Eighteen months from the most valuable startup in history to government bailout candidate. That's the speed record nobody wanted to set.
The SpaceX GPU story is the detail that tells you where this is heading. last year the chips were too valuable to sell. this year they're inventory you can't find enough tenants for. when the hoarder becomes the landlord, the thesis has changed and nobody sent the memo to the market cap.
The government stake framing is doing the quiet work here. "Investment" is the word you use when the numbers don't support "commercial return" but the politics don't allow "bailout." the playbook is identical to 2008 bank rescues, just wearing a hoodie instead of a suit. privatise the gains in the $150 billion valuation round, socialise the losses when the revenue doesn't show up. if OpenAI had a path to profitability it wouldn't need the government in the cap table. the cap table is the confession.
I’m not knocking it. Using LLMs as a writing tool can be very useful for certain situations and with good instructions you can completely change the writing style and beat that cliche “AI feel” out of it to fit your tastes.
It demonstrates laziness, lack of imagination, inability to put together a coherent argument and/or illiteracy when one uses it for comments on a blog that really are not that difficult to put together without using a chatbot.
Lots of tells - the speed record nobody wanted to set - is classic AI
The sentence structures e.g. "if OpenAI had a path to profitability it wouldn't need the government in the cap table. the cap table is the confession."
However the fact that the comment is AI generated doesn't entirely make it wrong
"Investment" is the word you use when the numbers don't support "commercial return" but the politics don't allow "bailout."
Is an entirely valid point. It's wrong because what the government should do is buy the assets it wants from the bankruptcy firesales... but it is exactly what governments have done historically
Because of the post training process consumer facing LLMs have a relatively strong default writing style in the absence of further instruction or context. It’s amazing to me the number of people who use default free ChatGPT and then claim it’s “dumb” or hallucinates or whatever. Absurdly under-appreciated the degree to which the outputs of this systems varies enormously between naive and highly technical users.
Thanks to Sam A for showing the US taxpayers possible dividends while eying at her pockets. How about socialization of the losses? With capitalists like this, who needs socialists?
I have been doing investing for almost 10 years and for the first time my broker has been "informing" me about an upcoming IPO. I'm happy to lose my own money but it won't be by investing in Space X stock.
I heard an author speaking about AI who casually asserted the cost of "intelligence" is dropping globally. I was unable to externalize the poetic irony.
Maybe a dead cat will bounce on Monday. Let's see what just happens on Tuesday.
The fact that Trump is talking enthusiastically about the American government investing in AI... is not a good look. The Chinese must really be laughing.
If you have a bank that's too big to fail, you pump in money and it carries on working.
If you have an AI company that's pursuing a completely unworkable technology, pumping in any amount of money isn't going to help. You're even more doomed.
The US economy just ran into an epistemological brick wall. You can't build "intelligence" by parsing language strings, however much compute you throw at the problem and how many texts you scrape. Any competent philosopher will tell you this.
Is there a precedent in history when an entire industrial economy has literally been based on hallucinations?
To me, this AI market slump isn’t a surprise it’s a predictable leadership failure. I’ve watched platforms optimise blindly for pure scale, assuming bigger always meant better. Instead, we willingly bypassed safety boundaries and ignored the real societal costs. We prioritised algorithmic hype over human utility, and now the system is finally breaking. I wish we had more news about things that really bring value and improve our lives. Not the typical cheering on how many people lost due to automation, how a certain model can code better than a person. This is all an hallucination from certain players. We need t focus in reality and not in "valuations".
The job numbers were unexpectedly high. Jobs numbers are released on the first Friday of every month. What does that mean?
The job of the Federal Reserve is to keep price stability and low unemployment. If job numbers are unexpectedly high, the likelihood of interest rate increases goes up. Higher interest rates make it harder for businesses to be profitable. AI has a lot of debt, and it is more sensitive to interest rates.
AI 'Tokens' need to be 'spent' on the highest ROI. I see AI’s initial generosity as generating publicity and investor interest, and as a way to determine where the highest ROIs are located so they can focus their efforts on those specific industries. AI was never going to give us fully automated communism. The lump-of-labor fallacy has repeatedly shown that a larger economy generates more work.
Do not be surprised if prices bounce back on Monday. Market moves on the first Friday of every month are expected and unrelated to AI.
