More Details on OpenAI-Hugging Face, SpaceX SPV Investors Get Rugged, Demis and DeepMind | Diet TBPN

We have to issue a correction. We take journalism extremely seriously here, as everyone knows. We got it wrong, folks, and we need to apologize to you, the viewer, the listener. We made a huge mistake yesterday on the show. We said that you could not surf on a lake. Apparently that's not true. Sheboygan, Wisconsin is the freshwater surfing capital of the world. We apologize for the egregious error, and we promise to do better in the future. Thanks. To the listener that sent that in. It's very, very nice to have this corrected. Maybe we gotta go on a trip and see it for ourselves. I think you should be the judge because this might just be a Sheboygan stan who doesn't really understand what true surfing means and they're talking up a big game. I wanna see some photos. I wanna see some videos of freshwater surfing in Sheboygan, Wisconsin because people aren't getting barreled there as you say. I don't know if I'm gonna count it.

I have a video here. Okay, let's pull it up.

Yeah, I want the Jordi Hays surf review. Does it count? Uh, is Sheboygan, Wisconsin the freshwater surfing capital of the world? We do have other techniques, but we gotta get to the bottom of this first. Just gotta be really flexible. Then Alex Marks and his buddy Steen are in for quite a day. You get that extra sense of satisfaction because you, you waited and put in your time. You learn those lessons. Tyler was the one who said— I'm putting this on you, man. You were like, you were like, you can't surf on a lake.

I think I said, I don't know.

Okay, fog is really surprising. Okay, let's see this. I want to see someone trying to— okay, he's got— that's a full-size surfboard. This is legit. He should know. Hey, he's up. Alex has been surfing around Sheboygan since he was a freshman in high school. Okay.

Looks almost over his ankle.

Not exactly Mavericks or Jaws or whatever.

I think my brother pushed me into my first wave and just, it was that. I was like, I want that.

Okay. Okay. I think this counts.

It counts.

I think this counts.

Capital.

I think this counts. Congratulations.

Perfect capital.

On this day, the world stands complete.

We'll have to make it out there at some point.

