Agreed. I am coming to suspect that there are approximately 2 positions on this subject: there are people who have previously weathered an AI winter and can therefore easily see the writing on the wall, and there are people who maybe don't quite understand the phenomenon being described by the term "AI winter".

It's not about how good or useful or potentially lucrative the technology is. Every previous catalyst for an AI winter has eventually become pervasive, changed the world, and made some people a lot of money. Every single one. And I do mean pervasive. If you could poof just one of them out of existence, my cell phone would become noticeably less useful.

What it's really about is how hype cycles interact with funding for basic research.

>> "LLMs have reached a plateau."

you should look at benchmarks such as ARC which went from "needs 10 years, currently at 0%" to almost solved within the least year. Also there is a revolution happening in math which the layman might be missing.

I don't care about the benchmarks. I care about how helpful coding agents are for my work. And I can barely tell the difference between the models this year and the models last year. Everyone's raving about Opus but I bet about 50% of people would be able to identify it in a blind test against Sonnet.

For ARC v1 it was found that it was much less resistant to brute force than intended/designed. This was improved in v2, which LLMs are currently doing less good at. Note also that ARC tasks are explicitly designed to be slightly-out-of-reach, things that are quite simple for humans, but current models are pretty bad at - designed to measure and enable progress. But yeah there are many interesting approaches, and ARC is interesting to follow both because it attempts to measure ability to adapt to new takas ("fluid intelligence"), and because we have not saturated it yet.

Many technologies plateau, but we don't say they all have winters. Terrestrial radio winter? Television winter? Automobile winter? Air travel winter? Nuclear power comes close in terms of its tarnished image and reluctance to reinvest.

I personally believe contemporary AI is over-hyped, but I cannot say with confidence that it is going to lead to a similar winter as the last time. It seems like today's products satisfy enough users to remain as a significant area, even if it doesn't greatly expand...

The only way I could see it fizzling as a product category is if it turns out it is not economically feasible to operate a sustainable service. Will users pay a fair price to keep the LLM datacenters running, without speculative investment subsidies?

The other aspect of the winter is government investment, rather than commercial. What could the next cycle of academic AI research look like? E.g. exploration that needs to happen in grant-funded university labs instead of venture-funded companies?

The federal funding picture seems unclear, but that's true across the board right now for reasons that have nothing to do with AI per se.

I think it's easy to talk past one another on this subject. To be sure, I love Claude code. I wish I could pay for several years in advance at today's prices.

But the consumer's is not the important perspective for the AI hype cycle. It's the investors' perspective. I'm going to guess that well over a trillion dollars has gone into "AI", including the major labs and all the little "agents for XYZ" startups popping up every day. From the investors' perspective, that better pay off several times over in profit, not revenue, in the next few years. Prices will go up, but competition is fierce and nobody will be able to command a high margin. How are they going to make trillions in profit with razor thin margins?

This will make the very mention of "AI" toxic for most investors sooner than later.

> This will make the very mention of "AI" toxic for most investors sooner than later.

I'm not so sure. I agree about the exceptionally off the charts expectations rivaling only the dotcom bubble, but with the current rate of progress I'd expect AI to be pervasive enough to disappear from investor relationship materials not because it's a buzzword, but because it's assumed to be incorporated just like companies don't mention that they're running Linux servers; companies saying that they're doing stuff without AI will be something to watch for.

It may be my academic career bias, but I feel like the real AI Winter was not due to commercial investor perception. It was a deeper collapse of the government grant funding and higher education agenda which throttled a generation of researchers.

I think of LLMs like brains or CPUs. They're the core that does the processing, but needs to be embedded in a bigger system to be useful. Even if LLMs plateau, there will be a lot of development and improvement in the systems that use these LLM. We will be seeing a lot of innovation going forward, especially in systems that will be able to monetize these LLMs.

> 3 things we know about the AI revolution in 2025:

> - LLMs are amazing, but they have reached a plateau. AGI is not within reach.

Source/citation?