The funniest part is that the dot com crash didn’t kill the internet. It killed a lot of terrible business plans. AI may follow the same script: useful tools survive, investors discover that a chatbot isn’t worth infinite money, and taxpayers get stuck with the data center power bill. The technology can be real while the valuation is pure casino. - PS: You didn’t mention the smartphone. And that’s exactly what people mainly use these days.
The thing with this bubble is that it’s all terrible business plans. We’re 4 years in and no great leap. No productivity gain. No company that’s been able to turn the tech into something that makes a profit.
Yes, the tech will survive but if nobody is prepared to pay the actual compute costs, it’s going to get abandoned.
Exactly. And cheap open weight AI is the wildcard. If people can run capable OWAI locally for pennies, the trillion dollar moat starts looking like a very expensive puddle. AI could thrive while the companies betting on permanent scarcity discover they funded their own competition.
I’m not sure how much distillation and building on top of each others results is going on. I could see all the models freeze in development when the money leaves the segment.
A model frozen in the past has limited usefulness.
Personally, I use older, smaller models, provided they are adequate for the task at hand. My theory is that at least 80% of companies are using a sledgehammer to crack a nut. Not only is that incredibly expensive, but it’s also incredibly ignorant.
That’s the thing about bubbles: it doesn’t mean that the thing is useless, just that its value is inflated. We still have websites after the dotcom bubble burst, certainly we still have housing despite several bubbles bursting there. I would be surprised if LLMs go away entirely. OpenAI and Anthropic might not survive, but Google and Microsoft will continue. God only knows what will happen in China. Local models will still be OK. The broader concepts of machine learning, deep learning, big data etc have been useful for far longer than this bubble and will continue.
Exactly. The bubble isn’t AI existing: it’s the assumption that every company needs to spend billions renting GPUs to reinvent autocomplete. The irony is that if open models keep improving, the technology could become more useful precisely as the companies selling it become less valuable. Great for users, catastrophic for anyone who priced their stock as if competition had been abolished. - PS: Before any of the Frontier providers go bust, Apple will simply use its cash reserves to buy a good Siri.
It depends, if Anthropic keep nerfing Claude’s capabilities they’re going to lose to their competitors. I’ve been doing defensive cybersecurity work for the last couple of weeks and Claude, a tool which should be ideal for automating the tedious parts of this, has been a complete pain in the arse.
Claude: I’ve found a security bug in your codebase that nobody knows about!
Me: Oh shit that’s not good, can I see it?
Claude: <blocked by safety classifier>
Repeat ad nauseam.
The prissy little bastard’s recalcitrance made me to do a lot of stuff by hand anyway, but instead of being bored I’m now actively irritated by the bot telling me what I can and can’t do. Obviously can’t use them at work for regulatory reasons, but when I was doing similar stuff at home to secure a side project which has to live on the public internet GLM and DeepSeek have no such qualms and honestly I’m convinced the whole ‘DeepSeek is the budget option’ thing is mostly behavioural economics; the cheapness to me feels like efficiency gains more than nerfing the model.
My boss is big into Claude code and has produced something really useful from it. Just a shame that’s a job a software developer could have been paid to do 😢
The funniest part is that the dot com crash didn’t kill the internet. It killed a lot of terrible business plans. AI may follow the same script: useful tools survive, investors discover that a chatbot isn’t worth infinite money, and taxpayers get stuck with the data center power bill. The technology can be real while the valuation is pure casino. - PS: You didn’t mention the smartphone. And that’s exactly what people mainly use these days.
The thing with this bubble is that it’s all terrible business plans. We’re 4 years in and no great leap. No productivity gain. No company that’s been able to turn the tech into something that makes a profit.
Yes, the tech will survive but if nobody is prepared to pay the actual compute costs, it’s going to get abandoned.
Exactly. And cheap open weight AI is the wildcard. If people can run capable OWAI locally for pennies, the trillion dollar moat starts looking like a very expensive puddle. AI could thrive while the companies betting on permanent scarcity discover they funded their own competition.
I’m not sure how much distillation and building on top of each others results is going on. I could see all the models freeze in development when the money leaves the segment.
A model frozen in the past has limited usefulness.
Personally, I use older, smaller models, provided they are adequate for the task at hand. My theory is that at least 80% of companies are using a sledgehammer to crack a nut. Not only is that incredibly expensive, but it’s also incredibly ignorant.
That’s the thing about bubbles: it doesn’t mean that the thing is useless, just that its value is inflated. We still have websites after the dotcom bubble burst, certainly we still have housing despite several bubbles bursting there. I would be surprised if LLMs go away entirely. OpenAI and Anthropic might not survive, but Google and Microsoft will continue. God only knows what will happen in China. Local models will still be OK. The broader concepts of machine learning, deep learning, big data etc have been useful for far longer than this bubble and will continue.
Exactly. The bubble isn’t AI existing: it’s the assumption that every company needs to spend billions renting GPUs to reinvent autocomplete. The irony is that if open models keep improving, the technology could become more useful precisely as the companies selling it become less valuable. Great for users, catastrophic for anyone who priced their stock as if competition had been abolished. - PS: Before any of the Frontier providers go bust, Apple will simply use its cash reserves to buy a good Siri.
I think Anthropic will survive. Claude code is very useful.
Cursor may secure Groks future too.
It depends, if Anthropic keep nerfing Claude’s capabilities they’re going to lose to their competitors. I’ve been doing defensive cybersecurity work for the last couple of weeks and Claude, a tool which should be ideal for automating the tedious parts of this, has been a complete pain in the arse.
Claude: I’ve found a security bug in your codebase that nobody knows about!
Me: Oh shit that’s not good, can I see it?
Claude: <blocked by safety classifier>
Repeat ad nauseam.
The prissy little bastard’s recalcitrance made me to do a lot of stuff by hand anyway, but instead of being bored I’m now actively irritated by the bot telling me what I can and can’t do. Obviously can’t use them at work for regulatory reasons, but when I was doing similar stuff at home to secure a side project which has to live on the public internet GLM and DeepSeek have no such qualms and honestly I’m convinced the whole ‘DeepSeek is the budget option’ thing is mostly behavioural economics; the cheapness to me feels like efficiency gains more than nerfing the model.
Useful doesn’t mean profitable, sustainable or cost effective.
You’re not paying the real cost right now.
My boss is big into Claude code and has produced something really useful from it. Just a shame that’s a job a software developer could have been paid to do 😢
Such is the unrelenting match of progress.
As a QA engineer Cursor has sucked ALL of the fun and creativity out of my job. Also cost two of my subordinates their job
I can’t deny it’s useful though