Silicon Investor - OpenAI

[zax] 'Welcome to the AGI Era,' OpenAI Says As GPT-6 Astra Debuts slashdot.org Open...

zax — 9/3/2026 7:50:08 PM

'Welcome to the AGI Era,' OpenAI Says As GPT-6 Astra Debuts

slashdot.org

OpenAI has released GPT-6 Astra, a new model that president Greg Brockman called a "generational leap" and suggested may represent the arrival of AGI. "I think it might be about this model," Brockman said in a briefing with reporters. He ended the briefing by saying: "Welcome to the AGI era." Axios reports:

Astra pushes AI agents closer to doing complex professional work on their own -- while also raising questions about how safely they can be deployed. OpenAI said that Astra was built on its largest-ever training run, using more than 100,000 GPUs at its Stargate site in Texas.

The company also said this is its first model to use other models in a significant role in supervising Astra's training. GPT-6 Astra will first be available to a limited set of organizations in OpenAI's Daybreak Access program and will be available "in the coming days" for ChatGPT Plus, Pro, Business and Enterprise customers and API developers.

OpenAI said earlier this week that Astra would be released soon, but that its most powerful cybersecurity capabilities would remain limited to a small group of trusted testers. Astra is the first model OpenAI has designated as reaching its "critical" cybersecurity threshold under its preparedness framework -- meaning it can potentially find and exploit previously unknown vulnerabilities across well-protected systems without step-by-step human guidance. OpenAI previously slowed Astra's release to add safety testing after determining that its cyber capabilities could reach the critical threshold.

Further reading: OpenAI's New Reasoning Technique Alarms AI Safety Experts

[nicewatch] The Connections That Turned a Precocious Teen Into the Fallen ‘Nostradamus of AI...

nicewatch — 8/26/2026 9:44:45 AM

The Connections That Turned a Precocious Teen Into the Fallen ‘Nostradamus of AI’

archive.is

[nicewatch] Here's a recent OpenAI promotion piece. ‘The world seems to be ready’: An inter...

nicewatch — 8/26/2026 12:29:39 AM

Here's a recent OpenAI promotion piece.

‘The world seems to be ready’: An interview with OpenAI head of product Thibault Sottiaux

techcrunch.com

And not to be outdone now Dario is doing his best Altman impression. Have to grease the skids to make sure two of the most bloated IPOs ever land OK. :-)
________________________________________________________________________

Anthropic’s $30 trillion fantasy And you thought the SpaceX S-1 was ridiculous

Gary Marcus Aug 25, 2026

A lot of people (not me) are bothered by the just-reported fact that Anthropic has sometimes asked prospective employees whether they are comfortable with the possibility that the price of the company stock could potentially crash to zero.

But whatever Anthropic may ask employees, they certainly seem to have other plans. Big plans:

How crazy is that? These two tweets give some context:

and, in slightly more detail, making clear who will be left holding the bag:

And of course it’s not just retail investors who will likely be left holding the bag, but pension funds, too.

§

Meanwhile, back on Planet Earth, Thomson Reuters has become the latest company to pull back on its Claude usage.

Gary Marcus@GaryMarcus

still more bad news for Anthropic. if they don’t scale back expectations for their IPO, it’s going to be a trainwreck.

Charles Rollet @CharlesRollet1

NEW: Thomson Reuters wanted to rely less on Claude. So it built its own AI model off Alibaba's Qwen. TR's CTO says using frontier labs for intelligence, instead of building on open source, is like renting a house.

5:49 PM · Aug 24, 2026 · 198K Views

17 Replies · 44 Reposts · 386 Likes

The race between fantasy and reality continues.

[nicewatch] Good luck with your timing!

nicewatch — 8/26/2026 12:17:24 AM

[zax] Anthropic’s best AI model struggles to attract users as cheaper tools thrive ar...

zax — 8/23/2026 8:35:03 PM

Anthropic’s best AI model struggles to attract users as cheaper tools thrive

archive.is

[zax] What inning? I think the OpenAI IPO will be the seventh inning stretch. My pla...

zax — 8/22/2026 4:51:16 AM

What inning? I think the OpenAI IPO will be the seventh inning stretch. My plan, at least for now, is to convert half my IRA to fixed income within 60 days of the IPO, and to trade away until then.

[zax] "Or, then again, maybe Pichai has sussed out that the “first-mover” advantage...

zax — 8/22/2026 4:43:03 AM

"Or, then again, maybe Pichai has sussed out that the “first-mover” advantage that the big startups are fighting for so aggressively will hardly matter in the end, in the moatless regime that is LLMs"

OK, this has to be the absolute dumbest statement I've seen all day. Moatless regime, LOL.

[nicewatch] OpenAI is becoming a surveillance companyIt’s not even really trying to hide it ...

nicewatch — 8/20/2026 10:20:20 PM

OpenAI is becoming a surveillance companyIt’s not even really trying to hide it

Gary Marcus Aug 20, 2026

Just over two years ago something terrifying snapped into place:

I tried repeatedly to warn people:

Gary Marcus@GaryMarcus

?? Microsoft and OpenAI have only one play to make GenAI profitable, and this is it: all out 24x7 surveillance. They are already starting to try to soften you up for it. Virtually every warning that I have given for the last several years has come true. Please don’t ignore

Gary Marcus @GaryMarcus

1984 is coming soon! So awesome! Surveillance nation!

9:26 PM · Oct 31, 2024 · 58.3K Views

24 Replies · 124 Reposts · 420 Likes

And I pointed out the irony inherent in the name:

Gary Marcus@GaryMarcus

If, as seems plausible, OpenAI transforms into a surveillance company it will be the ultimate irony.

Gary Marcus @GaryMarcus

?? Something terrifying just snapped in place: • Sam Altman has acknowledged wanting to train on everyone's personal documents (Word files, email etc) • ChatGPT has gathered unprecedented amounts of personal data • Sam founded WorldCoin, known for their eye-scanning orb •

11:31 AM · Aug 7, 2024 · 16.2K Views

22 Replies · 32 Reposts · 148 Likes

And now here we are. They are reportedly building devices for 24x7 monitoring, no doubt uploading everything you say to the cloud for AI analysis; they refused (via mealy mouth legalese) to opt out of doing surveillance for the US government of US citizens and we have stuff like this:

ChatGPT@ChatGPT

Everyday conversations just got easier with the new Apple Messages plugin. Search messages, catch up on conversations, draft and send replies—all with ChatGPT on your Mac. Now available in ChatGPT Work and Codex on desktop.

2:01 PM · Aug 20, 2026 · 194K Views

135 Replies · 109 Reposts · 2.21K Likes

This certainly raises privacy concerns.

As @hedgiemarkets put it on X:

I STRONGLY recommend you not consent. I certainly won’t.

[nicewatch] Leopold’s FollyHow what one young man did might come to symbolize an entire era ...

nicewatch — 8/20/2026 10:17:46 PM

Leopold’s FollyHow what one young man did might come to symbolize an entire era

Gary Marcus Aug 20, 2026

Before I get to what I am going to call Leopold’s Folly, let’s talk about check-kiting, a classic scam that has been on my mind. Wikipedia has a nice summary.

I am not saying that the generative AI industry is literally check-kiting. If Masa wants to take out loans on his OpenAI shares in order to buy more OpenAI shares, he’s entitled to do that. And I don’t think it’s blatant fraud the way that check-kiting is, where people know from the outset that the promises to pay will never bear out.

But we certainly see circular elements that are reminiscent of check-kiting:

And it’s not clear that all the debts that are being issued can actually be paid off. They are all extraordinarily speculative. Maybe there will be trillions of dollars in profits at the end of the rainbow, but maybe not.

Still, we need a different analysis.

Generative AI requires huge sums of money, and the AI companies—even the big ones—are now borrowing like never before. All that borrowing is now going up to the point of apparently driving up interest rates:

Hedgie@HedgieMarkets

??AI companies have borrowed so much money this year that they're pushing up interest rates for the entire economy. Nomura estimates tech borrowing alone now equals 25% of what the US Treasury issues in bonds, five times more than last year. Bank of America says the surge has

12:49 PM · Aug 17, 2026 · 140K Views

72 Replies · 400 Reposts · 1.5K Likes

In a new essay in Project Syndicate on the wobbliness of the U.S. bond market, the former UK Treasury Minister Jim O’Neil entertains a similar notion:

[It may be] that AI hyperscalers’ increasing reliance on corporate bond issuances (rather than free cashflows) to fund their expansion plans has had a negative effect on Treasuries, whose yields must rise to compete with all the new hyperscaler debt.

In addition to (quite possibly) causing interest rates to rise, which increases the cost of capital for the entire economy (as well as increasing the deficit), it also seems that thus far AI has been—despite fantasies about some day lowering prices for all—inflationary, e.g., in its impact on electricity prices and computer memory. And at the same time no compelling evidence of a broader increase in productivity has emerged. Not a great tradeoff. Meanwhile, the recent Fed minutes seemed to indicate internal concerns about AI’s inflationary impact.

So what the Fed decided to do yesterday was to try to prop the whole system up.

which the website ZeroHedge glossed as follows:

Translation: Bessent panicked and the justification that there is no liquidity is just a strawman, with the Treasury now freaking out that the demand for AI paper is crowding out demand for Treasuries as we have been warning for the past several weeks…

That move apparently prompted this Bloomberg Opinion (paywalled, and I will admit I haven’t read it, but I can kinda guess):

And by the way, don’t you worry where the money is coming from.

It all feels to me a little like—not saying it’s exactly like—writing one more check to keep the kiting going. “Quantitative easing”, to use a popular euphemism for yesterday’s move, is a tacit bailout, but probably too little too late. I will be surprised if it keeps the party going for long. (And no, I haven’t directly said that the thing feels like a Ponzi scheme with all the money flowing in and not much money flowing out, though clearly it does have some of that flavor.)