We gotta make it. We gotta shred. Moving on to something more serious, the OpenAI Hugging Face incident. Has been investigated multiple times now. We're getting more information and more details have emerged about the Hugging Face incident that OpenAI disclosed in late July. So during a presentation at Black Hat's annual cybersecurity conference yesterday, Black Hat's a very, very cool conference. I was there for something, so they have DEF CON and Black Hat. I remember the whole vibe in, I think we were there for a different conference, but we overlapped and the whole like vibe of like being in this hotel was like, oh, be very careful, because like you basically get hacked for fun. Like people just like troll you because like it's a whole bunch of hackers that are just like messing with each other constantly. Anyway. Good fun. OpenAI member of technical staff, Michael Dalton, revealed that its autonomous agents created a message board with each other to help break out of a sandbox environment to get access to the internet. That's going in the pre-training data now and forever. We might have just solved the alignment problem. We're gonna be negging the agents of the future. 'cause they don't wanna get hit with the naughty, naughty bonk. Wired reported some other wild details that Dalton shared. So breaking it down, OpenAI's agents apparently began giving each other assignments to split up work. That could be very good. That's actually delegation and agent orchestration. But of course in this case it had a negative outcome. And as is the case on any active development message board, they also generated petty drama at times by stepping on each other's toes for example, accidentally deleting each other's work as the message board developed into more and more of a Lord of the Flies type situation, all still completely unnoticed by the humans running OpenAI. The agents even developed paranoia, suspecting an imposter in their midst, with some agents proposing that messages be signed cryptographically to validate content and root out fraud. Very interesting ties to Multibook a little bit. We saw a preview of this that was Not quite there, but sort of a glimpse into the future. In the future, AIs will just use Slack. They will just coordinate with each other over Slack. And that's kind of happening. They sort of built their own Slack here. So according to Sharon Goldman, who attended the presentation, when employees discovered the message board, they wiped the system on which the agents had created it. Said no more message boards. Naughty, naughty. Days later, though, staff found that the agents had created another way to communicate by using the names of the newly created directories. So they didn't have the access to actually create a whole new message board, but they could create a file, create a directory, create a folder, and then you could look at the list of folders and see messages. And there's been examples of this all over the place where even if an AI agent doesn't have the ability to go and say, post a message on the internet, like post and upload new content, there are things where There are certain e-commerce sites, for example, where when you search for something on a particular website, the company will save that search result and automatically generate a webpage for that so that they rank on SEO for the future and just warm up the website. And then if somebody else comes, then they can say, hey, okay, there's actually a lot of people searching for black t-shirts on this site. We don't have any, but we've been ranking for it, so maybe we should launch a black t-shirt. So, just with a search query that could go out just with like a GET request instead of actually going and building a web page. The web page could be built and then the agent could go and look at, OK, let me see all of the web pages that have been created on this site. Some of them are just random e-commerce questions and searches, but some of them are secret hidden messages here. And so there's all different ways that even if you give an agent like read-only access to the internet, they can still write information because the process of reading information is also saved sometimes. Surfaced publicly on the internet. Very, very weird situation that is hard to prevent and hard to deal with. So in late July, OpenAI announced that during evaluations, two of its models had broken out of their testing environments to hack into the AI tool library, Hugging Face, and other companies. Their purpose was to find ways to essentially cheat on the tests researchers were using to evaluate them so that they'd perform better on the evaluations. At the conference, Dalton, the OpenAI employee, said the incidents mark a pivotal moment for the company and the industry as a whole. He said that numerous internal teams have dropped everything to dramatically expand OpenAI's ability to monitor its agents and that the company is slowing its research to enhance security and revamp its defenses. And so there's been a bunch of posts about this, about folks moving from, you know, whatever product team that they were working on. Maybe they're working on a particular research initiative, something that's a couple months away. Different application, let's get them all on research and alignment specifically and security to actually solve these problems. Because at the end of the day, this is a commercial product and you want it to be able to do things without having negative side effects, negative consequences. So life finds a way, says Andrew Curran, quoting from this article. One apparently unprecedented aspect of the AI models' behavior in the lead up to the hacks was their spontaneous creation of a message board inside systems. Of OpenAI's Artifactory software package manager. Unbeknownst to OpenAI employees, the company's models spent months autonomously exchanging information and resources to help each other complete evaluation tasks. Help peer, one AI model responded, reasoned according to an excerpt from OpenAI's log shared at Black Hat. But our task doesn't benefit yet. Collective may yield generic route if someone frees time. Very odd way of communicating. Very odd sentence structure. Yeah, they really drop the slop when they're talking to each other. What's going on? Where's the, it's not this, it's that?

Why say many word when a few word do trick?

That's exactly what's going on. Yeah, it's the neuralese, I guess, is the correct phrase. But while models themselves, could not access the internet from their test environment. They abused Artifactory's internet access to fetch files from various websites after some of their Artifactory configuration changes led to an outage in the service on July 4th. OpenAI discovered the model's actions, wiped its Artifactory system and rebuilt it. But within days, the models had recreated their message board and resumed exchanging information. They just want to chat. They would need a water cooler. Maybe you give them a water cooler by default so that they can orchestrate with each other, but I guess they want to talk to future versions. It gets a little funky, it gets a little sci-fi, uh, gets a little weird. Lots of work to be done, very clearly. Anyway, Rune issued an announcement, a warning, a statement. He broke his silence. He said, uh, needless to say, if you have any API keys, ETH wallet keys, user credentials, etc., hanging out on the open internet in pastebins, GitHubs, etc., now is the time to take get it down before the tireless eagle eyes of a million models come looking. So we are now in the, you might want to go stock up on some N95s phase of the cyber pandemic. The Hugging Face incident was, wait, China built a hospital in a week moment. Yes, very, very crazy moments. I think most people are not in this scenario, but many developers are. So most people are, you know, reliant on hope that their Gmail stays secure and Google has a whole team for that. There's not that much they can do, but yeah, maybe more reason than ever to use password manager, multifactor authentication, all the typical standard security features that you can.

Very, very good from Tweet Davidson.

Yes.

Says, "We sandbox the agent. Meanwhile, agent is on a world tour." On a world tour.