Respectfully, I think you are so anti AI that it does bias your writing. I agree with your take on private companies getting bailed out. Cronyism, and socialism are things I don't want either. But that doesn't mean all things AI are bad. Writing code has changed forever and one person can be as productive as two or more in some cases. Yes, we're overvaluing AI companies, but the value isn't zero either. After the bubble bursts, we'll still write most code using AI. The pace of software writing is forever changed. That said, I enjoy your takes. You're thought provoking. Thanks for what you do.
You missed more than one. Gary is not and never has been anti AI. I'm really disgusted with people making that accusation simply because he's pointed out what is now undisputed truth, that LLMs and scaling *alone* will never achieve AGI. It's my understanding that Claude *is* combining LLMs with symbolic tools. People who try to shoot the messenger of facts they don't want to face would do well to examine their own mental reflexes and thought patterns before attacking ethical and knowledgeable messengers like Gary Marcus.
🤔- modelling ai on a version of the PFC/Limbic concept of the human brain (RL/DL, search/value splits) might want to read the DSM 5 perhaps? Remember your mother’s words ‘don’t run while carrying scissors’. Watching the late stage ‘Person of Interest’ tv show should be mandatory in ai oversight boards.
He has praised coding publicly many times. But i underrsand is -- He is always able to provide more balanced (pro -cons analysis) on AI than anyone else.
And yes! AI coding is here to stay. I use it a lot.
Gary is not anti-AI. Neither am I. I suspect those of us who became interested in Cog Sci and AI before Tech Bros were a thing, have a slightly different take on the industry.
Shout out to a UBC Philosopher who recently said to me "I'm glad someone (me) is still positive about this subject." Altman and Musk will never change my enthusiasm for the field, if not the companies and their inflated valuations.
We should call these models “industrial models”, not “frontier”. Frontier implies innovation and exploration. These models are just big and use resources at industrial scale.
Before the dot.com crash, Cisco was the most admired company and its CEO, John Chambers, was the most admired CEO. They could almost name their price for their servers, which were the backbone of the internet. After the dot.com crash, people could buy "lightly-used" Cisco servers at significant discounts, Cisco's stock went down by approximately 65%, and Cisco never again attained the primacy it held before the crash. Of course, many other high-flying companies at that time, including others that were also core infrastructure, saw their stocks go down by even larger amounts and many were unable to stay in business. That could easily happen again.
A critical distinction exists between the 2008 financial crisis and the current AI bubble: the capacity for a government rescue. Today, the U.S. national debt-to-GDP ratio hovers at a staggering 122%, on a trajectory toward a breaking point as trillions of dollars are required just to service existing debt. Given these severe fiscal constraints, combined with a profound lack of political will in Congress to fund corporate interventions, AI companies face a market reality where a federal bailout is highly improbable
What's most comical is essentially AI has been pushing since 2024, and these bears just get destroyed for months and months and months, and then whenever there are red events they all come out of the woodwork like they called something or that even having down days proves anything at all. Until the EARNINGS REPORTS show weakness, this party is far from over. I welcome lower prices.
Eighteen months from the most valuable startup in history to government bailout candidate. That's the speed record nobody wanted to set.
The SpaceX GPU story is the detail that tells you where this is heading. last year the chips were too valuable to sell. this year they're inventory you can't find enough tenants for. when the hoarder becomes the landlord, the thesis has changed and nobody sent the memo to the market cap.
The government stake framing is doing the quiet work here. "Investment" is the word you use when the numbers don't support "commercial return" but the politics don't allow "bailout." the playbook is identical to 2008 bank rescues, just wearing a hoodie instead of a suit. privatise the gains in the $150 billion valuation round, socialise the losses when the revenue doesn't show up. if OpenAI had a path to profitability it wouldn't need the government in the cap table. the cap table is the confession.
It seems no one wants to ask the obvious question: 'If your AI is so intelligent, why can't it tell you how to achieve profitability?' ;)
sam actually once said he would try that!
So by now, shouldn't GPT-5-5 be able to help the poor guy out?
Truly a monumental point there! My guess is they ask that all the time and don't like the answers.
That’s like asking “how can I travel faster than light?”
Having read Ed Zitron, even a Chatbot probably knows that isn’t possible.