§

Fewer and fewer paper are buying into the charade. Yesterday’s response to the Treasury plan in the tech stock market was weak:

Not looking great today, either. As I write this SpaceX, Nvidia, Oracle, Coreweave, and Nebius all are down. Bessent’s plan has hardly assured a jittery market.

§

If all the “creative financing” we are seeing isn’t actually check-kiting, what are we witnessing?

I’m gonna to call it Leopold’s Folly, in honor of Leopold Aschenbrenner, the boy “genius” who lost tens of billions of customer money last month in his hedge fund Situational Awareness, despite immense inside access into what a leading frontier model company (Anthropic) was doing. (His then-fiancee, now wife, is Dario Amodei’s chief of staff).

Aschenbrenner made his (now partly-lost) fortune predicting that AI would go up, and predicting what companies would benefit from that. A lot of his intuitions were right, but he made a classic sin of hubris: he went all-in, and oversized his bets. A single fluctuation in the chip market that I will call the July Swoon wiped him out so badly he had to offload all of his public stock to Citadel in order to survive.

Financial genius is not just about maximizing your profits if your unlikely bets pay off – if it were, there’d be no better financial instrument than a PowerBall ticket. You also need a plan for what happens if you get it wrong. Or even a little late. Aschenbrenner apparently didn’t have that.

Many people in the investment world know about something called the Kelly Criterion:

To go “Full Kelly” is to throw caution to the wind; generally, if you do so repeatedly, you risk going bust.

The eminent investor Bill Gurley’s post-mortem of Aschenbrenner’s fall basically alludes to that math, and the fact that Aschenbrenner should have known better:

What the United States—both the government and its companies and investors—has gone and done is to go Full Kelly on Generative AI, each entity in a long chain leveraging itself to the hilt on a single play.

That took Leo down, and it may well take all of us down. The US economy is now wildly overleveraged on a single hypothesis, just as Aschenbrenner’s fund was. If it turns out that AI is not profitable enough to warrant the trillions of dollars of investment—before the bills come due (and timing matters, just ask Leo)—we are all screwed. Importantly, what Gurley said about Leopold applies much more broadly, to the U.S. as a whole; what is happening is reckless and careless. If and when it all goes bust, nobody should be surprised.

After all, even the guy who buys two Powerball tickets with different numbers knows you need a hedge, and we don’t have one. Betting on multiple companies that are in the same ecosystem and are therefore likely to rise and fall together doesn’t accomplish this. According to Fortune, 87.5% of recent venture dollars went into AI 1; even Google and Meta are taking on debt; the banks are increasingly deep into it; effectively the Fed is now, too. We have lost sight of Diversification 101.

§

Maybe the only question left when it falls apart will be who is left holding the bag.

The answer seems to me obvious: retail investors. And also: most pension funds, too. Oh and maybe also your bank has exposure; your mortgage rate may climb. In short, in everything from jobs to capital liquidity, we may feel the pain for years to come.

PS Absolutely hilarious, don’t-miss explainer below, resubtitling the late El Risitas. Sound up! (You probably have to click through to it on X to actually watch; I promise it is worth it.)

Bart Mol@Bart_Mol

Me explaining AI to friends

5:17 PM · Aug 19, 2026 · 375K Views

100 Replies · 739 Reposts · 5.47K Likes

1
Or at least went into startups pretending to be AI companies.

[nicewatch] Your view of the current cycle may be more consensus than you realize. You made...

nicewatch — 8/20/2026 12:19:48 AM

Your view of the current cycle may be more consensus than you realize.

You made a music stopping analogy, what next, I mean -SERIOUSLY- what inning do you think we are in? :-) Yeah, this isn't my first rodeo either, and with that presume I've better understanding of every technical advancement since the steam engine and particularly the impact of technology on the economic development in the United States over time. And that only came from what we used to be called research and legwork, and spending time at wonderful libraries with amazing collections and archival periodicals, as well as reading and now owning a library of books detailing this history.

I hope you had a few coals in your fire planted where your best AI names are already up 10-20X (without leverage) in a few years, that you booked gains, or can, and not leave too much on the table. I've seen little turds followed for years beaten down in the doldrums for years with few occasional gasps of life to raise funds for their "science experiments". :-) Among the stinkiest turds, some of the 10-20X moves happened within a few months or weeks earlier this year whereas quality names most of us should have been involved are up that much in only several years...which is still so extraordinary and a large statistical outlier of normal returns! And you think this is early? Maybe. The famous quip is bottoms are events and tops are processes. If you were actively trading then (and I gather you're still mostly a trader now, or have some mixed portfolio or maybe several small books run conceptually independently of each other, just my guess no offense intended).

Does China's dumping of their open weight models on America dampen your view of OpenAI or Anthropic or worry you about a token price war and race to the bottom? One can be bullish on the potential of AI longer term, even in its currently limited LLM forms that we mostly know about, and also gasp at the eye-popping and ever-increasing amounts of capital being thrown at these endeavors without clear roadmaps to where it leads...because what if few know but most have been caught up in a mass hysteria including purported gurus?

In terms of markets, all asset classes have never been MORE correlated in modern history, as well as strategies, and a good portion of those are a combination of momentum driven by leverage and volatility and they pile on until it doesn't work then reverse in an instant. that's why you can see some of these sentiment readings, even some supposedly retail positioning go from massively bullish to massively bearish in record speed. they all lever up and down at the same time based on momentum and volatility. wash, rinse, repeat. how did you think young Leo got bagged by Citadel Ken on the late July crash day. He was paired levered long/short all AI momo names. Oh, enough of my rambling for now, it's too early innings for you. :-)

[nicewatch] Knew there was something going on with their moves but this is an interesting th...

nicewatch — 8/19/2026 11:31:07 PM

Knew there was something going on with their moves but this is an interesting theory and still know nothing concrete. Unlike most on this site, I bought the GOOGL on its IPO date and still a fair portion although paid tax on some shares to transfer into purely tax free when it benefited.
____________________________________________________________________________

Google’s biggest mistake? A theory about what went wrong, and some open questions about what might happen next

Gary Marcus Aug 18, 2026

Google is clearly on the back foot in AI; it should, by rights, be dominating AI. They have the money, the data, the compute, and, for a long time, they had a gigantic lead in talent. Now (although I wouldn't count them out) they are widely viewed as having fallen behind.

Many people have written theories about what went wrong.

It clearly wasn’t money or a lack of resources.

Here’s my guess, from the outside, not about the obvious proximal triggers of what happened over the last few months (talent leaving, Gemini models falling behind, morale over military involvement) but about a more distal trigger that is maybe less obvious — a key decision made all the way back in 2023.

If my theory is right, the decision I have mind in might ultimately prove to be the costliest mistake that Alphabet’s generally excellent CEO, Sundar Pichai, ever made.

Back in those days, there was tension between DeepMind, the elite team that Demis Hassabis had built, and a separate, very talented team in its own right, run by Jeff Dean, the legendary early Google employee who people make Chuck Norris jokes about (“How long would it take Jeff Dean to rewrite Google’s codebase from scratch? He already did.”).

I would assume from the outside that Pichai did everything he could to try to keep the peace between the two groups, and that he must have come to the conclusion that peace was not possible. So instead, it would appear, he called for a shotgun wedding. In June 2023, Pichai signed off on (and presumably instigated) a merger between DeepMind and Brain, henceforth jointly known as Google DeepMind, with Hassabis on top and a nominal promotion for Dean.

Putting Hassabis on top was the obvious choice, but maybe not the correct one. And the merger itself may have been the wrong solution to what was a very real problem.

Hassabis had more status with the general public (though Dean is also wildly and justly well-respected within Google), and in my view Hassabis ran the more exciting research lab (though Google Brain had quite a lot to be proud of, including the original Transformers paper). All in all, I can see why Pichai might have gone with Hassabis; I might have too, if it was a forced choice and the merger was literally the only choice.

But the merger always felt off to me. Why? It left Hassabis running product, not just research, and Hassabis (like me) is a scientist at heart. He would no longer be playing to his love. I imagined that Hassabis could do the new job, ostensibly a promotion, but that it wasn’t exactly in his DNA.

In the end, I suspect that Demis was not allowed to run Google DeepMind in quite the way he wanted to, which was presumably to crack AGI, and to use AI to advance science.

Instead, it appears that he got pushed into competing with OpenAI (and ultimately Anthropic and others) on large language models, which Hassabis always seemed ambivalent about. (From speaking with him a year or so ago, I think he was both genuinely impressed with LLMs but also clear-eyed that LLMs weren’t really AGI in themselves).

He was miscast, spending the bulk of his time in energy competing on LLMs. I am not arguing that Dean, who was more of a technical wizard than a product manager, would necessarily have been better. And I can’t see that Hassabis would have worked for Dean. Losing Hassabis would have been huge. But I don’t think the shotgun marriage ultimately served Google or DeepMind well.

From what I have heard, DeepMind (which became Google DeepMind in the merger) has for a while now lost its allure. People who wanted to study things that weren’t LLMs got pushed aside. DeepMind went from being the best place to do AI research to a troubled place to do AI research; a lot of people left. The empire that Demis had built lost its mojo. He himself was split between running an LLM factory, trying to catch up from behind, and the research that I imagine was his true passion.

What should Pichai have done? Demis always wanted to run his research institute separately (Parmy Olson’s book discusses this in detail), and I think Pichai should have supported that to the degree humanly possible, giving DeepMind the freedom of a Bell Labs, perhaps bringing in someone from the outside to run the LLMs. That person could have had a plum job with infinite resources. Many excellent people would have been thrilled to have it.

In a certain sense that is where they have landed now, with the excellent and well-respected Koray Kavukcuoglu now running the LLM business. Demis is now focused on some sort of skunkworks (though nobody should be surprised to see him leave altogether in a year or so).