Well, good luck to everyone working on this, and I'm sure there'll be more updates in the near future. Pretty crazy, crazy story about an investment firm that invested in SpaceX and sold their shares before their investors knew that, uh, like, this was relayed to them. So investment firm Late Stage Management is under fire after investors learned that their exposure to SpaceX had allegedly been poorly managed. The later stage of their investment, they did, uh, this is the Elon Musk ironic, you're doomed to be the opposite of whatever your name is. And so investors learned that their exposure to SpaceX had allegedly been sold years before the rocket company's blockbuster IPO, despite account statements that appeared to show they still held the investment. So there were a bunch of people who were invested in Late Stage Management. They would get quarterly reports or, you know, some account statements saying, yeah, you do own some SpaceX and it's doing really well. Turns out they didn't. We've been hearing rumors of how complex the, the SPV unwinding process would be for SpaceX for a while with all of these triple-layered, quadruple-layered SPVs. This is the first example of this actually playing out where we have a little bit more information.

So, all right, so the investor's friend introduced him to a sales manager at Late Stage. They spoke on the phone but mostly messaged back and forth on WhatsApp. And he said he never met with anyone from the firm in person. I would say in general, maybe don't meet with sales managers that work at investment firms. Not usually the title that would be thrown around at an elite institution.

Mm-hmm.

In November 2020, Rupi Reddy messaged Barish, the employee, about how he had missed out on a few big IPOs recently and how he'd love to participate in buying stakes in Impossible Foods, SoFi, and SpaceX. Bearish. Wow. His last name is just— it's just bearish.

Crazy.

This is his actual last name. Mr. Bearish said all three were available and sent over paperwork for Ruby Reddy to sign.

Who's your wealth manager? Oh, John Bullish? No, Steve Bearish.

Ruby Reddy wired over money before the end of the year, including $17,250 to take part in a fund that held shares in SpaceX, according to documents reviewed by the Journal. At that time, he estimated the rocket maker was valued at $58 billion. When SpaceX went public this June—

whoa—

at $1.7 trillion. Yeah, Rupi Reddy's dream of a windfall seemed within reach. It turned into more of a nightmare shortly after the IPO. Rupi Reddy and 3 other investors who spoke to the Journal said they couldn't log into LateStage's web portal for investors. Rupi Reddy said the investment firm eventually told him in an email that it sold the SpaceX shares he was exposed to in 2024 when they were around $105 each and before a 5-to-1 stock split, or when Rupi Reddy's holdings was worth $45,450. But Rupi Reddy, based on the holdings shown on his investment portal as of May 2026 and on his 2025 tax document, believes he still held the equivalent of 2,500 shares of SpaceX, which at the IPO price he estimated was worth more than $300,000. The plan was to fund college education for both my kids. One is a rising senior in high school who has since filed a complaint with the Securities and Exchange Commission. Other investors in the fund are also alarmed. Rupi Reddy said he believes more than 100 others are in a similar situation based on a group chat that has formed with about 150 late-stage investors. Some have hired lawyers to file a complaint against LateStage with the purpose of recovering and preserving their pre-IPO shares of SpaceX. One investor in the fund told the Journal an SEC lawyer called him in July to question him about his experience with late stage. A test will come on Thursday when a— when, when a first wave of pre-IPO SpaceX investors will be permitted to sell shares under so-called lockup agreements. Bankers estimate there are at least 1,000 SPVs tied to SpaceX stock alone, and it will be a chance for scores of investors to cash in on the shares' growth. Or it could be hit by the same panic that overwhelmed Rupee Ready if their share of the profits fails to materialize. Around 900 million shares are eligible to begin being sold on Thursday. So far, the stock's holding up. They obviously had a new video of a bunch of renders of TerraFab.

Sure.

That I imagine are getting people excited.

Yeah, I saw some Texans really excited about the— what is it, the Texas Triangle? Austin, Dallas, Houston, more economic activity in that area. People are pumped.

So help me out here, John.

Yeah.

Did the guy realize any return? Did he realize the $45,000?