Im GUFFAWING
Ironic Gary liking this comment that was written by ChatGPT.
the first two sentence are great, am willing to admit
It was? How can you tell? What did I miss?
“Doing the quiet work here” is classic GPT phrasing.
Oh wow! Apprecciate it! Guess I'm not that familiar with it's outputs yet.
“the cap table is the confession” is also a dead giveaway
I’m not knocking it. Using LLMs as a writing tool can be very useful for certain situations and with good instructions you can completely change the writing style and beat that cliche “AI feel” out of it to fit your tastes.
It demonstrates laziness, lack of imagination, inability to put together a coherent argument and/or illiteracy when one uses it for comments on a blog that really are not that difficult to put together without using a chatbot.
Don’t worry,
You haven’t missed anything important.
Lots of tells - the speed record nobody wanted to set - is classic AI
The sentence structures e.g. "if OpenAI had a path to profitability it wouldn't need the government in the cap table. the cap table is the confession."
Doing the quiet work as noted earlier
Pangram is quite definitive
https://www.pangram.com/history/320a6467-8d14-4ae5-b0ce-8e1935af3929?ucc=YhBVDVN3Je7
However the fact that the comment is AI generated doesn't entirely make it wrong
"Investment" is the word you use when the numbers don't support "commercial return" but the politics don't allow "bailout."
Is an entirely valid point. It's wrong because what the government should do is buy the assets it wants from the bankruptcy firesales... but it is exactly what governments have done historically
Because of the post training process consumer facing LLMs have a relatively strong default writing style in the absence of further instruction or context. It’s amazing to me the number of people who use default free ChatGPT and then claim it’s “dumb” or hallucinates or whatever. Absurdly under-appreciated the degree to which the outputs of this systems varies enormously between naive and highly technical users.
posted, and immediately reposted by Carl Quintanilla of CNBC: https://x.com/GaryMarcus/status/2063354070606462989?s=20
Privatize the gains, socialize the losses, externalize the effects ('ESG' investing) :/
PSE investing 🤔😂
Thanks to Sam A for showing the US taxpayers possible dividends while eying at her pockets. How about socialization of the losses? With capitalists like this, who needs socialists?
I have been doing investing for almost 10 years and for the first time my broker has been "informing" me about an upcoming IPO. I'm happy to lose my own money but it won't be by investing in Space X stock.
FINALLY. She sits back and pulls up a chair. Popcorn all around people 🍿
Except the Canadian gov is investing my tax dollars in this, instead of housing co-ops. Nuts.
Dude. Don't even get me started on the level of anger I have about how my tax dollars are being used. I'm an American...
I heard an author speaking about AI who casually asserted the cost of "intelligence" is dropping globally. I was unable to externalize the poetic irony.
It would have been more accurate to just say “intelligence (of the real as opposed to artificial kind) is dropping globally”.
What just happened?
Maybe a dead cat will bounce on Monday. Let's see what just happens on Tuesday.
The fact that Trump is talking enthusiastically about the American government investing in AI... is not a good look. The Chinese must really be laughing.
If you have a bank that's too big to fail, you pump in money and it carries on working.
If you have an AI company that's pursuing a completely unworkable technology, pumping in any amount of money isn't going to help. You're even more doomed.
The US economy just ran into an epistemological brick wall. You can't build "intelligence" by parsing language strings, however much compute you throw at the problem and how many texts you scrape. Any competent philosopher will tell you this.
Is there a precedent in history when an entire industrial economy has literally been based on hallucinations?
To me, this AI market slump isn’t a surprise it’s a predictable leadership failure. I’ve watched platforms optimise blindly for pure scale, assuming bigger always meant better. Instead, we willingly bypassed safety boundaries and ignored the real societal costs. We prioritised algorithmic hype over human utility, and now the system is finally breaking. I wish we had more news about things that really bring value and improve our lives. Not the typical cheering on how many people lost due to automation, how a certain model can code better than a person. This is all an hallucination from certain players. We need t focus in reality and not in "valuations".
"If scale was “all you need”, or if AGI was actually nigh, Elon would be hoarding LLMs, not leasing them."
I'm assuming LLMs should be GPUs. FYI.
The job numbers were unexpectedly high. Jobs numbers are released on the first Friday of every month. What does that mean?