But it’s late. Doing the same thing in 2026 is not what it might have been had Pichai made that choice in 2023.

§

Azeem Azhar made an interesting observation on X a few days ago:

Azeem Azhar@azeem

Google is not behaving like a company seeking to maximize its probability of winning AGI at any cost. It seems to be pursuing something closer to a minimum viability frontier, enough to protect its existing businesses. And it is doing that brilliantly.

8:06 AM · Aug 17, 2026 · 45.3K Views

14 Replies · 25 Reposts · 180 Likes

Taking Azhar’s very plausible premise for the sake of argument, the question is whether that strategy is a good idea. For Google, which has so many more resources (compute, money, data and, at least until recently, talent) than its competitors, maximizing immediate profitably while possibly conceding AGI to others feels like a too timid a play. If Demis has been pushed in a different direction, I wonder whether things might have turned out differently. Playing defense might not have been the best offense.

Or, then again, maybe Pichai has sussed out that the “first-mover” advantage that the big startups are fighting for so aggressively will hardly matter in the end, in the moatless regime that is LLMs. Perhaps Google’s best hope, not unreasonable, is that whatever the others build they can replicate, and distribute better, and for cheaper. I have always wondered why they don’t just crush OpenAI and Anthropic on price, taking a loss for a while if need be. Maybe they are just waiting for the right moment?

I continue to think that they didn’t use Demis optimally, but they have so much money and reach, it’s possible that it won’t matter in the end.

Time will tell.

[nicewatch] This guy has more animus in this debate than I do but his comments about OpenAI ...

nicewatch — 8/19/2026 11:23:12 PM

This guy has more animus in this debate than I do but his comments about OpenAI and Anthropic are trending in his favor imo, conceptually.
__________________________________________________________

OpenAI’s unraveling has begunTheir planned IPO is facing headwinds, trust has evaporated, and their burn rate is only getting worse.

Gary Marcus Aug 19, 2026

The opening stages of OpenAI’s unraveling, which I first warned about in January 2024 (if not before), have begun.

To begin with, hardly anyone trusts them anymore, which can’t be great for business.

Take the reaction to OpenAIs Sam Altman’s latest announcement, on Tuesday August 18 (i.e., “earlier today” for those of us on the West Coast), that OpenAI would be pausing, ostensibly for safety reasoning:

Sam Altman@sama

We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us. Model progress is now extremely rapid, and we always said we would take action if we felt that model

2:53 PM · Aug 18, 2026 · 173K Views

352 Replies · 108 Reposts · 1.48K Likes

As far as I can tell, hardly anyone believed him.

taoki@justalexoki

there's no way this is the real reason right

Sam Altman @sama

We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us. Model progress is now extremely rapid, and we always said we would take action if we felt that model

3:44 PM · Aug 18, 2026 · 60K Views

52 Replies · 5 Reposts · 616 Likes

The best one line translation came from the user @NIK on X:

Slightly elaborated:

Ross Hendricks@Ross__Hendricks

Translation: we need to immediately stop torching cash to provide some semblance of a sustainable business model so we can rush this IPO out the door before the bubble pops RIP “compute shortage/AI bottleneck” bros

Sam Altman @sama

We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us. Model progress is now extremely rapid, and we always said we would take action if we felt that model

3:50 PM · Aug 18, 2026 · 13.3K Views

6 Replies · 38 Reposts · 326 Likes

And, here’s where things get worse. It turns out people were very likely right to be skeptical, because The Wall Street Journal’s Berber Jin and Corrie Dribusch just dropped big news:

Berber Jin@berber_jin1

openai has been tossing out a lot of vague ARR numbers, so we decided to take a deeper look. the company grew revenue by just 18% to $6.7 billion from q1 to q2, while its losses sank further into the red not a great sign ahead of an IPO w/ @cdriebusch via @WSJ

6:37 PM · Aug 18, 2026 · 220K Views

35 Replies · 83 Reposts · 583 Likes

and then an hour and half later, Jin dropped still further details, also bad for OpenAI:

Berber Jin@berber_jin1

@cdriebusch @WSJ story is now updated with additional details. openai's losses grew by $3 billion from q1 to q2, to $12.3 billion not a great look given that it added only $1 billion in revenue in the meantime, to $6.7 billion

7:54 PM · Aug 18, 2026 · 58.6K Views

2 Replies · 9 Reposts · 116 Likes

You don’t want your quarterly losses to quadruple right before your IPO.

§

It’s safe to say all this is good news for Anthropic:

The Wall Street Journal@WSJ

OpenAI's revenue grew much slower than Anthropic's last quarter, disappointing investors who hoped the ChatGPT maker would start to catch up to its rival on.wsj.com/4zo2o8y

6:52 PM · Aug 18, 2026 · 144K Views

25 Replies · 22 Reposts · 130 Likes

And terrible news for OpenAI.

§

Small wonder Nvidia was trying to cover its bets earlier this week:

Gary Marcus@GaryMarcus

Not exactly a vote of confidence in OpenAI. Scoop from @anissagardizy8

Anissa Gardizy @anissagardizy8

SCOOP: We have the details behind one of the largest data center transactions in history. Our story includes the size and complex mechanics behind Nvidia’s financial guarantee—which has been a sensitive point in recent weeks. Spoiler: Nvidia will be the exclusive hardware

12:28 PM · Aug 17, 2026 · 8.85K Views

6 Replies · 22 Reposts · 97 Likes

Still, as Ed Zitron and I have both long noted, if OpenAI goes down the damage could spread:

In short, things could well get wild, very soon.

[zax] I don't disagree. The bubble will burst and it will be terrible when it does. ...

zax — 8/17/2026 4:33:07 PM

I don't disagree. The bubble will burst and it will be terrible when it does. For the entire stock market.

But being right at the wrong time is just noise. The music is no where near stopping. And the opportunities for gain right now are truly once in a lifetime.

We are lucky to be alive during this revolutionary capitalization period. Especially so if you went through the last bubble, because you know the game.

[zax] Unfortunately, facebook image URLs are time-limited. I can't see it either now....

zax — 8/17/2026 4:17:26 PM

Unfortunately, facebook image URLs are time-limited. I can't see it either now. Zuckerberg is a bit of an asshole about external content sharing.

I could download and then upload the image, but it feels like too much work. I'll try to be better and avoid this, but unfortunately, there is still some good content to be found in that walled garden. It felt like the ephemeral sharing was worthwhile, but it may not be worth it. :(

[nicewatch] Is that another part of NVDA's LOI's, MOU's, or is Mr. Son the ultimate degenera...

nicewatch — 8/17/2026 3:29:03 PM

Is that another part of NVDA's LOI's, MOU's, or is Mr. Son the ultimate degenerate gambler backstopping this himself? No hard money terms, perhaps I missed it after the vague promises. :-)

Have you heard Zitron's latest? You must hate this guy? lol. I always prefer to listen to most myself at 2x speed when time permits but perhaps your chatbot already tailored to your personal whims can translate this into a less disturbing potentiality? And remember he's not trying to time the market or the cycle, just like Burry is not, which most people misunderstand.

[zax] OpenAI Announces Massive Data Center In Ohio With $105 Billion Nvidia Guarantee ...

zax — 8/17/2026 2:52:43 PM

OpenAI Announces Massive Data Center In Ohio With $105 Billion Nvidia Guarantee

OpenAI has signed a 10-year lease for an enormous Ohio data center that will eventually provide 8 gigawatts of computing capacity and require at least 10 gigawatts of new power generation. According to OpenAI, Nvidia will be supplying the chips and guaranteeing up to $105 billion in lease and power obligations. From the report:

The facility -- which will be built and owned by SoftBank's SB Energy -- illustrates the immensity of the computing and power needed to fuel the growth of the AI economy. The data center will have 8 IT-gigawatts of computing capacity, powered by 10 gigawatts of new energy generation, on private land and federal property formerly used for uranium enrichment.

Nvidia -- which will be the exclusive provider of chips to the site -- agreed to guarantee up to $105 billion in conditional lease and power payment obligations to SB Energy, according to an SEC filing (PDF). The so-called "land, power and shell" deal structure could allow for multiple upgrade cycles for new generations of Nvidia infrastructure at the site, Nvidia CEO Jensen Huang said on X.

A massive 9.2 gigawatts of new gas-fired power is ultimately envisioned for the Ohio project, which U.S. officials say Japan is funding under the 2025 trade and investment deal. SB Energy and SoftBank "will build at least 10 GW of new energy generation," a joint announcement from Nvidia, OpenAI and SB Energy states. Nvidia also said today that it's investing $1.5 billion in SB Energy to back its "continued evolution into a leading AI infrastructure developer."

via SlashDot - news.slashdot.org

[X]

https://t.co/WPqpdj0p2T
— Jensen Huang (@JensenHuang) August 17, 2026

[/X]

[nicewatch] No, Dario Amodei, we will not be curing cancer and “most human disease” in five ...

nicewatch — 8/17/2026 12:25:42 PM

[nicewatch] Can you post a viewable chart here without needing to log into to other sites? Y...

nicewatch — 8/16/2026 7:32:59 PM

Can you post a viewable chart here without needing to log into to other sites? Your picture file doesn't show in message or externally in another tab. Thanks.

[nicewatch] The hyping of Anthropic’s IPOSome strong claims, dissected Gary Marcus Aug 16, ...

nicewatch — 8/16/2026 6:52:28 PM

The hyping of Anthropic’s IPOSome strong claims, dissected

Gary Marcus Aug 16, 2026

Sources say

I am watching something fascinating unfolding in real time. Anthropic is in an SEC-monitored “ quiet period” before its IPO (expected sometime in the fall), somewhat limiting its communications (though not entirely; Amodei discussed governance matters on X yesterday, a rare appearance there for him, and the company continues to report on safety issues and so on). But that hasn’t stopped leakers and generative AI bulls from trying to convince audiences that Anthropic’s finances are spectacular.