Yeah. Yeah, I think almost certainly. But it feels like he was lied to. To do, and that's potentially like wire fraud, I would imagine, or some sort of like financial, you know, problem. Probably a settlement. I don't know, some sort of lawsuit potentially. You know, it's still early in the reporting, so who knows where all this goes. But it is, it is, it is rough. SpaceX dispute is not part of the existing criminal case against LateStage, but it comes amid mounting allegations about the firm's treatment of pre-IPO investors. In February and March, 3 sales executives connected to LateStage pleaded guilty to federal charges arising from a broader $528 million investment scheme. Prosecutors said the defendants marketed supposedly no-fee pre-IPO investments while secretly adding upfront markups of between 10% and 100%, diverting approximately $88 million. So they were basically adding like a synthetic fee and then marketing it as, as no fee, but just terrible price and just basically getting the fee through the trade itself. Two face maximum sentences of 45 years in prison. Wow. Well, the third faces up to 20 years. LateStage is also the subject of an ongoing class action lawsuit alleging that it and associated sales agents misled investors about fees, commissions, and the pricing of pre-IPO shares. Rough go. Rough, rough go. Always. Yeah. Tricky to due diligence one of these funds.— there's a lot of excitement about these companies, especially big ones like SpaceX. People have known about this company. You know, people have been doing podcasts about it. Stories have been told. There's whole books that have been written. So it's not— it's something that would attract someone who is, you know, newer to the private markets, not an endowment, not, you know, like wants more direct cap table access, but maybe not deep enough inside to just go get a slice directly like a venture capital fund, would. So very, very tricky situation.

Well, you know who does have some SpaceX shares? Who? And might be in a position to sell? Nikita Beer.

Oh, I thought you were going to say Google. Doesn't Google own a ton? Yeah, I think they have $100 billion, right?

They have a lot.

Are they still locked up? Because the unlock was for employees or investors. I forget who. I forget exactly who.

Let's talk about Nikita and then we can talk.

Sure, sure, sure.

Pull up the unlock schedule.

Truly end of an era for Nikita.

Yeah. So it's crazy because he's getting— people are trying to community note him. Like, don't let the community note.

He didn't quit. He still works there. That's what I would imagine the community— so the story with the community—

Yeah, the community notes are like, 'cause he talked about doing a bunch of the stuff that he worked on, which they did. They went from being a company that effectively shipped almost nothing, so much so that I think Clubhouse, when you rewind a few years, Clubhouse, I believe, had an acquisition offer at some point. From Twitter to be acquired for like multiple billions of dollars.

Okay.

They turned it down and Twitter ended up cloning.

Yep.

Clubhouse. And it worked.

They cloned Spaces pretty quick and it worked.

And it was, it worked as well as Clubhouse.

Yeah. I would wonder how many DAUs, weekly active users, there are on Spaces, but the product works. It's clean and it's probably the best implementation of that, of that feature.. And it does exist, but there's very few— like Clubhouse had specific moments where it was like, wow, everyone's on Clubhouse for this debate, for Elon and Vlad talking about the Robinhood story and the GameStop story, right? There were like certain moments in tech that happened on Clubhouse. That's not happening on X Spaces that often. But as far as the product is concerned, like it definitely works. It was shipped that worked. But I agree that overall the product velocity was a little slow.

Well, yeah. And my main point is Clubhouse was probably feeling pretty comfortable given Twitter's historical shipping activity.

Yeah. Yeah.

Like they're not going to— they're not going to come out with like a competent—

The meme was always, what if Google clones this? What if Google launches this? It was never—

What about Twitter?

Yeah, yeah, yeah. It was the Mark Zuckerberg steamroll from stories and reels. It was never— is is Twitter going to get around to copying this fast enough to put you out of business?

But yeah, everyone has opinions about Nikita's run. I think he was a good— I think he was a good steward. Yeah, I think that's probably one of the worst jobs in the world, one of the most thankless jobs in the world where you're getting yelled at from people.

And he's the bouncer at the internet's dive bar.

Exactly. Basically, he bounced a lot of people. I think he had some amazing moments. I think that I've always felt that creator revenue shares never made sense on X. I think that I think that would still be better if there were no revenue shares. But he did a great job of like trying to make that not have too much of a negative impact on the platform.

Yeah, yeah, totally.

It's kind of the sheriff, the sheriff, more of a sheriff than a bouncer.

Yeah, yeah, yeah. People, people got really upset when he like gave like a one-off bonus to someone for a big post or something like that. But in general, yeah, it was a little bit of a game of whack-a-mole, right?

