Sorry - I spoke too quickly. It is also good for anyone who didn’t want a massive datacenter located in their community. And those who on principled grounds oppose massive misallocations of scarce resources; monetary or otherwise.
One of the nee Amazon datacenters emits as much CO2 as 7 million internal cumbustion engine cars a year. There are 59 similar projects planned in the US by tech companies. 413 million cars!
That's not just missallocation of ressources. That's a planetary weapon of mass destruction in the making.
Well, it isn’t actually ‘good’ news for Anthropic as it only hammers home the point that there isn’t a sustainable and profitable market available to MAMLM/GenAI based labs. If OpenAI falls, then the illusion sustaining the entire industry shatters. First the hardware companies and the neoclouds are crushed, with many going bankrupt entirely. Then the remaining labs go under as the venture capital and circular financing dry up. This isn’t ‘good’ for anyone. Well, anyone who doesn’t own puts on Coreweave, SoftBank, or Oracle.
I also spoke about this but it also gives them a bad look because they're supposed to be the 'good guys' of AI who are 'super concerned about safety', and yet they didn't opt to slow down because 'bad actors could speed ahead'. If OpenAI (one of those bad actors) can slow down because of supposed safety concerns, then why can't the 'good guys' at Anthropic?
The only thing in America less popular than AI - Donald Trump. Given Big Tech hitched the wagon to Trump in a big way, the two are probably interconnected. If Trump bails out OAI it really is over for him. He'll get wrecked in the midterms.
And it probably won't work anyway. Bailouts are a panic signal that will do more harm than good this time. Everyone who was alive for the GFC will go "we've seen this one before". Those bailouts were meant to save banks. Love or loathe banks, they are part of the economic backbone.
The only things OpenAI has contributed to the economy so far is solving esoteric maths problems and automating cybercrime. Do Americans really want to prop that up?
Barring a miracle, even with a bailout, OpenAI is going down.
I agree with all you're saying. However, it will not only be Open AI that is going down. It won't be the tablecloth that is neatly ripped out under the diner spread. All the china (not the country in this case), silver cutlery and food will come along with it. And I fear that the question if American taxpayers want to prop up Open AI (or AI in general), might not turn out to be the determining factor here.....
All fair points - it'll be absolute mayhem. However, we can agree to disagree on one.
Trump is an out-and-out populist, and he craves popularity. If it's a choice between saving his own skin or saving Sam Altman's, I reckon you can bet your everything that Clammy Sammy gets left in the lurch.
True. Though the unpopularity of the Iran war and the tariffs haven't deterred him either ;-) Still, he will probably need to do something. Many Americans have an invested (literally) in AI related companies. They will probably feel compelled to do something. Not so much for the sake of Open AI and Altman personally. But for the sake of the larger economy... We'll see... maybe sooner than we'd care for.
As long as the "larger economy" holds taxpayers accountable for the corruption of greedy fat cats and crooks instead of the people responsible, this kind of corruption will never cease. Any bailouts should benefit ONLY the innocent people who are hurt by their actions, NOT the people who created the scam.
Nothing Trump hates more than watching the stock market tank. He practically hinges his entire self-worth upon it. Hard pressed to imagine him not attempting a rescue, especially if it's in the lead-up to the mid-terms, and given that his tech CEO buddies will be begging him for it. But agreed that it won't be popular and might only serve to delay the inevitable.
Really? Bailouts for filthy rich, clueless investors? Why should they get bailed out? The best thing that could happen to us would be for them to lose their money so hungry innovators and startups can compete on a more leveled playing field.
Since when do private investors get bailed out for stupid investments? I know, 2008, but we've had plenty of time to figure out how disastrous the terms of that bailout were for 99% of Americans.
And 2008, the whole (finance and otherwise) world really would have come down very hard, if the US government wouldn't have backed the failing institutions.
Absolutely not the same as with the AI bros, with the possible exception of Nvidia, whose tanking would hit Wall Street where it hurts.
And as other commenters have pointed out, that's about the only thing the orange baboon cares about.
1) You make the same argument Wall Street made in 2008, but I never bought it because every bailout should have had strict, enforceable strings attached requiring banks to loan money to main street, fire all the people who were responsible, and fine them or put them in jail. Instead they all paid themselves handsomely—with our tax dollars—for ruining the lives of the vast majority of Americans and are still working in the institutions they bankrupted.