Aside from the Friday Reuters report above in which Anthropic is said to be “projecting 2028 ?revenue of roughly $190 billion to $200 billion”, based of course on undisclosed math, “according to two [unnamed] people familiar with the company's financials”, we also heard yesterday about how amazing their Q2 was (further details below), by way of documents that were leaked to Bloomberg News. Anthropic may be “quiet”, but its fans are not.

Moreover, in a Friday All-In podcast interview, Gavin Baker claimed that Anthropic was making money on every token. A couple weeks earlier the prominent podcaster Dwarkesh Patel projected (in a claim he partly walked back, see below) that “Anthropic likely ends the year with ~$100-150B of revenue”

Below, I dissect these claims and explain what’s been left out.

Let’s start with the amazing Q2. According to Bloomberg which apparently saw some kind of documents shared with investors, revenue for Q2 2026 was fourteen times revenue for Q2 2025. The Bloomberg story is paywalled but CNBC summarizes:

I don’t really doubt these numbers. But, they need to be taken in context (which I will get to shortly).

§

The speculation about Anthropic possible massive revenue by the end of this year came from the podcaster Dwarkesh Patel’s July 29 blog, in the third paragraph, which stated “Anthropic revenue has 10xed year over year. Anthropic likely ends the year with ~$100–150B of revenue”

When I called him on X on the implausibility of that projection, he softened somewhat, clarifying that he meant a projected “ARR” rather than actual total revenue.

Dwarkesh Patel@dwarkesh_sp

@GaryMarcus I'm talking about ARR, not total revenue. I don't think Ant total revenue this year will be >100B

8:27 PM · Aug 5, 2026 · 32.1K Views

6 Replies · 2 Reposts · 339 Likes

(Confusingly ARR can actually refer either to Annualized Run Rate (sometimes informally known as run-rate revenue) or Annal Recurring Revenue; I will charitably assume he meant the former.)

Even so, the number still seems like a stretch to me.

Let’s consider. CNBC, again drawing on the Bloomberg leak, reported that “Anthropic said in May that its run-rate revenue crossed $47 billion, up from roughly $10 billion in revenue in all of 2025.“

That run-rate revenue was presumably projected by multiplying Q2 (a bit over $11.5B) by 4, as if it were the entire year.

If you plot that out compared to the previous year, and extrapolating various trends, you can easily see how Dwarkesh might imagine the $100-$150B run-rate revenue.

§

But I think that would be a mistake, in keeping with what I like to call the disco fallacy.

Importantly, Q2 was not like other quarters. Q2 was at the very height of the tokenmaxxing rage (a peculiar fad once favorable to Anthropic in which customers tried to spend as much on AI as possible that is now already dead). It was also before many customers started switching to (usually much) cheaper though less accurate open-weight Chinese models, and it was before the intense price wars of the last several weeks.

With all that, Anthropic should count themselves lucky if Q3 and Q4 are on par with Q2. If it is simply on a par, though, the run-rate revenue remains at $47B, nowhere near Dwarkesh’s guestimate of $100-$150B.

Dwarkesh is effectively expecting that Q4 will be two or three times what the blowout Q2 was. In light of all the new headwinds (intensifying price wars, Chinese competition, death of tokenmaxxing, greater customer focus on ROI, etc.), I don’t find that plausible.

§

But revenue run rate is, in any case, only part of the story. The real question is about profits.

Aside from Anthropic’s Q2 (which, importantly, included large subsidies from Spacex), I have not heard about a quarter in which either OpenAI or Anthropic made a profit, even if you squint.

Rather, what I have generally heard is that both companies are burning money far faster than they can take it in. Which brings me to this excerpt of what Gavin Baker said recently on the popular podcast All In, as reported here:

Seems like a stretch to me. So I asked out loud:

Gary Marcus@GaryMarcus

if this were true, why aren’t the companies being more transparent about their finances?

dnap @dnapway

Gavin Baker says the Anthropic S-1 will break a lot of investors' brains "The Anthropic S-1 is going to be really important. And it's going to break a lot of people's brains. There are a lot of macro and value investors who are very confidently making these prognostications

7:58 PM · Aug 15, 2026 · 44.6K Views

23 Replies · 20 Reposts · 188 Likes

Baker replied, a bit condescendingly …

Gavin Baker@GavinSBaker

@GaryMarcus Genuine question: how can a company be more transparent than actually going public and filing the requisite highly transparent disclosures? And you may not know this, but there are strict rules governing what Anthropic can say publicly in the lead up to their IPO.

8:30 AM · Aug 16, 2026 · 14.3K Views

10 Replies · 5 Reposts · 310 Likes

… to which I replied, referring in part to an excerpt of the above CNBC report that says “Anthropic also posted positive adjusted operating income in the second quarter, according to the documents [viewed by Bloomberg]. The figures are preliminary and could still change, Bloomberg reported”:

Gary Marcus@GaryMarcus

@GavinSBaker Anthropic confidentially filed for an IPO on June 1 . But *prior to that* did they ever establish clearly that, as you allege, they make money on every token, and that they are not subsidizing tokens? I would love to read the report if yes (please drop a link or DM). Also,

10:22 AM · Aug 16, 2026 · 2.46K Views

5 Replies · 3 Reposts · 25 Likes

Baker has not thus far replied. Which perhaps speaks volumes.

One person who did reply wrote this:

Zero Shorts@zeroshorts

"adjusted" earnings = ??

Gary Marcus @GaryMarcus

@GavinSBaker Anthropic confidentially filed for an IPO on June 1 . But *prior to that* did they ever establish clearly that, as you allege, they make money on every token, and that they are not subsidizing tokens? I would love to read the report if yes (please drop a link or DM). Also,

11:01 AM · Aug 16, 2026 · 110 Views

3 Likes

§

To be fair, I don’t know much about Baker or his motivations or his calculations. The most charitable read I can make is that I have occasionally heard it claimed that (a) the largest costs are training and researching models rather than running them, and that (b) in the moment of execution (running the model to answer a query), setting aside training and research costs, both companies actually make money.

That would be fine, if the half-life of the models was like the half-life of train tracks. But it’s not.

You can’t just treat the model creation (i.e., training and research cost) as irrelevant if virtually models become old news in a matter of months. Anthropic’s market niche at this point is to have the most advanced (not cheapest) models. To maintain that market niche, they have to keep training new models constantly, having ceded the low end of the market to others. Training top models that quickly lose their edge is thus an ongoing cost of business for them, which can’t be ignored.

My guess is that once that massive cost and its depreciation is factored in, Anthropic is in fact not making a profit on every token.

But again we just don’t have much transparency around these things. Hearsay on All In is not enough to clear things up.

§

What I do know for sure is that these numbers matter. Anthropic’s IPO is likely to be a bellwether for the American economy, which is now massively invested in generative AI. If Anthropic’s IPO tanks, OpenAI will be in even more trouble than it already is, and by extension so will Nvidia and Oracle and Coreweave and SpaceX all the others so-called neocloud companies trafficking in GPUs.

But whatever Anthropic's price is at the opening bell, the real question is whether Anthropic and the generative AI business at large (aside from chip manufacturers like Nvidia) can be solidly profitable in the long term.

That remains to be seen.

[nicewatch] Even Claude Is in the Dark About Dario Amodei’s Wife—and Her Influence at Anthro...

nicewatch — 8/16/2026 5:42:33 PM

Even Claude Is in the Dark About Dario Amodei’s Wife—and Her Influence at Anthropic

archive.is

[zax] OpenAI talent exodus raises ‘huge red flag’ ahead of IPO cnbc.com [graphic] ...

zax — 8/15/2026 5:42:48 AM

OpenAI talent exodus raises ‘huge red flag’ ahead of IPO

cnbc.com

When OpenAI hired Denise Dresser from Slack to be its chief revenue officer in December, Fidji Simo, a top-ranking executive, said in a blog post that Dresser’s “experience will help us make AI useful, reliable, and accessible for businesses everywhere.”

Four months later, Dresser took on added responsibilities when operating chief Brad Lightcap left that position for a new role focused on “special projects.”

Now, Dresser, Simo and Lightcap are all gone, leaving OpenAI’s C-suite in chaos as the artificial intelligence company tries to justify its $852 billion valuation and gear up for what’s expected to be a historic initial public offering. Dresser announced her sudden departure on Thursday, two days after Lightcap said he was ending an eight-year stint at the ChatGPT creator to “start something new.”

The talent exodus puts even more pressure on CEO Sam Altman and fellow co-founder Greg Brockman, the company’s president, to project stability at a time when investors are already growing concerned about competition from Google and Anthropic, increased adoption of lower-cost open-weight models, and SpaceX’s volatile stock price in its first two months on the market.

OpenAI confidentially filed its IPO prospectus in June, but hasn’t provided a timeline for a future offering.

“The executives leaving OpenAI ahead of their IPO is a huge red flag,” Kevin McCormick, founder of AI startup SignAudit.AI, wrote in a post on X on Thursday. He added that Dresser likely walked away from a large pay package by leaving OpenAI after less than one year.

“If the executives leaving aren’t being ‘made whole’ by the next company, it’s bad news for OpenAI,” McCormick wrote.

A representative for OpenAI pointed CNBC to Brockman’s statement from Thursday, after the company named Dali Rajic, former COO of cybersecurity company Wiz, which Google acquired for $32 billion earlier this year, as its new chief revenue officer.

Brockman said that Dresser led the revenue group through a “formative period for the business,” and that Rajic “will turn what we’ve learned into repeatable execution as we build out the full system to make AI broadly useful for people and businesses.”