The job of the Federal Reserve is to keep price stability and low unemployment. If job numbers are unexpectedly high, the likelihood of interest rate increases goes up. Higher interest rates make it harder for businesses to be profitable. AI has a lot of debt, and it is more sensitive to interest rates.
AI 'Tokens' need to be 'spent' on the highest ROI. I see AI’s initial generosity as generating publicity and investor interest, and as a way to determine where the highest ROIs are located so they can focus their efforts on those specific industries. AI was never going to give us fully automated communism. The lump-of-labor fallacy has repeatedly shown that a larger economy generates more work.
Do not be surprised if prices bounce back on Monday. Market moves on the first Friday of every month are expected and unrelated to AI.
https://en.wikipedia.org/wiki/Lump_of_labour_fallacy
Respectfully, I think you are so anti AI that it does bias your writing. I agree with your take on private companies getting bailed out. Cronyism, and socialism are things I don't want either. But that doesn't mean all things AI are bad. Writing code has changed forever and one person can be as productive as two or more in some cases. Yes, we're overvaluing AI companies, but the value isn't zero either. After the bubble bursts, we'll still write most code using AI. The pace of software writing is forever changed. That said, I enjoy your takes. You're thought provoking. Thanks for what you do.
i literally wrote an essay on how claude code was the biggest innovation in the last few years of AI.
Obviously I missed this. Apologies.
You missed more than one. Gary is not and never has been anti AI. I'm really disgusted with people making that accusation simply because he's pointed out what is now undisputed truth, that LLMs and scaling *alone* will never achieve AGI. It's my understanding that Claude *is* combining LLMs with symbolic tools. People who try to shoot the messenger of facts they don't want to face would do well to examine their own mental reflexes and thought patterns before attacking ethical and knowledgeable messengers like Gary Marcus.
thank you 🙏
Thank YOU for all you do in service to our country, humanity, and the world.
🤔- modelling ai on a version of the PFC/Limbic concept of the human brain (RL/DL, search/value splits) might want to read the DSM 5 perhaps? Remember your mother’s words ‘don’t run while carrying scissors’. Watching the late stage ‘Person of Interest’ tv show should be mandatory in ai oversight boards.
PREACH. Well said ma’am.
He has praised coding publicly many times. But i underrsand is -- He is always able to provide more balanced (pro -cons analysis) on AI than anyone else.
And yes! AI coding is here to stay. I use it a lot.
Gary is not anti-AI. Neither am I. I suspect those of us who became interested in Cog Sci and AI before Tech Bros were a thing, have a slightly different take on the industry.
Shout out to a UBC Philosopher who recently said to me "I'm glad someone (me) is still positive about this subject." Altman and Musk will never change my enthusiasm for the field, if not the companies and their inflated valuations.
We should call these models “industrial models”, not “frontier”. Frontier implies innovation and exploration. These models are just big and use resources at industrial scale.
Well considered insights. Thank you Gary.
“if Elon thought AGI was nigh, he wouldn’t be giving compute away” facts, facts, facts
Before the dot.com crash, Cisco was the most admired company and its CEO, John Chambers, was the most admired CEO. They could almost name their price for their servers, which were the backbone of the internet. After the dot.com crash, people could buy "lightly-used" Cisco servers at significant discounts, Cisco's stock went down by approximately 65%, and Cisco never again attained the primacy it held before the crash. Of course, many other high-flying companies at that time, including others that were also core infrastructure, saw their stocks go down by even larger amounts and many were unable to stay in business. That could easily happen again.
A critical distinction exists between the 2008 financial crisis and the current AI bubble: the capacity for a government rescue. Today, the U.S. national debt-to-GDP ratio hovers at a staggering 122%, on a trajectory toward a breaking point as trillions of dollars are required just to service existing debt. Given these severe fiscal constraints, combined with a profound lack of political will in Congress to fund corporate interventions, AI companies face a market reality where a federal bailout is highly improbable
What's most comical is essentially AI has been pushing since 2024, and these bears just get destroyed for months and months and months, and then whenever there are red events they all come out of the woodwork like they called something or that even having down days proves anything at all. Until the EARNINGS REPORTS show weakness, this party is far from over. I welcome lower prices.