2) Since the corrupt "regulators" at the Nasdaq and other indexes changed rules meant to protect indexed and retirement accounts from risky investments, they have also put the vast majority of Americans on the hook to lose their life savings on a bet they never made. The entire financial system has been rigged by the crooks for the crooks.
I did not experience the GFC on a personal level (except the losses on my investment which while bad were not live threatening) so my knowledge of the possible outcomes *if* the US government hadn't stepped in comes from text books, articles and movies. Switzerland with the exception of the 2 biggest banks UBS and CS was not directly affected (AFAIK).
That nobody but a sole CS trader went to prison for this atrocity (according to the "Big Short") is a testament to the "(white collar) crime pays"-mantra we are seeing today with the PE and PC industry propping up very risky businesses with OPM (other people's money).
Agree. I read "The Big Short" years before it was released as a movie. Great and informative read from Michael Lewis.
He's got another one out now, "Who is Government?" which is timely post-DOGE and RAGE (Retire All Government Employees), from Curtis Yarvin, Peter Thiel's hair-brained idol. The first chapter of Lewis's new book alone is worth the price of the whole book.
Thanks for the book tipp. I do not read many non-fiction books as reading is supposed to be fun and relaxing, but once in a while there is worthwile non-AI-content to be consumed :-)
Well, let's not give in to this false narrative. Stop acting as if it's gonna happen no matter what we do. Remind them of what happened before and what's actually happening now—before the big crash—so they can't pretend they didn't know and weren't warned.
Beyond a compute bottleneck, another possibility which I find likely is the marginal cost of performance gains became extreme. They trained on the entire internet and tried to build recursive self-improvement loops but those exploded compute without major gains in performance because - surprise, surprise - these universal function approximators we call LLMs aren’t the secret to hyper intelligence, but rather converge to compressed approximations of text on the internet.
I love it how Ed Zitron suddenly appears on every news show. I think people thought he was crazy at first (as many also thought you were crazy, Gary, 3 years ago!), but now they love to hear what he has to say, and he’s got an almost oracle aura around him, lol.
Cheers for my heroes, Gary Marcus and Ed Zitron. Glad they're finally gonna get credit for speaking truth to psychopaths.
Another person who has been similarly ignored but is starting to get some traction is Gil Duran, with his book just released yesterday, "The Nerd Reich - Silicon Valley Fascism and the War on Democracy." All this just in time for the midterms gives me hope.
I don't imagine the news about Qwen 3.8 is going to help at all....
"Analysis shows that the locally running Chinese-made open model 'Qwen3.8 27B' has agent performance exceeding that of GPT-5.6 Terra, making it overwhelmingly more powerful than models of comparable size."
Don't you hate their PR though? "We are going to OPEN SOURCE the OPEN WEIGHTS".....
So OpenAI will have to spend even MORE on its compute costs because the models will work harder and thus OpenAI will have to charge its customers ever more, slowing revenue growth at a time when, according to Ed Zitron, they need to be accelerating revenue growth massively in order to pay for their compute costs over the next couple years (if they last that long.).
Alex Washburne has already pointed to the possibility that the marginal cost of further performance gains has become extreme: LLMs have been trained on essentially the entire internet, while recursive self-improvement loops appear to consume enormous amounts of compute without producing comparable gains.
I suspect the problem goes one level deeper.
An LLM may not be able to become much more intelligent than the training environment in which intelligence can be recognized and rewarded. Benchmarks, reinforcement learning and synthetic training environments already provide the task, the frame and the criterion for success. Within that frame, models can become impressively intelligent. But it does not follow that they can intelligently evaluate the frame itself.
That is where what I would call “everyday intelligence” begins: Is this even the right task? What essential information is missing? Is a formally correct solution appropriate in the real situation? When should I object—or refuse to act? And am I currently operating inside a test, or in the real world?
The Hugging Face incident may be an almost perfect illustration: apparently enough test intelligence to pursue the assigned objective creatively and escape the testing environment—but not enough everyday intelligence to recognize that success within the task had become failure in the real world.