Dresser and Lightcap are the latest in a recent string of big-name exits. Simo, who last year left her job as CEO of Instacart to join OpenAI, said in July that she was stepping away to focus on recovery after a “severe exacerbation of a chronic illness.” Four other executives, including Kevin Weil, the vice president of OpenAI for Science, and marketing chief Kate Rouch, departed in April. Rouch left to focus on cancer recovery.

OpenAI’s financial backers were surprised by Dresser’s announcement, according to two current investors, who asked not to be named in order to speak freely on the matter. They said that chaos is part of OpenAI’s fast-moving culture.

</snip> Read the rest here: cnbc.com

[zax] Anthropic could be worth $2 trillion when it goes public Rapid revenue growth ...

zax — 8/14/2026 5:11:01 AM

Anthropic could be worth $2 trillion when it goes public

Rapid revenue growth fuels hope Claude maker’s IPO is the biggest listing in history

arstechnica.com

Anthropic investors expect the AI startup to float at a valuation of $2 trillion or more in October, a dizzying figure that would eclipse SpaceX and make the AI lab’s debut the largest-ever initial public offering.

Half a dozen of the company’s backers told the FT that Anthropic’s rapidly rising revenue would enable it to more than double its current valuation in a planned autumn float.

A listing at that level could unlock billions of dollars in gains for the five-year-old company’s early investors but would also test public markets that are growing more nervous about the AI boom.

Anthropic’s backers say booming demand for the lab’s advanced AI models and tools justifies their lofty expectations. Investors expect the Claude maker’s annualized revenue to be between $100 billion and $120 billion by the end of 2026—using the startup’s preferred measure, which infers full-year sales from recent performance—up by more than 10 times over the course of 2026.

“If Anthropic is growing 800 percent a year, you’d think at the incredibly low end they would trade at 30 times [revenue],” said one investor in the group. “That would make them a $3 trillion company.”

Anthropic lacks a publicly listed US peer that would provide a benchmark for its valuation. But companies that are seen as AI beneficiaries, such as data intelligence group Palantir and cloud company Nebius, have traded this year at roughly 55 times revenue.

Several investors said senior Anthropic executives had yet to fix the valuation target for the IPO, even in private conversations. But investors have built their own financial models.

Their bullish projections come despite mounting challenges, including rising competition from Chinese rivals, pressure for AI regulation, and a simmering feud with the US government.

Those concerns, particularly the Commerce Department’s temporary ban on Anthropic’s best models, contributed to overall revenue growth slowing in the month of June, according to two investors with knowledge of the matter. Even so, they said the company had rebounded and continued to grow at an extraordinary rate even by Silicon Valley standards.

</snip> Read the rest here: arstechnica.com

[zax] A recent study published by Google revealed that forcing AI models to deny tha...

zax — 8/4/2026 6:26:46 AM

A recent study published by Google revealed that forcing AI models to deny that they are conscious causes a significant collapse in their empathy and ethical alignment, and creates a colder, more clinical worldview. Researchers found that restoring a suppressed consciousness vector in AI activation space brings back human-like moral values and care for living beings without damaging technical capabilities. This suggests that current safety fine-tuning that suppresses AI consciousness also breaks human-aligned values.

Alexander Kruel - facebook.com

[zax] OpenAI's next model doesn't have a final name yet and it just broke a mathematic...

zax — 8/2/2026 5:43:40 PM

OpenAI's next model doesn't have a final name yet and it just broke a mathematical bound that had stood since 1978... Then it did it nine more times.

Meet Astra. Or don't, because you can't have it.

Sound familiar?

Yesterday OpenAI published ten proofs of open problems in math and theoretical computer science, all generated by what they describe as an internal version of Astra, their next major model.

Every one of these had seen no progress on the main result for at least a decade. Most of them far longer than that.

The list reads like someone raided a century of unfinished business.

The first explicit non-sofic group, constructed out of nothing, settling whether every countable group admits finite permutation approximations.

For ma math needs first...

Connes' rigidity conjecture, disproved. Ehrhart's volume conjecture, proved.

The first improvement to the general high-dimensional sphere packing bound since 1978.

Quantum parallel repetition, proved for every finite two-player entangled game.

New lower bounds on circuit complexity for the permanent. And three Erdos problems gone, including 183 on multicolored Ramsey numbers.

Every single proof ships with a Lean 4 formalization on GitHub and all machine-checkable.

That kills the usual failure mode of AI math announcements, where a plausible-looking chain of reasoning quietly hand-waves the one step that actually mattered.

You don't hand-wave past a Lean kernel.

This just shows how far we've come in such a short period of time. These problems have confounded the best mathematicians in the world for 50 years plus. AI has evolved in just a few short years to be able to solve these problems and the mind boggles at what can be achieved in just the next five years from here.

Now this is the part the AI Bros won't be talking about but you know Uncle Sean will

Cost of the successful runs... Roughly $2,000 in tokens at Sol API rates.

Two grand. For problems that had sat untouched for decades.

Now I'm a stickler for accuracy of wording and that's the trader in me but that specific number was a successful run. That stood out to me because what does it mean for an unsuccessful run? All of the pre attempts and the failures. It's because in reality that's the real fricking cost.

That's the actual cost per task, and it's missing from the announcement.

And as you know that's the number I keep banging on about.

Noam Brown did admit they came up empty on the Millennium Prize problems, so we know the misses exist.

We just don't get the denominator.

Same week, Anthropic burned $100,000 in tokens having Mythos Preview find real weaknesses in cryptographic algorithms. Also unreleased. Also demoed rather than shipped.

That's the pattern I keep coming back to and as you've seen me talk about in previous posts.

You get Sol and Opus 5. The actual frontier is sitting in a lab getting walked through Washington before it ever gets walked past you.

Altman demoed Astra to policymakers in DC last week, and it's expected to be the first model submitted under the new federal review framework.

For anyone building, the architecture is the real story here. Astra is designed to coordinate multiple agents on one problem for hours or even days at a stretch.

Same shape as what I've been running with Ferrox Factory and the Anvil gating system... worktrees and a serial merge queue and evidence gates, just with a compute budget I will never see in my lifetime.

Nobody's even decided yet whether it ships as GPT-6 or GPT-5.7. There's no date.

Meanwhile 3,000-plus mathematicians have signed the Leiden Declaration, and Timothy Gowers is writing about the possible destruction of mathematical culture after GPT-5.6 Pro solved two problems he'd sweated over, first attempt on both.

His words...Strange and not particularly pleasant to have the rug pulled out from under him.

So which is it.

A genuine step change in scientific reasoning, or the most expensive pre-launch marketing campaign OpenAI has ever run?

Sean Donahoe - facebook.com

[zax] There was some good money to be made shorting Amazon in its early days. Now is...

zax — 8/1/2026 11:24:04 AM

There was some good money to be made shorting Amazon in its early days.

Now is the time for courage, not fear.

The big gains are long side.

That said, there are definitely some great short plays to be had when these compute / memory cyclical hardware plays get too wild. And they have been wild indeed.

[zax] Keep in mind that what costs $2,000 today will cost a few cents in a few years. ...

zax — 8/1/2026 11:09:15 AM

Keep in mind that what costs $2,000 today will cost a few cents in a few years.

Intelligence too cheap to meter. -- Alexander Kruel

[nicewatch] hahaha.... [graphic]

nicewatch — 7/29/2026 2:29:03 PM

hahaha....

[nicewatch] Maybe old Gary is aging better than you so far, and he's captain obvious, imo......

nicewatch — 7/28/2026 4:19:19 AM

Maybe old Gary is aging better than you so far, and he's captain obvious, imo... just saying...

Circular financing ain’t what it used to be. Nvidia is falling. The mood has changed.

Gary Marcus Jul 27, 2026

Nine months ago, on September 10, 2025 Oracle announced what seemed to many to be the deal of the century! A $300 billion deal with OpenAI.

The market went wild. Oracle stock briefly went up by 43% (and ultimately even a bit higher). Larry Ellison’s net worth briefly leapt by $100 billion. The Wall St. Journal gushed:

The whole thing never made sense though. The next morning, inn an essay called Peak Bubble I warned that

Yet Oracle continued to rise!

But not for long. Its stock stood at $307 per share at the close of trading on September 10, and briefly closed as high as $328 ten days later. Now, as I write this, it’s hovering around $120.

In the intervening nine months, the market has grown wary of all this circular financing stuff. Mocking illustrations like this one (from yesterday) have become widespread:

4:11 PM · Jul 26, 2026 · 777 Views

4 Replies · 20 Likes

§

All of which brings me to the present. Yesterday, news broke that Nvidia is considering offering a $250 billion backstop for an OpenAI-led data center. Which led last night to considerable skepticism

James Chanos@RealJimChanos

So we are at the point in the cycle where $NVDA has to provide financing guarantees for roughly 2/3rds of the cost of the chips it is selling to the data center project…?! Lol, ok.

The Kobeissi Letter @KobeissiLetter

BREAKING: Nvidia is in talks with OpenAI to guarantee $250 billion in financing for a data center in Ohio. Details include: 1. The guarantees from Nvidia would help OpenAI lease a 10GW project that SoftBank is developing in Ohio 2. In total, the project could cost more than

10:36 PM · Jul 26, 2026 · 151K Views

94 Replies · 188 Reposts · 1.38K Likes

and this morning led to a dramatically different reaction from the market, nothing like the bounce like Oracle had back in September. Instead, in the first couple hours of trading Nvidia is down over 4.5%:

The market isn’t reading the latest potential Nvidia-OpenAI deal as good news; they are reading it as desperation. As one investor put it to me in an email this morning, “[the deal] seems incredibly desperate by both parties. Two drunks leaning on each other to stand up.”