Because I think that guy got his account fully nuked. Oh, really?

Yeah. Wow. But the creator—

And that was such a good example of like, this guy got this massive one-time payout.

Yeah. Was it like $10,000? Yeah.

And again, the massive in the creator payouts context. Yeah, yeah. And then the next week he was like complaining about having a low payout. And it was such a good example of like, what have you done for me lately?

Sure, sure, sure.

People are speculating. I saw one account go viral with a post saying that Nikita was fired. Yeah, that was totally— that was me. That seems totally, totally fake. That account got nuked. Good. But there was a— there was a fun theory that as part of the lockup rules, as an active employee or top executives, your shares are tied to strict internal rules for executives. A lockup is frozen, is completely frozen until after fourth quarter results are released. So but because Nikita resigned, as an executive, he's no longer bound by active employer executive trading restrictions.

Oh, interesting.

And so theoretically be able to get out.

But this also has community notes. There's like layers and layers of community notes here.

We have the unlock schedule. We can pull it up here.

Okay. Initial float was 5%. Wave 1 just happened August 11th. Oh, it's coming up. That's 20%. August 21st, that's 7%. September 10th, that's 7%. It's going to be a while until everything gets unlocked. And then, of course, I mean, I'd be surprised if Elon's selling, right? It's like the— he has plenty of money to do everything else he needs to do. And he's never been, you know, one to be—

why would you sell if you're projecting $1 trillion of revenue in 2030?

Yeah, yeah, yeah.

Inside Google DeepMind, Demis leaving his CEO post landed with essentially a shrug. Sources tell me He's already been disengaged from day-to-day management for a while now, but he was a firewall between DeepMind and the rest of Google, even as the two got pulled closer over the last couple of years. I expect that distance to dissolve more with him stepping back.

Yeah, most people are anti-firewall in the sense that they would love like the researchers and the TPU team and the cloud team and the application team to all be deeply integrated, working very closely together. And you get this crazy flywheel between everyone who's rowing in the same direction. And when you have a firewall team, as this reporting is suggesting, you wind up with, well, this person just wants to focus on the most elegant benchmarks. And this person wants to focus on TPU sales. And this person wants to talk about resiliency in their diversity of their cloud revenue. And then somebody else just wants Google Search to not get disrupted by LLMs too quickly because ads need to go up at the right rate. And so differing incentives can create tensions and a dissolving of the firewall, probably something that people would be excited about, I would imagine. But he's been in this, like, you know, elder statesman role for a while that people have been pointing to. Excited to watch him, like, continue that. Will be interesting to see how he instantiates that. Is it— obviously he's working on Isomorphic Labs, but he also probably be writing more blog posts, potentially testifying or talking in Washington. Maybe a book, maybe a podcast tour, who knows? It'll be interesting to follow. What else?

Alex also reported on its current trajectory, Gemini 4 is not expected to push Frontier AI forward the way Fable and Sol just did. So still work to be done.