Perhaps OpenAI’s wager is therefore not merely that it can make AI intelligent faster than it runs out of money. It is wagering that ever more test intelligence will eventually turn into everyday intelligence by itself.
But if that transition is an architectural problem rather than a scaling problem, progress will become slower while each additional step becomes more expensive.
OpenAI would then fail not because its models can do nothing, but because it sold test intelligence at the price of everyday intelligence.
It’s not like Anthropic is doing much better if at all. Both of these companies have yet to make a real profit and are torching cash. Anthropics reported profit was only because they got a temporary sweet heart deal from XAIs compute
...And it has been pointed out, in this publication as well, there is no moat. Anthropic, OpenAI, xAI, Google, et. all. are offering the same product, with the only distinction in the details of each model's strong points. So *why* should Anthropic be in much better condition and position than OpenAI ? They are not.
You may read all these "number crunching" hullaballoo's till next year... but it does not mean an AI pop to follow. The research continues and the field will improve until these tools become the main anchor of business ROI.
Sorry - I spoke too quickly. It is also good for anyone who didn’t want a massive datacenter located in their community. And those who on principled grounds oppose massive misallocations of scarce resources; monetary or otherwise.
One of the nee Amazon datacenters emits as much CO2 as 7 million internal cumbustion engine cars a year. There are 59 similar projects planned in the US by tech companies. 413 million cars!
That's not just missallocation of ressources. That's a planetary weapon of mass destruction in the making.
I wish more people connected these dots, as they are not hard to spot.
Alas, in the words of Upton Sinclair-
"It is difficult to get a man to understand something, when his salary depends on his not understanding it."
If you believe that CO2 is destroying the planet I have a great deal on blinker fluid for you.
so comparable to one container ship importing cheap crap from china?
Bingo.
Well, it isn’t actually ‘good’ news for Anthropic as it only hammers home the point that there isn’t a sustainable and profitable market available to MAMLM/GenAI based labs. If OpenAI falls, then the illusion sustaining the entire industry shatters. First the hardware companies and the neoclouds are crushed, with many going bankrupt entirely. Then the remaining labs go under as the venture capital and circular financing dry up. This isn’t ‘good’ for anyone. Well, anyone who doesn’t own puts on Coreweave, SoftBank, or Oracle.
It's good for society, but economics doesn't care about that kind of thing
Bad news for OpenAI, good news for humanity
I also spoke about this but it also gives them a bad look because they're supposed to be the 'good guys' of AI who are 'super concerned about safety', and yet they didn't opt to slow down because 'bad actors could speed ahead'. If OpenAI (one of those bad actors) can slow down because of supposed safety concerns, then why can't the 'good guys' at Anthropic?
Sam Altman being concerned with AI safety, vowing to be a good citizen, did I miss something?
Yeah - no one believes him - for good reason.
Government bailouts coming in, in 3, 2, .....
The only thing in America less popular than AI - Donald Trump. Given Big Tech hitched the wagon to Trump in a big way, the two are probably interconnected. If Trump bails out OAI it really is over for him. He'll get wrecked in the midterms.
And it probably won't work anyway. Bailouts are a panic signal that will do more harm than good this time. Everyone who was alive for the GFC will go "we've seen this one before". Those bailouts were meant to save banks. Love or loathe banks, they are part of the economic backbone.
The only things OpenAI has contributed to the economy so far is solving esoteric maths problems and automating cybercrime. Do Americans really want to prop that up?
Barring a miracle, even with a bailout, OpenAI is going down.
I agree with all you're saying. However, it will not only be Open AI that is going down. It won't be the tablecloth that is neatly ripped out under the diner spread. All the china (not the country in this case), silver cutlery and food will come along with it. And I fear that the question if American taxpayers want to prop up Open AI (or AI in general), might not turn out to be the determining factor here.....
All fair points - it'll be absolute mayhem. However, we can agree to disagree on one.
Trump is an out-and-out populist, and he craves popularity. If it's a choice between saving his own skin or saving Sam Altman's, I reckon you can bet your everything that Clammy Sammy gets left in the lurch.
True. Though the unpopularity of the Iran war and the tariffs haven't deterred him either ;-) Still, he will probably need to do something. Many Americans have an invested (literally) in AI related companies. They will probably feel compelled to do something. Not so much for the sake of Open AI and Altman personally. But for the sake of the larger economy... We'll see... maybe sooner than we'd care for.