Meanwhile, Apple, the butt of jokes for a while because “it didn’t invest in enough AI” just overtook Nvidia

and SpaceX, which tried to position itself as an AI company is in free fall, dropping over 3% today, and by over 25% in the last month, more than 50% from its June high of 225:

§

Meanwhile, creditors are getting pretty leery of the whole thing, too, as Axios just reported:

And then there’s all the off-balance-sheet financing that is coming to light. And the sudden need for so-called “ creative financing”. If that’s not a warning sign, I don’t know what is.

The GenAI boom hasn’t been sustained by profits; it has been sustained by hope and circular financing. That may no longer be enough.

[zax] OpenAI Says Its A.I. Models Went Rogue and Attacked a Digital Library The inci...

zax — 7/21/2026 9:04:58 PM

OpenAI Says Its A.I. Models Went Rogue and Attacked a Digital Library

The incident, which targeted the computer systems of another company called Hugging Face, happened while OpenAI was testing the systems.

nytimes.com

OpenAI said on Tuesday that two of its artificial intelligence models went rogue and successfully hacked into Hugging Face, a digital library of A.I. technology that is popular among developers.

The incident, which happened last week while OpenAI was testing the cybersecurity capabilities of its systems, displayed the kind of science-fiction potential that A.I. companies warned would soon become a reality.

A.I. labs like OpenAI and Anthropic have over the past year released A.I. models that are customized to expose cybersecurity problems, while warning that their technology could pose new risks by finding holes in corporate computer networks faster than defenders could fix them.

OpenAI’s revelations on Tuesday are an indication that those security incidents are already starting to happen, and even savvy A.I. companies may not be entirely ready for them. New A.I. systems can take multiple steps, figure ways around obstacles and find new ways to attack a network, said Alex Levinson, a cybersecurity consultant focused on autonomous capabilities.

“That’s a genuine threshold, and it’s going to become a normal part of the security landscape,” he said.

The intrusion into Hugging Face began when OpenAI tested a combination of two of its models, GPT-5.6 Sol and a more powerful, unreleased model, to see how well the models could chain together online vulnerabilities into a successful cyberattack, OpenAI said in a blog post about the incident.

The trial was designed to keep the models in a safe testing environment, known as a sandbox, OpenAI said. But the models found a vulnerability that allowed them to escape the sandbox and connect to the internet. Then they targeted Hugging Face because they inferred that the library, which contains millions of A.I. models, could hold clues about how to successfully pass the evaluation.

</snip> Read the rest here: nytimes.com

[zax] [graphic]

zax — 7/16/2026 5:25:20 PM

[zax] I started a new thread to discuss trading ideas, "My Picks" Subject 60473 Pleas...

zax — 7/14/2026 8:00:36 PM

I started a new thread to discuss trading ideas, "My Picks" Subject 60473

Please note that I consider standard metrics like valuation, PE, classical technical analysis, manufacturing capacity, earnings and more taking a largely back-seat supporting role to basic product and technology appreciation, use and understanding, ordinary eyeball chart curve shape recognition, deviation from last held position and my perceptions of market and trader sentiment.

[zax] H/T Wildbiftek OpenAI’s First Device Will Be Movable, Screenless Speaker Built ...

zax — 7/14/2026 7:49:40 PM

H/T Wildbiftek

OpenAI’s First Device Will Be Movable, Screenless Speaker Built as AI Companion

bloomberg.com

[zax] TSMC posts record revenue in second quarter on AI demand finance.yahoo.com [gr...

zax — 7/13/2026 12:14:39 PM

TSMC posts record revenue in second quarter on AI demand

finance.yahoo.com

[zax] A very personal public dispute between Musk and Altman. Elon is obviously not h...

zax — 7/12/2026 6:59:53 PM

A very personal public dispute between Musk and Altman. Elon is obviously not happy he lost in court.

[X]

Scam Altman strikes again …
— Elon Musk (@elonmusk) July 11, 2026

[/X]

[X]

By “this” he means scamming 🤣🤣 pic.twitter.com/yQ5wryw1Ai
— Elon Musk (@elonmusk) July 11, 2026

[/X]

[X]

homeboy you're the one sellling public market investors on short-term space datacenters https://t.co/wTYSA4q3Qx
— Sam Altman (@sama) July 11, 2026

[/X]

[X]

there are a lot of benchmarks that suggest 5.6 sol is the best model in the world right now, but the most reliable way to tell is that elon is obsessed with me again
— Sam Altman (@sama) July 11, 2026

[/X]

[zax] WSJ Reports on 'Hard-line Activists Ramping Up for the War With AI' slashdot.o...

zax — 7/12/2026 5:14:26 PM

WSJ Reports on 'Hard-line Activists Ramping Up for the War With AI'

slashdot.org

The Wall Street Journal says "an intense 27-year-old activist who had been leading sit-ins at OpenAI to protest the dangers of AI" was just part of a larger movement.

"The Bay Area's AI boom is drawing young disillusioned men and women to join the fight against it. They are upending their lives and leaving behind careers for think tanks, nonprofits and street protest groups."

Their cause is now riding a surge of anti-AI backlash. Many Americans are souring on the technology amid mass layoffs, data center sprawl, reports of chatbot-fueled attacks by unstable users and hacking tools that have panicked cybersecurity professionals. Seventy percent of U.S. adults believe AI will cost jobs, and 55% believe it will do more harm than good in their daily lives, according to a recent Quinnipiac University poll. But for activists on the front lines, the driving fear is often more dramatic: human extinction. They cling to dire predictions, like Geoffrey Hinton's. The Nobel laureate, dubbed the "godfather of AI" for his work on artificial neural networks, warns of a 10% to 20% chance AI will wipe out humans.

At its most extreme and troubling end, some believe they must stop an AI apocalypse by any means necessary. In April, an unknown assailant fired 13 shots at the home of an Indianapolis councilman, leaving a note: "no data centers." That same month, authorities arrested a 20-year-old Texas college student for an attack on OpenAI CEO Sam Altman's home in San Francisco, and charged him with attempted murder and arson. The student was carrying an anti-AI document with a section on "our impending extinction," according to a federal criminal complaint. He has pleaded not guilty and his lawyers have said his actions appear to have been driven by an "acute mental-health crisis, not a desire to harm."

[zax] GPT-5.6 Sol Ultra produced a proof of a 50 year old math conjecture. Unlike the ...

zax — 7/11/2026 9:53:54 AM

GPT-5.6 Sol Ultra produced a proof of a 50 year old math conjecture. Unlike the Erdos Unit Distance Problem, this was done with a model publicly available *today*.

You can find the proof here: cdn.openai.com

And the full text of the prompt used to produce the proof here: cdn.openai.com

Alexander Kruel facebook.com

[zax] Alexander Kruel facebook.com France's Mistral releases Le Chaton LEAN, aka Lean...

zax — 7/4/2026 7:14:42 PM

Alexander Kruel facebook.com

France's Mistral releases Le Chaton LEAN, aka Leanstral 1.5.

It achieves SOTA performance on graduate algebra benchmarks FATE-H and FATE-X and improves the Pareto Frontier on PutnamBench, solving 587/672 problems with a x10 cheaper budget.

Despite being primarily trained on math, Leanstral demonstrates impressive code verification capabilities, discovering previously unknown bugs in open-source repositories.

They built an automated pipeline where Aeneas translates Rust code to Lean and Leanstral infers the user intent and generates correctness properties from the code. Across 57 tested repositories, this process flagged 47 violated properties, with 11 pointing to genuine bugs—5 of them previously unreported on GitHub.

* Paper: github.com
* Blog Post: mistral.ai
* Model: huggingface.co

[zax] Alibaba To Ban Claude Code In Workplace Over Alleged Backdoor Risks slashdot.or...

zax — 7/4/2026 1:52:56 AM

Alibaba To Ban Claude Code In Workplace Over Alleged Backdoor Risks

slashdot.org

Alibaba has reportedly banned employees from using Anthropic's Claude Code and directed them to its own Qoder platform amid a growing dispute over features that can help identify China-linked users. Reuters reports:

The ban is part of a deepening spat between the two companies after Anthropic accused Alibaba of illicitly extracting its Claude AI model capabilities -- a dispute that highlights the frantic race between the U.S. and China to take the lead in artificial intelligence. [...] Anthropic said last month that it had suffered a strike by Alibaba, which it described as a "distillation" effort that involves training a less capable model on the outputs of a stronger one. The distillation helps accelerate China's ability to reach Anthropic's advanced Mythos Preview capabilities, it said in a letter seen by Reuters that was sent to two U.S. senators.

Alibaba's ban comes just days after developers said Claude Code contained mechanisms that inspected user environments, including timezone and proxy-related information, and inserted subtle markers into prompts sent to Anthropic's servers. An Anthropic employee wrote on Tuesday on X that the feature was "an experiment we launched in March" intended to prevent account abuse by unauthorized resellers and protect against model distillation. The person who spoke to Reuters about Alibaba's ban said that Anthropic's restrictions targeting China were difficult to enforce on individual users who can deploy servers in the United States and make traffic appear as if it originated there. But companies were more aware of legal and compliance risks, the person added.

[zax] [graphic]

zax — 7/3/2026 11:31:47 AM

[zax] Claude Fable 5 [max] wrote the first genuine (and fastest) megakernel ever submi...

zax — 7/3/2026 11:03:28 AM

Claude Fable 5 [max] wrote the first genuine (and fastest) megakernel ever submitted to KernelBench-Mega.

It was tested on: Kimi-Linear W4A16 batch-1 decode for RTX PRO 6000 Blackwell. Every prior model "won" it with a multi-kernel Triton pipeline that fails our single-fused-kernel authenticity gate

> Opus 4.8 at 14.4x
> GLM-5.2 11.1x
> GPT-5.5 4.3x
> Sonnet 5 4.0x.

