EDIT: The fact that I'm downvoted for saying that tells you that not only we're in a massive bubble, but the author knows that he's personally invested in the bubble getting bigger.
I appreciate the author trying to address this. The only issue is the author didn't define what 'pops' in the bubble.
The 'what' that pops, probably, IMHO, is the investment and resources ecosystem dumped, poured, thrown, redirected, etc. to AI.
Just like the .com boom, the costs and money thrown at AI are outsized and will 'pop" as it matures. A 'pop" and leveling down to something perhaps not so outrageous is coming.
As the author states, .com popped but left the Internet still around and expanding. AI is exactly that kind of bubble.
But if they think the bubble is the idea that "Modern AI will significantly transform business and the global economy, and that it'll lead to massive unemployment for knowledge workers", then—in my opinion—they're just wrong.
AI will one day transform the world but I do not believe modern AI will make that happen, that will have to wait for futuristic AI. I guess that makes AI not a bubble but I can still label modern, i.e. current, AI a bubble, right?
Part of why AI is concerning is that literally nobody can actually predict what the outcome of these models is going to be even a couple of years out, the decision tree just instantly blows out and becomes impossible to track.
I really dislike the title and the article. You are basically stating that AI is not a bubble, but everything around AI is a bubble, which people obviously mean when they say AI is a bubble. Same was the case with dot com crash, where people didn't LITERALLY mean .com domains would crash, but everything around the domains (& Internet) would crash.
Does not help that the author is overinvested into the AI space.
>Do people actually believe that if you just "add AI" to what they have, they'll instantly become millionaires?
Looking at recent YC batches, YC seems to believe so.
He says bubbles are false beliefs about reality that everybody figures out is false. The claim that "[AI] will displace tens or hundreds of millions of knowledge workers in the next 3-10 years" is that false belief inflating the bubble. Big tech and investors are dumping money into AI thinking at least a sizeable portion of the white collar workforce is going to be replaced with $2000-$4000/month subscriptions to AI agents (I am so sick of the term "agent") acting almost autonomously. We're hitting the wall with what's feasible with LLM scaling laws even if you're dumping countries GDP worth of money into making bigger and better models. Is spending $500 billion to $1 trillion dollars to get something that performs marginally better than the best LLM today worth it?
Why should I trust someone that's making money from AI to tell me that AI isn't a bubble?
Especially when the entire argument seems to hinge on the author's dislike for the word Bubble. The bubble that pops isn't the actual thing, it's the money that was propping the thing up...
I don’t feel like this article addresses the question of whether we’re seeing an AI bubble or not at all.
To me this is very simple: certain companies have invested billions upon billions in AI data centres that need to see a very high return on investment in the years ahead to avoid a “pop”. I think nvidia is also being valued is if it’s a certainty that we will continue to build many such data centres every year for the foreseeable future.
These expectations are now being faced with the reality that AI isn’t making us that much more productive, and the improvements in LLMs are slowing down, so it’s not clear there will be a continued will to pay expensive subscription for these AIs, and certainly not clear that companies are willing to accept higher prices that may be need to get ROI.
There’s another way this could go down: I have a sneaking suspicion that we can get 90% of the productivity benefits of LLMs with models that can run locally. A combination of an algorithm breakthrough, better tuned models, better software tooling around them, and better hardware in the hands of end users, may get us to that 90%. That could end up making a lot of AI companies redundant.
1. The leading companies in LLMs: Anthropic and OpenAI are both not profitable and rely on huge amount of continuous investment (from "successes" like SoftBank who continuously invested in WeWork) to even stay afloat.
2. Each model from GPT3.5 to GPT 4 to 5 has been 10x more expensive to train. However, we have long ago entered the era of diminishing returns where 4 was a big improvement from 3.5 but 5 was a mediocre improvement (and to some, a regression) from the previous model. That suggests, unless a miracle occurs, 6 will be extremely expensive to train to imperceptible improvements.
5. Anthropic are very early in their journey to be already scraping the barrel, yet they recently 180 on their ToS and admit that they, in fact, will be training on your chats (suggesting that they can't get more/quality data elsewhere): https://www.anthropic.com/news/updates-to-our-consumer-terms.