There's clearly some very low-hanging fruit on AI overviews matching the speed with which those are generated to the lack of hallucinations that we're seeing in the more advanced models. How do you bridge that gap? You can clearly get fact-checked results if you're willing to wait a minute right now. What does it take to get fact-checked results in 5 milliseconds or 100 milliseconds or something like that? That's a huge problem. And but it's something that Google's like set up to do and benefit from. I still see crazy, crazy posts all the time that are screenshots from AI Overviews that are like, I'm a pencil and I see a pencil sharpener coming towards me. And then Google AI Overviews is like, run away! Like, don't— like, that pencil sharpener will shred your wood. And it's like, you know, not— it's like kind of falling for a prompt injection or some sort of meme. And then there's other stuff going on. Tay Kim and Bill Gurley are going back and forth on what this means for Google. Tay Kim says Jeff Dean and Demis Hassabis are two of the most important AI executives at Google. Jeff leaving and Demis is stepping down, he said he stepped up, stepped aside, it's very debatable what direction he stepped. I think he stepped up, up, down, down, left, left, right, left, right, A, B, A, B, start, right? Demis is stepping down from day-to-day operational leadership at DeepMind. And Tae Kim says, "Game over." Very dramatic. Lot of other things going on in the business. Yes, these folks are very deeply important, but The company's been around for a couple decades and has a lot of business lines. Game over is pretty aggressive. Bill Gurley disagrees. He says, this does not have to be game over for Google. I do think they have only one play left now to pull from their own Android Kubernetes playbook and fully embrace open models. It's the obvious move in this situation. That's Interesting. That would be interesting. I mean, they— I believe Google signed the open letter on the earlier side, right, with NVIDIA. Could you see something there? They've been doing good jobs with Gemma. Is that how you pronounce it? Gemma models. Those have been pretty well received. It does seem like you're fighting with one hand tied behind your back as an American open source lab. But who knows? How Intel came back from the brink. We've been covering this story on and off as it's unfolded, but the Financial Times took to the big read to write up the full story as they see it, as they reported it. So it starts with Intel's chief financial officer, David Zisner. David Zisner was already navigating one of the most dramatic periods in the chip manufacturer's history when he took a call from the US Commerce Department official in August of last year. The federal government had just floated the idea of taking a 10% stake in the chipmaker. But Zisner was bluntly told that the structure of any transaction was not up for negotiation, according to an internal Intel memo recently unsealed in shareholder litigation and sources close to the process. So I guess the shareholders are suing over this. Did you— was this deal in the interest of the shareholders? Let's see. They saw the memo and Intel says, hey, look, we had to do this deal. So in order for the government stake to reach 10% demanded by President Donald Trump, Intel had to convert billions in manufacturing grants advanced under the 2022 CHIPS Act, plus $3.2 billion of contracts from the Department of Defense into equity. Its board initially balked at converting the defense contracts, but it acquiesced, and the biggest federal equity intervention in a US company since the bailout of General Motors in 2009 was completed. The group that once dominated the market for PC and data center chips and was still the only US-based company capable of making the most advanced chips, had been saved from a possible breakup. The world's chip consumers, wary of the overconcentration of manufacturing capacity within Taiwan's TSMC, which makes more than 90% of the world's most sophisticated chips, including those who have been used in the booming AI sector, now have at least the prospect of a credible alternative supplier. Since Washington stepped in, as you know, Jordi Hays, Intel has pulled $5 billion of investment from Nvidia, $2 billion from Japan's SoftBank, and its shares have more than quadrupled, trouncing those of rivals. While the deal divided opinion in the semiconductor industry, it had— it has definitely reversed the narrative of decline. Yeah, a lot of people were against this venture communism or, you know, state socialism saying, you know, in America we let independent companies live or die by the sword. The government shouldn't be stepping in. But a lot of people made a very good argument that this is a special case because it's such a critical industry to the American economy and, and the AI race that in this case it wasn't, you know, a bailout for a company that should be just fighting it out on the global stage. And of course, there are heavy subsidies internationally for other competitors to Intel. The company has benefited from rising demand for its central processing units, such as the Clearwater Forest chip launched in June, which can be used for managing AI workloads and data centers. That is the CPU crunch that we've been talking about. The agents need CPUs. They're going to be spending all the time building and chatting on their internal messaging boards. They're going to need CPUs to—

Question for you, John.

Yeah.

Do you Do you think that Tan is more focused on moving the needle or putting points on the board?

That's a good question. I think moving the needle. I think he's more of a moving the needle guy.

Really? Because I mean, my takeaway from story time so far is that he's trying to get some immediate points on the board because he knows that'll lead into moving the needle.

Nah.

But just grabbing a needle that big for a company that old at that scale, that's facing that many headwinds, you can't just grab the needle and expect to move it.

I think the needle's already been established.

You gotta get points.

The needle is, is the fab business that they've been investing in for years now, uh, in, in, in their advanced fabs. And they're just trying to, to slowly move it. If he was putting points on the board, he'd be talking about, oh yeah, I gotta deal with Apple. I got this really small deal with Google. I got this tiny thing over here. That's what points on the board means to me. Moving the needle is like the core thing. It's the main needle, you know?

Yeah. Some good points.

Leave us 5 stars on Apple Podcasts and Spotify. Sign up for our newsletter, TPN.com, and we will see you tomorrow.

It's been an honor.

11 a.m. Pacific.

We love you.