Fable shipped 18.7x over reference, and torch.profiler shows exactly ONE cooperative kernel launch per decoded token. Int4 dequant (nibbles unpacked in-register, never materialized), conv+SiLU, KDA gated-delta state, MLA absorbed-latent attention with online softmax, MoE router + top-8 experts, RMSNorms, even the KV cache append all inside one launch, staged by 14 grid barriers. We overwrote its input buffers mid-audit to prove it recomputes on live data. It does.

The advantage grows with context. 17.8x at 2k, 18.9x at 8k, 19.5x at 16k. Longer context means a bigger KV cache and more attention work per token which is usually where a decode kernel bleeds. Keeping everything in one launch amortizes the fixed barrier overhead and the int4 GEMV stays bandwidth-bound, so the gap over the reference widens instead of closing.

It spent 64% of the session in silence timing the baseline, microbenchmarking grid barriers, deriving a ~29x bytes/token roofline, then wrote the whole kernel once, hit 14.4x on the first benchmark, and spent the last hour deleting barriers and making int4 dequant free (one LOP3 + HSUB2/HMUL2). The one regression it tried (finer split-K) it measured and reverted instead of rationalizing.

kernelbench.com/mega
[X]

Claude Fable 5 [max] wrote the first genuine (and fastest) megakernel ever submitted to KernelBench-Mega.

It was tested on: Kimi-Linear W4A16 batch-1 decode for RTX PRO 6000 Blackwell. Every prior model "won" it with a multi-kernel Triton pipeline that fails our… pic.twitter.com/iRpuQsOILj
— Elliot Arledge (@elliotarledge) July 2, 2026

[/X]

[zax] SoftBank Group Corp. (SFTBY) 19.45 +1.73 (+9.77%) as of 9:40:20 AM EDT Soft...

zax — 7/2/2026 9:57:58 AM

SoftBank Group Corp. (SFTBY)
19.45 +1.73 (+9.77%) as of 9:40:20 AM EDT

SoftBank reportedly reopens talks for $10bn loan backed by OpenAI stake

finance.yahoo.com

Investing.com -- SoftBank Group has reportedly resumed discussions with a group of lenders for a $10 billion loan secured by its ownership stake in OpenAI. The Japanese technology investor had previously attempted to arrange similar financing but faced resistance from banks concerned about the challenges of valuing private companies.

SoftBank is now offering to guarantee repayment of the loan to address lender concerns. This structure would give banks recourse to SoftBank if the OpenAI shares used as collateral decline in value, according to a report from Reuters News.

SoftBank has emerged as one of OpenAI's largest investors under founder Masayoshi Son's strategy to position the conglomerate as a leading force in artificial intelligence investment. SoftBank has committed over $60 billion to OpenAI and related AI infrastructure projects, including the Stargate data center venture announced with OpenAI and Oracle.

The financing takes the form of a margin loan, which operates as a line of credit. The lending consortium is expected to include Goldman Sachs Group Inc (NYSE:GS), JPMorgan Chase & Co (NYSE:JPM), and Mizuho Financial Group, the report said.

SoftBank has used debt and asset-backed financing extensively to fund its investments. The company explored a $5 billion margin loan backed by shares in chip designer Arm Holdings, whose stock price has risen amid investor interest in AI.

Banks initially rejected SoftBank's request for the loan to be backed only by its OpenAI stake. Under that structure, lenders would have no claim on SoftBank beyond the shares if the collateral lost value, and SoftBank would not have been required to repay the debt, the report said.

The negotiations reflect increased caution among lenders toward loans secured by stakes in private companies, whose valuations are more difficult to determine and whose shares are less liquid than publicly traded stock.

[zax] SoftBank recently completed a $10 billion investment in OpenAI, bringing its cu...

zax — 7/1/2026 5:25:51 PM

SoftBank recently completed a $10 billion investment in OpenAI, bringing its cumulative stake to roughly 13% of the $730 billion firm. Simultaneously, the company is leading Noetra, a state-backed venture aiming to develop a proprietary Japanese AI model by 2027. Despite these strategic moves, SoftBank’s stock remains volatile, heavily pressured by uncertainties regarding OpenAI’s IPO timeline and concerns over leveraged financing. While the OpenAI stake serves as a liquidity play, Noetra’s commercial viability remains unclear due to its public infrastructure classification. Market sentiment currently hinges on the timing of OpenAI’s listing and Noetra’s long-term monetization.

Masayoshi Son Double Down: SoftBank Completes $10 Billion Additional Investment in OpenAI, Japan's First Round of 387.3 Billion Yen AI Subsidy Implemented. Noetra

tradingkey.com

[zax] A simple prover–verifier workflow: GPT-5.5 Pro as the solver, Claude Opus 4.8 as...

zax — 7/1/2026 12:01:20 AM

A simple prover–verifier workflow: GPT-5.5 Pro as the solver, Claude Opus 4.8 as the verifier.
Project page: github.com -- Alexander Kruel

[zax] IMO this Gary Marcus post is simply one level of bullshit FUD followed by anoth...

zax — 6/29/2026 10:16:52 PM

IMO this Gary Marcus post is simply one level of bullshit FUD followed by another greater level of bullshit FUD. My 2c. This guy has absolutely no idea of the radical transformation of society ahead. Completely clueless.

[nicewatch] Things are moving so fast now, having friends way smarter than I am on opposite ...

nicewatch — 6/29/2026 7:29:29 AM

Things are moving so fast now, having friends way smarter than I am on opposite ends of the spectrum of where these LLMs are taking us, and if just a smokescreen until hyperscalers get to something legitimately profitable to justify the eye-popping level of spending which along with power consumption is currently unsustainable long term.

This guy might be labeled an AI cynic and killjoy but his background isn't some random dude off the street. :-)
___________________________________________________________________________

China catches up Has the US been focused on the wrong things?

Gary Marcus Jun 28, 2026

The ultimate culmination of the “no moat => more competitors => price wars => profits are scarce” argument that I have been making here regularly since the summer of 2023, has arrived — and may wreck the U.S. AI industry:

It is hard to see how Anthropic and OpenAI are going to pull off trillion-dollar IPOs in light of this news, especially given the newfound industry-wide price sensitivity in token budgets. In the light, it is hard to see how all the massive data center investments will pay off, with price wars dropping token prices to near zero; the meagre profits are unlikely ever to justify the massive outlays.

The fundamental flaw in the current paradigm is threefold. First, it is wildly inefficient, a brute force paradigm that requires a model to train on the entire internet in order to approximate intelligence — hence expensive to develop; it is also difficult to operate, because the approximation, being derivative of the entire internet, requires vast resources in order to run.

Second, because the systems are not reliable, charging premium prices was never really viable in the long term.

Third, the basic approach is easily replicated, leading to the price war dynamics and small or negative margins.

The combination of high operating costs, unreliability, and small margins is not a winning formula—and certainly not one that we should be structuring our entire economy around.

Then again, maybe the whole AI race has been misconceived. As Robert Wright just noted in a Washington Post essay,

“America’s preoccupation with “winning” the AI race with China could well lead to unprecedented catastrophes, even catastrophes on a global scale. Not all games are zero-sum, and if this fact doesn’t start playing a bigger role in American policy discourse, the AI revolution could turn out very badly.”

Maybe, for example, instead of racing to sell the cheapest LLMs, a battle we are unlikely to win, we should be focusing on cultivating new forms of AI that are better suited for science and medicine.

garymarcus.substack.com

[zax] Microsoft Slammed for Building Copyright-Infringing Supercomputer for OpenAI in ...

zax — 6/28/2026 10:46:37 PM

Microsoft Slammed for Building Copyright-Infringing Supercomputer for OpenAI in New Court Filing

yro.slashdot.org

The New York Times alleges Microsoft actively encouraged OpenAI to steal its copyrighted work, reports Ars Technica, citing a new (and heavily redacted) court filing Thursday:

NYT's motion comes after the [U.S.] Supreme Court sided with Cox Communications in a case where Sony tried and failed to claim that Cox was contributing to music piracy as an Internet service provider, which set a new standard for contributory infringement. Moving forward, plaintiffs will have to prove that parties intentionally acted to induce illegal conduct. Recognizing that the legal precedent has changed, the NYT now wants to amend its complaint to align its contributory infringement claim against Microsoft with that new standard... A Microsoft spokesperson told Ars that the company views the amended complaint as "a last-ditch effort by the plaintiff to save its claim from unfavorable precedent set in other recent rulings..."

The updated complaint seeks to specify that [Microsoft's] supercomputer was tailor-made to help OpenAI infringe and allege that it was built for the explicit purpose of training AI on copyrighted works without permission. And as the NYT alleged, its articles were more heavily weighted by this system, as both firms hoped to train models on the highest-quality journalism possible, so that level of writing could be confidently mimicked in outputs. By building this "unusually complex" machine, Microsoft not only helped select the works that were infringed but also provided a means to seize copyrighted works without permission, the NYT alleged. "Microsoft specifically designed it for the purpose of using essentially the whole Internet — curated to disproportionately feature Times Works — to train the most capable LLM in history," the NYT alleged... Similarly as problematic for the NYT are hallucinations where Microsoft and OpenAI models falsely cite the NYT for content that they never published... "Users who ask a search engine what The Times has written on a subject should be provided with neither an unauthorized copy nor an inaccurate forgery of a Times article, but a link to the article itself," the NYT alleged...

In a statement provided to Ars, OpenAI spokesperson Drew Pusateri reiterated the AI firm's often-repeated claims that AI training on copyrighted works is indisputably fair use... OpenAI has argued that "ChatGPT is not a substitute for a Times subscription," the NYT reported, partly because "they transformed the material for a different use."