6. OpenAI and Anthropic are both connected to CoreWeave, Nvidia, Microsoft, Google. What is funny is that CoreWeave is also connnected/dependent on Nvidia, and so is Microsoft and Google. Those companies' stocks are a major part of the US economy at the moment. https://www.reuters.com/business/dominant-ai-trade-confronts.... "Meanwhile, the combined market cap of the 10 biggest AI plays, including Nvidia, Broadcom and Microsoft -- stood at $18 trillion, BCA said in a note last week. That amounts to about 33% of S&P 500 stock market capitalization, up from around 15% in late 2022, according to BCA."
10 companies in the same sector, comprising 33% of the S&P capitalization is a 100% bubble, and if anything, one of the biggest ever to exist. And I, for one, can't wait for it to pop so I don't have to read hype booster articles like this on the front page of HackerNews.
The bubble is AGI, which was pitched as imminent (it’s even part of the OpenAI-Microsoft agreement language). It will pop when people realize it’s not remotely near.
“AI will change how business is done” is not a bubble because every innovation changes how business is done, it’s just a matter of degree. Various forms of AI have been changing businesses for a decade at least.
Because something could very easily pop soon, but this article is just laying the groundwork for an argument that "it wasn't AI popping, that was the belief you could run a AI adoption program and fire half your staff popping."
At some point, what's the difference? Obviously LLMs are here to stay in some form or another.
What the stock market does with that is an entirely different animal and honestly I'm not optimistic.
This article feels self defeating. It’s rewriting the meaning of the word bubble itself. There is a speculative asset bubble. Whilst I think there is some correctness in saying that drawing parallels with the Dotcom boom/bust is intellectually lazy I still think this acts as a notable precedent. Is there potential for AI usefulness long term? Yes, will it take exactly the same form as now or do the products and services exist now? No. So there is some utility in comparing the two. Either way this article reads like a bit of cope.
The article says a bubble is something that pops but in the context of AI I think most people mean a financial bubble as in "a period when current asset prices greatly exceed their intrinsic valuation, being the valuation that the underlying long-term fundamentals justify". Which we probably have. (https://en.wikipedia.org/wiki/Economic_bubble)
A bubble that will pop is the one where they say that AI will replace creative humans. This is certainly wrong, because it requires consumers to stop being fans of actual people.
Consider XKCD. Will I lose interest in Randall Munroe because an AI thing will auto-generate any number of XKCD-style comics? Of course not.
The proliferation of AI has been compared to the proliferation of polyester.
28 comments
[ 2.7 ms ] story [ 53.2 ms ] threadOh yes, this must be why chatbot companies announced $25 billion of new funding last week.
EDIT: The fact that I'm downvoted for saying that tells you that not only we're in a massive bubble, but the author knows that he's personally invested in the bubble getting bigger.
So of course he'll say "it isn't a bubble".
The 'what' that pops, probably, IMHO, is the investment and resources ecosystem dumped, poured, thrown, redirected, etc. to AI.
Just like the .com boom, the costs and money thrown at AI are outsized and will 'pop" as it matures. A 'pop" and leveling down to something perhaps not so outrageous is coming.
As the author states, .com popped but left the Internet still around and expanding. AI is exactly that kind of bubble.
AI will one day transform the world but I do not believe modern AI will make that happen, that will have to wait for futuristic AI. I guess that makes AI not a bubble but I can still label modern, i.e. current, AI a bubble, right?
- Virtual Reality: big hype in the early 90s (arcades, movies like Lawnmower Man) through to use cases today like surgical training, aviation training
- Mobile video calls: hyped in early 2000's with 3G and pre-iPhone devices. Actually took off with 4G and 5G plus iOS and Android phones
- 3D printing: back in 2013 we were expecting "a 3D printer in every home" ... today valuable in industrial prototyping
Looking back at 2025 we'll be saying "Remember when they said everyone would lose their jobs to AI..."
Anybody telling you otherwise is lying to you.
https://companiesmarketcap.com/juniper-networks/stock-price-...
Looking at recent YC batches, YC seems to believe so.