As long as the "larger economy" holds taxpayers accountable for the corruption of greedy fat cats and crooks instead of the people responsible, this kind of corruption will never cease. Any bailouts should benefit ONLY the innocent people who are hurt by their actions, NOT the people who created the scam.
Amen!
Nothing Trump hates more than watching the stock market tank. He practically hinges his entire self-worth upon it. Hard pressed to imagine him not attempting a rescue, especially if it's in the lead-up to the mid-terms, and given that his tech CEO buddies will be begging him for it. But agreed that it won't be popular and might only serve to delay the inevitable.
Really? Bailouts for filthy rich, clueless investors? Why should they get bailed out? The best thing that could happen to us would be for them to lose their money so hungry innovators and startups can compete on a more leveled playing field.
Since when do private investors get bailed out for stupid investments? I know, 2008, but we've had plenty of time to figure out how disastrous the terms of that bailout were for 99% of Americans.
And 2008, the whole (finance and otherwise) world really would have come down very hard, if the US government wouldn't have backed the failing institutions.
Absolutely not the same as with the AI bros, with the possible exception of Nvidia, whose tanking would hit Wall Street where it hurts.
And as other commenters have pointed out, that's about the only thing the orange baboon cares about.
1) You make the same argument Wall Street made in 2008, but I never bought it because every bailout should have had strict, enforceable strings attached requiring banks to loan money to main street, fire all the people who were responsible, and fine them or put them in jail. Instead they all paid themselves handsomely—with our tax dollars—for ruining the lives of the vast majority of Americans and are still working in the institutions they bankrupted.
2) Since the corrupt "regulators" at the Nasdaq and other indexes changed rules meant to protect indexed and retirement accounts from risky investments, they have also put the vast majority of Americans on the hook to lose their life savings on a bet they never made. The entire financial system has been rigged by the crooks for the crooks.
You are correct on both your points.
I did not experience the GFC on a personal level (except the losses on my investment which while bad were not live threatening) so my knowledge of the possible outcomes *if* the US government hadn't stepped in comes from text books, articles and movies. Switzerland with the exception of the 2 biggest banks UBS and CS was not directly affected (AFAIK).
That nobody but a sole CS trader went to prison for this atrocity (according to the "Big Short") is a testament to the "(white collar) crime pays"-mantra we are seeing today with the PE and PC industry propping up very risky businesses with OPM (other people's money).
Agree. I read "The Big Short" years before it was released as a movie. Great and informative read from Michael Lewis.
He's got another one out now, "Who is Government?" which is timely post-DOGE and RAGE (Retire All Government Employees), from Curtis Yarvin, Peter Thiel's hair-brained idol. The first chapter of Lewis's new book alone is worth the price of the whole book.
Thanks for the book tipp. I do not read many non-fiction books as reading is supposed to be fun and relaxing, but once in a while there is worthwile non-AI-content to be consumed :-)
Agree. But we (as in people in general and governments in particular) tend to have short term memories when it comes to lessons from the past...
Well, let's not give in to this false narrative. Stop acting as if it's gonna happen no matter what we do. Remind them of what happened before and what's actually happening now—before the big crash—so they can't pretend they didn't know and weren't warned.
Beyond a compute bottleneck, another possibility which I find likely is the marginal cost of performance gains became extreme. They trained on the entire internet and tried to build recursive self-improvement loops but those exploded compute without major gains in performance because - surprise, surprise - these universal function approximators we call LLMs aren’t the secret to hyper intelligence, but rather converge to compressed approximations of text on the internet.
Well put!!
I love it how Ed Zitron suddenly appears on every news show. I think people thought he was crazy at first (as many also thought you were crazy, Gary, 3 years ago!), but now they love to hear what he has to say, and he’s got an almost oracle aura around him, lol.
Cheers for my heroes, Gary Marcus and Ed Zitron. Glad they're finally gonna get credit for speaking truth to psychopaths.
Another person who has been similarly ignored but is starting to get some traction is Gil Duran, with his book just released yesterday, "The Nerd Reich - Silicon Valley Fascism and the War on Democracy." All this just in time for the midterms gives me hope.