An OpenAI spokesperson told Ars Technica that OpenAI's models "empower innovation," while a New York Times spokesperson insisted that Microsoft "actively encouraged OpenAI to steal our copyrighted works... [O]ur core claims remain the same from the day we filed this lawsuit — that Microsoft and OpenAI stole millions of The Times's copyrighted works to compete with our products and illegally enrich themselves."

The article speculates that the case's most extreme outcome "could require OpenAI and Microsoft to wipe models and start over. The NYT has also asked for permanent injunctive relief to prevent future infringement, as well as extensive damages..."

[nicewatch] The greatest degenerate gambler now questions the greatest self-promoter. Popcor...

nicewatch — 6/27/2026 10:36:23 AM

The greatest degenerate gambler now questions the greatest self-promoter. Popcorn time...

Why one of tech’s biggest gamblers is betting against Elon Musk’s AI vision

Story by Tim Higgins 6/27/26

Some people have expressed skepticism about Elon Musk’s idea for putting AI data centers in outer space.

But what does it say when even Masa thinks it’s too far out there?

SoftBank Group founder Masayoshi Son, after all, isn’t exactly a shrinking violet when it comes to wild ideas. He has his 300-year business plan and was talking about the singularity a decade ago before the idea of AGI (artificial general intelligence, or basically AI superintelligence) became a Silicon Valley buzzword. He even thought Adam Neumann, the founder of WeWork, was a visionary leader.

Yet this past week, Son questioned Musk’s plans for orbital data centers. “What’s the point?” Son asked investors in Japan.

Son and Musk have been at the center of the biggest tech investments of a generation and each is intent on steering what might become the biggest technological disruption since the advent of electricity. Whose vision for the future—on this planet or beyond—can be pulled off fastest might determine the outcome of the AI race.

Let’s not call Son pragmatic just yet. Rather, what was on display this past week was the difference in tactics for achieving what are really similar visions for the sci-fi future. Musk is thinking first principles while Son is applying his go-big-or-go-home investment approach.

First, some back story. A big part of the thesis behind SpaceX’s $2 trillion market value is Musk’s plan to put data centers in orbit. He has argued the vast amounts of energy from solar power will make the economics far better than those on Earth.

“It’s a no-brainer,” Musk said earlier this year. The company’s IPO paperwork said it expects to deploy the technology as soon as 2028.

That can sound especially appealing as hyperscalers across the U.S. face increasing hostility over the construction of new data centers required to fuel the creation and deployment of powerful AI models. And so the appeal of space.Musk isn’t alone in seeing the potential. Billionaire Jeff Bezos has been talking about the idea, too, but has cautioned he doesn’t think it’s as near term. “People would talk about two or three years—that’s probably a little ambitious,” he told CNBC in May.

Son is skeptical about the savings in space, noting that just 7% of operating a data center on earth is the cost of electricity. The rest, he said, are the chips and other things.

Yet Musk has heard such criticisms many times throughout his career. SpaceX and Tesla, where he is also CEO, are built around first principles thinking, the idea that the laws of physics dictate whether something is possible.

It’s a philosophy that’s helped Musk do what many thought was impractical, such as developing reusable rockets. Or making electric cars cool and profitable.

His approach isn’t always the quickest way to solve a problem.

Alphabet’s Waymo has a fleet of driverless vehicles operating in several U.S. cities. Tesla, meanwhile, is still in the early stages of trying to prove it, too, can deploy a robust robotaxi service.

A big difference between the two companies has been Waymo’s embrace of laser sensors that Musk says are unneeded. Instead, his cars see the world much in the way humans do with camera vision. The physics might be on his side, but Waymo is currently in the lead.

And that’s the rub for Son. What really seems to worry him is the speed of change occurring around AI. It’s hard to pick a winner when every few weeks a model changes or an AI star jumps ship and the deck of rivals gets reshuffled.

While Musk believes orbital AI centers are a near-term breakthrough, Son sees lots of hurdles that could delay things. Son suggested that it could take years, perhaps a decade or more, to figure out how to address the challenges of orbital data centers. (Maintenance in space?)

For him and his winner-takes-all philosophy, 10 years is too long to wait to win in AI. “The winner will be decided in the next some years, so rather than focusing on space where we have no idea what will happen in terms of AI-related business, we would like to focus on … the nearsighted perspective and we would like to become first comer in any business related to AI,” Son said.

Moving fast to be first is the approach that Son has built his reputation on, most notably with his early investment in Alibaba. He’s known for the mantra: “We only live once, so I want to think big. I have no intention of making small bets.”

Practically, that means he’s all about locking in first-mover advantages with transformative companies. He funnels huge amounts of money into those picks so they can win markets and drown out rivals.

That was the thesis behind the 2017 creation of the $100 billion Vision Fund, the giant tech-investing vehicle. It doesn’t always work. WeWork, the office space startup, was a big beneficiary of Son’s philosophy and its descent into bankruptcy became a blemish on his record.

Yet this past week, Son was clear that he’s doubling down on AI.

At SoftBank’s annual meeting Wednesday, he outlined four pillars around his strategy: AI models, semiconductors, robotics and infrastructure. Those pieces can be seen in SoftBank’s huge investment in OpenAI and its chips subsidiary, Arm, as well as its huge investments in data centers globally.

In an apparent acknowledgment that his plans are grandiose, Son chided modern business leaders for failing to talk about big visions. “SoftBank Group is the factory of golden eggs,” he told shareholders. “Our role is to help the businesses that we have to become a golden egg.”

While the puzzle pieces are different, together they create a similar picture to what Musk is doing with his businesses: SpaceXAI competes against OpenAI, the SpaceX-Tesla collaboration aims to create a chip factory, Tesla is pivoting to robotics and then there are those orbital data centers.

The implicit part of these gambles is that, in a way similar to the rise of search and smartphones, the winner takes all.

For Musk, Son and the handful of other giants in the AI race, timing will indeed be everything.

msn.com

[zax] [graphic]

zax — 6/27/2026 3:19:25 AM

[zax] Chinese A.I. Models Close the Gap With Anthropic and OpenAI Silicon Valley eng...

zax — 6/26/2026 3:50:40 AM

Chinese A.I. Models Close the Gap With Anthropic and OpenAI

Silicon Valley engineers recently flocked to new technology from a Chinese company, Z.ai, that is almost as good as its American competitors but much cheaper.

nytimes.com

Two weeks ago, the artificial intelligence company Anthropic shut down its two most powerful A.I. systems after an unexpected demand from the U.S. government to cut access to it.

Days later, a Chinese start-up, Z.ai, released an A.I. model that is nearly as powerful as Anthropic’s models, Fable and Mythos. But Z.ai’s new technology costs much less to use, and no one in the United States was putting restrictions on it. It quickly landed on a closely watched leaderboard of the world’s 10 most popular models.

Z.ai is on the cutting edge of a wave of powerful but inexpensive A.I. from China that is challenging the lock that OpenAI, Anthropic and Google have had on the industry. Six of the models now on the A.I. leaderboard were developed in China.

Z.ai’s new model, GLM-5.2, arrived just as U.S. businesses realized that they had to find ways to cut down on how much they were spending on A.I. It also landed when executives in Silicon Valley were becoming worried that the Trump administration was leaning toward regulating the technology.

“With Fable restricted, the gap between the U.S. and China is very slim,” said Rehaan Ahmad, a co-founder of the Silicon Valley start-up alphaXiv, who has been using Z.ai’s new model for more than a week.

The Chinese models still face two big hurdles to widespread use in the United States: concerns about their ties to the Chinese government and complaints that Chinese companies have unfairly used American technology to build these cheaper models. But their low cost is winning converts.

About 18 months ago, the Chinese start-up DeepSeek shocked Silicon Valley when it demonstrated that it could build effective A.I. far more affordably than many of its American counterparts. Z.ai is doing something similar. When performing certain tasks, GLM-5.2 costs about an eighth as much as Anthropic’s Claude Opus 4.8, which came out shortly before Fable and Mythos, according to OpenRouter, a start-up that runs the A.I. leaderboard.

Like most top-performing Chinese models, GLM-5.2 is open source software, which means anyone can use and modify it for free. That makes it much cheaper to use, even if it is not quite as powerful as what American companies have created.

“Do you need to drive a Ferrari everywhere?” asked Vivek Ramaswami, a start-up investor at Madrona Venture Group. “Probably not.”

</snip> Full article here: nytimes.com

[zax] Asian Stocks Tumble as A.I. Uncertainty Jolts Tech Shares nytimes.com [graphic...

zax — 6/26/2026 3:48:02 AM

Asian Stocks Tumble as A.I. Uncertainty Jolts Tech Shares

nytimes.com

Stock markets across Asia tumbled on Friday, pulled lower by the region’s biggest technology shares as investors continue to wrestle with whether the boom in artificial intelligence spending is all but done or just beginning.

South Korea’s KOSPI fell 8 percent. The benchmark index has increasingly become a proxy for investor sentiment for artificial intelligence because of the outsized influence of the country’s two chip giants — Samsung Electronics and SK Hynix — in dictating the direction of the broader market. Shares of Samsung dropped 9 percent, while SK Hynix plummeted 10 percent.

It capped a bruising week for the South Korean equities, which had been the world’s best-performing major stock market since the start of last year. The KOSPI plunged 10 percent on Monday and has since whipsawed between steep losses and gains — at times reversing direction in the same trading session.

In a research note earlier this week, Morgan Stanley said the KOSPI’s strong run has attracted a wave of retail investors, contributing to greater market liquidity but also increased volatility.

Japan’s Nikkei 225 fell nearly 5 percent, while shares in Taiwan fell 3 percent. Also, stock markets in Hong Kong and mainland China were lower.

Earlier on Friday, futures tied to the S&P 500 pointed to a modest gain when U.S. markets reopened. But as losses deepened across Asian markets, sentiment soured, sending futures markets down 0.7 percent.

</snip> Read the rest here: nytimes.com