He says bubbles are false beliefs about reality that everybody figures out is false. The claim that "[AI] will displace tens or hundreds of millions of knowledge workers in the next 3-10 years" is that false belief inflating the bubble. Big tech and investors are dumping money into AI thinking at least a sizeable portion of the white collar workforce is going to be replaced with $2000-$4000/month subscriptions to AI agents (I am so sick of the term "agent") acting almost autonomously. We're hitting the wall with what's feasible with LLM scaling laws even if you're dumping countries GDP worth of money into making bigger and better models. Is spending $500 billion to $1 trillion dollars to get something that performs marginally better than the best LLM today worth it?
Especially when the entire argument seems to hinge on the author's dislike for the word Bubble. The bubble that pops isn't the actual thing, it's the money that was propping the thing up...
To me this is very simple: certain companies have invested billions upon billions in AI data centres that need to see a very high return on investment in the years ahead to avoid a “pop”. I think nvidia is also being valued is if it’s a certainty that we will continue to build many such data centres every year for the foreseeable future.
These expectations are now being faced with the reality that AI isn’t making us that much more productive, and the improvements in LLMs are slowing down, so it’s not clear there will be a continued will to pay expensive subscription for these AIs, and certainly not clear that companies are willing to accept higher prices that may be need to get ROI.
There’s another way this could go down: I have a sneaking suspicion that we can get 90% of the productivity benefits of LLMs with models that can run locally. A combination of an algorithm breakthrough, better tuned models, better software tooling around them, and better hardware in the hands of end users, may get us to that 90%. That could end up making a lot of AI companies redundant.
2. Each model from GPT3.5 to GPT 4 to 5 has been 10x more expensive to train. However, we have long ago entered the era of diminishing returns where 4 was a big improvement from 3.5 but 5 was a mediocre improvement (and to some, a regression) from the previous model. That suggests, unless a miracle occurs, 6 will be extremely expensive to train to imperceptible improvements.
3. The latest "features" from OpenAI:
- study mode https://openai.com/index/chatgpt-study-mode/
- the rumor for an office suite https://www.computerworld.com/article/4021949/openai-goes-fo...
show that they are running out of ideas what to do.
4. Anthropic is settling on huge cases: https://news.ycombinator.com/item?id=45142885 while struggling with usage vs monetization: https://news.ycombinator.com/item?id=44715471
5. Anthropic are very early in their journey to be already scraping the barrel, yet they recently 180 on their ToS and admit that they, in fact, will be training on your chats (suggesting that they can't get more/quality data elsewhere): https://www.anthropic.com/news/updates-to-our-consumer-terms.
6. OpenAI and Anthropic are both connected to CoreWeave, Nvidia, Microsoft, Google. What is funny is that CoreWeave is also connnected/dependent on Nvidia, and so is Microsoft and Google. Those companies' stocks are a major part of the US economy at the moment. https://www.reuters.com/business/dominant-ai-trade-confronts.... "Meanwhile, the combined market cap of the 10 biggest AI plays, including Nvidia, Broadcom and Microsoft -- stood at $18 trillion, BCA said in a note last week. That amounts to about 33% of S&P 500 stock market capitalization, up from around 15% in late 2022, according to BCA."
10 companies in the same sector, comprising 33% of the S&P capitalization is a 100% bubble, and if anything, one of the biggest ever to exist. And I, for one, can't wait for it to pop so I don't have to read hype booster articles like this on the front page of HackerNews.
“AI will change how business is done” is not a bubble because every innovation changes how business is done, it’s just a matter of degree. Various forms of AI have been changing businesses for a decade at least.
Because something could very easily pop soon, but this article is just laying the groundwork for an argument that "it wasn't AI popping, that was the belief you could run a AI adoption program and fire half your staff popping."
At some point, what's the difference? Obviously LLMs are here to stay in some form or another.
What the stock market does with that is an entirely different animal and honestly I'm not optimistic.
Consider XKCD. Will I lose interest in Randall Munroe because an AI thing will auto-generate any number of XKCD-style comics? Of course not.
The proliferation of AI has been compared to the proliferation of polyester.