Gary Marcus argues OpenAI faces WeWork-like collapse and Anthropic relied on a SpaceX subsidy

1. Agreed that OpenAI is in deep trouble; that’s why I have long suggested that it might be the WeWork of AI but 2. Anthropic is not out of the woods; their best quarter was exactly when tokenmaxxing - already declining - was at its peak, and relied on a big one time subsidy for SpaceX.

even if @scaling01 turns out to be wrong about some of these, I respect the specificity.

@beffjezos that is bs; i have conceded points three times just in the last week

you however will never concede the things i got right

and llms remain unprofitable

The zeitgeist has radically changed this months. Here’s what’s likely to happen next: - OpenAI will struggle - with the decline of tokenmaxxing Anthropic will struggle (aside from this quarter) to make a profit - Google will catch up to Anthropic - some Chinese companies might, too - LLMs will become commodities; margins will be very very thin - Most of the companies that invested massively in them will struggle to make back their investments - Nvidia will eventually decline, once all of the above becomes widely recognized.

I’m going to let you in on a secret. I made a mistake once.*

In August 2024.*

It’s true. I actually got something wrong. I predicted that there would be a collapse of the AI bubble (which in my judgement is still pending and possible, but hasn’t happened yet), and got the timing wrong!

You can see the error below; a lot of people like to drag with me it. And my timing prediction was definitely wrong.

But people who run around posting that one tweet are missing the forest for the trees in many ways:

• Pointing to one failed prediction is a cheap shot in the larger context of so many predictions that I have gotten correct. Sixteen out of my 17 high-confidence predictions for 2025 were correct, for example. (Summary at the essay “Six (or seven) predictions for AI 2026 from a Generative AI realist”).

• The tweet in which I got the timing wrong actually got a lot right, way ahead of its time. The tweet refers to an essay (“Generative AI Still Needs to Prove Its Usefulness”) that I had just written that was ultimately published in December 2024. In it I warned correctly that “essentially every big company seems to be working from the same recipe, making bigger and bigger language models, but all winding up in more or less the same place” which has basically proven true (indeed Larry Ellison just conceded it 18 months later). • I also warned there that “Unless [OpenAI comes out] with some major advance worthy of the name of GPT-5 before the end of 2025 that is decisively better than what their competitors can offer, the bloom will be off the rose. The enthusiasm that propped up OpenAI will diminish”, which by now pretty much true; by any reasonable measure OpenAI’s has indeed shrunk. • What I said overall – that the industry might not be profitable – is still largely true. And the view that I expressed than has gone from sounding weird to something that other people contemplate constantly, and largely for the very reasons I pointed out, ahead of their time.

*Now here’s the asterisk, and one more secret. In the course of making more concrete predictions than anyone else I know, I have actually have made a number of mistakes, not just one. But the good thing about me is that I concede them regularly, e.g., one to @EthanMollick last week, another to @GergelyOrosz a couple days ago.

If you look around, you will see that’s rare.

@GaryMarcus Token maxxing is at currently at the stage of going to the gym once a month versus everyday :D

-Gary Marcus will never admit defeat

place your bets

Hot take on what comes next, after the sudden decline of tokenmaxxing:

- OpenAI and Anthropic will not be monopolies - Chinese LLMs will succeed a lot more - Intelligence will become too cheap too meter - Companies will increase AI adoption by 100x - We will forget about toxenmaxxing and indulge in profit maxxing

💃💃💃💃💃