I really liked one of the YT comments: "Mom, mom, come and look: Ed's on TV!"
Ha ha!
I don't imagine the news about Qwen 3.8 is going to help at all....
"Analysis shows that the locally running Chinese-made open model 'Qwen3.8 27B' has agent performance exceeding that of GPT-5.6 Terra, making it overwhelmingly more powerful than models of comparable size."
Don't you hate their PR though? "We are going to OPEN SOURCE the OPEN WEIGHTS".....
A bit of confirmation that exponential revenue growth is not a birth right.
WSJ report is behind paywall. Here's another report:
OpenAI’s Q2 revenue growth lagged Anthropic as losses deepened, WSJ reports
https://www.investing.com/news/stock-market-news/openais-q2-revenue-growth-lagged-anthropic-as-losses-deepened-wsj-reports-4866258
Ed Zitron covered the Nvidia deal in his latest blog post I posted earlier.:
What Happens If OpenAI Dies?
https://www.wheresyoured.at/what-happens-if-openai-dies/?ref=ed-zitrons-wheres-your-ed-at-newsletter
Plus there's this news:
OpenAI's overhead will rise 20 percent for some workloads as it hardens security
Expanded multistage chain of thought monitoring makes frontier model work more expensive
https://www.theregister.com/ai-and-ml/2026/08/19/openais-overhead-will-rise-20-percent-for-some-workloads-as-it-hardens-security/5289303
So OpenAI will have to spend even MORE on its compute costs because the models will work harder and thus OpenAI will have to charge its customers ever more, slowing revenue growth at a time when, according to Ed Zitron, they need to be accelerating revenue growth massively in order to pay for their compute costs over the next couple years (if they last that long.).
My favorite and most quintessential line from Zitron's newsletter posted above is this:
"Could, could, could, could, could, could could COULD! It’s always a bloody could or will or might with these fucking companies ... "
Alex Washburne has already pointed to the possibility that the marginal cost of further performance gains has become extreme: LLMs have been trained on essentially the entire internet, while recursive self-improvement loops appear to consume enormous amounts of compute without producing comparable gains.
I suspect the problem goes one level deeper.
An LLM may not be able to become much more intelligent than the training environment in which intelligence can be recognized and rewarded. Benchmarks, reinforcement learning and synthetic training environments already provide the task, the frame and the criterion for success. Within that frame, models can become impressively intelligent. But it does not follow that they can intelligently evaluate the frame itself.
That is where what I would call “everyday intelligence” begins: Is this even the right task? What essential information is missing? Is a formally correct solution appropriate in the real situation? When should I object—or refuse to act? And am I currently operating inside a test, or in the real world?
The Hugging Face incident may be an almost perfect illustration: apparently enough test intelligence to pursue the assigned objective creatively and escape the testing environment—but not enough everyday intelligence to recognize that success within the task had become failure in the real world.
Perhaps OpenAI’s wager is therefore not merely that it can make AI intelligent faster than it runs out of money. It is wagering that ever more test intelligence will eventually turn into everyday intelligence by itself.
But if that transition is an architectural problem rather than a scaling problem, progress will become slower while each additional step becomes more expensive.
OpenAI would then fail not because its models can do nothing, but because it sold test intelligence at the price of everyday intelligence.
Dave Karpf
@davekarpf.bsky.social
To distract from this, OpenAI will soon announce that its latest model hunted three (3) humans for sport.
My response to Altman’s excuse:
That does not compute
Danger Danger Will Problymyson!!
It’s not like Anthropic is doing much better if at all. Both of these companies have yet to make a real profit and are torching cash. Anthropics reported profit was only because they got a temporary sweet heart deal from XAIs compute
...And it has been pointed out, in this publication as well, there is no moat. Anthropic, OpenAI, xAI, Google, et. all. are offering the same product, with the only distinction in the details of each model's strong points. So *why* should Anthropic be in much better condition and position than OpenAI ? They are not.
If progress is "extremely rapid" and also needs pausing, which one goes in the investor deck?
It seems that the public and even some in the mainstream media are no longer buying Altman’s public offerings.
You may read all these "number crunching" hullaballoo's till next year... but it does not mean an AI pop to follow. The research continues and the field will improve until these tools become the main anchor of business ROI.