get from a recursion-like platform that can automate, you know, underwriting or research. It can even train custom AI models for you and keep training them autonomously and so forth. So that's what I'm excited about to share today. And a ton of amazing big enterprises are using it today. We internally at Labelbox use Recursion across the board for all kinds of tasks. Awesome. Really appreciate you giving us so much of your time, Manu. Thanks for having me. And Alex, you are unbelievable. It's so great to have a builder like you on the show. You're invited back every week and then in between every day to talk about what you're building. But where can people follow what you're building? I know you have a YouTube. You got an X. You're everywhere. I got a YouTube and an X if you just search my name, Alex Finn. But also I am building meethenry.ai, Henry Intelligent Machines. It is an AI agent that will build you businesses. So it will learn about you, your skills, what you're good at, the unique challenges in this world you're positioned to solve and then help you solve those challenges. I dream of a world where you wake up and you have AI agents building you businesses and you do the things you love. You go out, enjoy the world, enjoy the sunshine, and you have value being created for you back at home. And that's Henry Intelligent Machines. So sign up for the wait list coming very, very, very soon. All right. Wow. So this is going to be a dedicated startup for you. Is that right? You're going to try to make a garden? I've already raised. I've raised. We're doing it. What about Uncle J-Cow? Can I get a slice? Let's talk. Let's do it. Let's not talk. Let's just send me the details and let me take a quick review. And let's do it. I want to get on that cap table. There you go. All right. Great job, everybody. Great job, Lon. And we'll see you next time. Bye-bye.