The AI Bubble Monitor | Hacker News Reader

The AI Bubble Monitor

cepr.net

66 pointsmooreds37 comments

> I have a lawyer friend who pays $20 a month for an Anthropic system.

I have a lawyer friend who's seen a dip in clients reaching out. Turns out, they're having Claude draw up contracts.

I used Claude to help me draft some letters to negotiate severance after I was laid off. It was really useful and I did manage to get a bit extra from my former employer

We're going to be in for a really rough time as more and more stuff gets replaced by AI and it becomes harder for people to earn incomes

Soon it'll be your AI message going to an AI on their end making the severance call with no human in the loop (if it wasn't already!).

I used Claude to help me rewrite the garbage job description they had for my position, based on the work that I was actually doing.

It was so much better. I hesitate to say 10x better, but it sure felt like it.

I have a lawyer friend who receives Claude contracts for drafting. Oh dear, the level of bullshit they find, apparently quality is classes below vibe coding. But as long nobody checks...

Maybe they're using the free version...

Or probably even moreso - reviewing contracts

Low end - absolutely.

High end? Not yet. Corporate contracts are getting human only oversight with us. We have tried to compare but found that the signal vs noise generated doesn't compete with our expensive legal department.

ai seems to be generating a K shaped ROI, But symmetrical. Like the rumsfeld quote: you have known knows, unknown knowns, known unknown and unknown unknowns.

If you know your use case and know AI limits, that's known knowns. The rest follows.

The article asks the 25 trillion dollar question. ‘What is the near term demand for AI?’ And the answer is nobody knows; especially not these companies. I think the bubble is almost inevitable. Everyone is racing to gain market domination as they see it as winner take all, but it just doesn’t appear to be shaking out that way. There are five very competent US based models alone and then I don’t even know how many open weight models. Unless the industry coalesces around 2-3 models soon I don’t think anyone makes any money on this in the next 10 years+.

There's no real incentive for the industry to coalesce around 2-3 models. I think it's very unlikely to happen.

Training is expensive. That seems like an important incentive.

I agree. Which is why I think the bubble is going to pop in the nearish term.

Do you think there will be no consolidation in the pop?

Nobody wants to admit that while partial to one company or another, brand and all that, I highly doubt that within 1-2 years we can tell the difference between model output for 90% of our tasks.

Except for the 'vibe' of some of the models with their writing or testing style I am not certain I could tell you which model is which unless asked.

Switching from model to model at the moment has been seamless without extensive 1 to 1 bench-marking but most of my tasks are never 1 to 1.

Completely agree. And for companies, if things are that indistinguishable, it is going to come down to cost and enterprise security / privacy. Sure companies will be slower to switch, but they will. It will be a race to the bottom.

It could be like the last SOTA lab to run out of money wins

The more relevant question for the short term is, how much of the earnings originate from debt? Everyone points to the PE ratio as evidence the bubble isn't that bad, but if you subtract all of the borrowed money, how much of those earnings are coming from cash flow from actual AI demand? I'd say 10% at best, which would effectively 10x the PE on many of those companies (particularly the chip stocks).

”China’s bubble already collapsed, leading to a plunge in its AI-tech index of 40 percent” This was news to me, is there any good coverage on what happened?

Here's a brief overview:

https://www.reuters.com/world/china/china-ai-bubble-has-alre...

South Korea too.

The actual makers of the actual silicon are taking a bath

https://finance.yahoo.com/markets/stocks/articles/south-kore...

Meanwhile here in the west the money keeps going round and round like some sort of "self sustaining economy"

Shameless self-promotion: https://artificialliabilities.com for a nice story-like AI infra overview. Based of FRED, SEC, and other publicly available data.

the only realistic demand is see is similar to why Azure, digitalocean etc are still breahing. There are incompetent companies that cannot live without certain services, so they'll stay dependent. A lot of companies are doing agentic engineering, and the rat race is even stronger there. You need a product that is better than a vibe-coded app, but you aren't competent enough to get there. These companies are going to die out at some point, and what will be left is the better half.

Even that won't make back the 3 trillion that is needed.

From what I am seeing about 10% of our AI spend has any ROI. You can count 90% loss of $XX but the ROI we get from that 10% is 200%-10000% of our spend.

Getting and retaining customers is 99% of profit which is why sales, marketing, advertising, and creativity is always in high demand. After that, quality.

AI solves problems that do not exist, it creates problems and then promise to solve them, that's why there's neither a market for them nor any profit.

Given the debt and deficits the government is running I worry they're not going to haveh te financial muscle for a rescue package when this implodes.

Not a problem if the govt takes back the minting of money from the Fed.

The author, Dean Baker an economist at a non-profit policy group, called the housing bubble. Not wealthy, he sold his condo in 2004 and bought a house in 2009.

Doesn't the title presuppose the outcome?

So what would you suggest as an alternative?

Some great charts in there that emphasize how absolutely ridiculous SpaceXS(?)ITesla is.

My take on "bubble" activity from HN chatter: https://www.alwaysleaveanote.com/hacker-news/bubble-talk/

The AI companies are assuming that there will be demand in the future for their products. That is why they are trying to scale up to meet the demand.

They are hoping rather than assuming because if there isn't sufficient demand (sufficient paying demand, not people who wouldn't pay just using it for free because they can), there is no way back for them that isn't an amusingly catastrophic collapse, and they are scaling up as if the demand is a forgone conclusion because it just isn't worth trying to plan for the other option.

That's true, but it seems like Anthropic actually has been capacity-strapped fairly often this year.

Yes the AI bubble is real, but the good things is, it is expaning like an unverse which has enourmous posibilities. I could has said that it will burst soon but after LAYA and JEV, we all are going to witness the lots of changes.

The stages of AI bubble are basically like this:

1. Denial: “No way these companies are worth anything. Zero profits.”

2. Anger: “Idiots! This will end like the 1999 dotcom crash!”

3. Bargaining: “Ok maybe if it goes lower I’ll… I’ll buy!”

4. Depression: “It just keeps going up…”

5. Acceptance: “I missed out, it was so obvious, now I’ll just buy in.”

That right there is a pretty good demonstration of the alternative position. "This time is different."

Reply on news.ycombinator.com