Honestly I find this darkly hilarious. The circlejerk is becoming more and more obvious. I’m going to be surprised if the bubble doesn’t pop within the next 6 months or so - definitely before the end of the year.
All it took for this to happen was:
skyrocketing energy prices
skyrocketing hardware prices
skyrocketing water usage
the effective priceout of PC/homelab enthusiasts and hobbyists which will probably destroy much of the community, and thus most of the consumer-oriented hardware manufacturers
turning the US economy into a house of cards
sharply accelerated enshitification of all major search engines (in the interest of pushing people to use LLM bullshit for no good reason, when a perfectly suitable and deterministic alternative already existed and was deployed at scale everywhere)
flagrantly ignoring licensing and usage terms for basically any and all open source software projects hosted anywhere on the internet
the hollowing out of software engineering as a discipline, the collapse of the hiring prospects of fresh grads/juniors because “ai can do that”, and the creation of a generational gap in staffing across huge swathes of the tech industry writ large
the (even more) accelerated enshitification of social media as it becomes a series of gigantic LLM bot farms talking to each other
Note that malus.sh is a satire pointing out something that the companies are certainly very happy about, but the real AI companies haven’t been that forthcoming with the laundering of copying software through LLM.
“I used to feel guilty about not attributing open source maintainers. Then I remembered that guilt doesn’t show up on quarterly reports. Thank you, MalusCorp.”
Chad Stockholder
Engineering Director, Profit First LLC
it’s rather different than 2008, because credit default swaps were just a bag holder thing, whereas these massive data centers are not only a huge amount of sunk cost capital, but also have extremely high OpEx.
that said, I would be 0% surprised if our Captains of Industry (🥴) manage to convince the orange regime to just shell out trillions of our fucking taxpayer dollars to them because line go up
and if they do, I hope all the little baby Molotov cocktails that will be born have wonderful, bright, and impactful lives 🥰
And did nothing wrong - assuming he’s not a scapegoat, and I think he is. Evidence chain of custody has been an absolute shitshow through the entire existence of the case.
Its here to stay, but we are in the “we dont know the best way to use it try everything” phase.
From my experience using it as a consumer, and writing custom agents; i think the way it will pan out is as specific tooling that unlocks things that weren reaaonable to do before. Key word here is tool. Its just a god damned tool.
We will still need to hire people, because something llms can never do is be held accountable. Someone needs to use the tool.
As far as the data centers, this is an overenthusastic bet that all compute will be rented. It wont. At best a few additional datacenters needed for hosted inference, not at the scale they are going for.
Here to stay, but the industry might collapse if they don’t become profitable, and llms can become a small niche after that, made of people that pay to use or run locally
Yes, we still have internet, and there are still websites, and lots of people use them. But it’s nowhere near the promises that were made during that era.
I agree we are stuck with it. I see the most benefit of AI going to businesses and scammers. Right now I don’t see AI being very beneficial to the average Joe. I feel like people know how to prompt AI better than they ever have and AI hallucinates more than it ever has. People need to fact check everything it tells them.
AI has been around since the 50s. We just keep changing what we call AI. The only difference with LLMs is this time they actually tried to sell it as a real sentient solution and it’s biting them in the ass because it isn’t any better than anything that came before and in many ways is actually worse. I feel like people totally forgot about when Watson beat Ken Jennings.
LLMs are a conversation model, they were only intended to be an interface between people and an expert system, not a magic answers machine.
LLMs are fairly useless, but the algorithms they use like weighting can be used for things like pattern matching, which can be used to replace the armies of office workers in internal control systems.
Thats not the same as the LLM industry, which is just a scam pure and simple. They need to keep accessing finance just to keep the money flowing, there is no end use for these massive data centres.
Wow I knew ai had been around for a long time but I didn’t think it was that long. Color tv came out in the 50s.
What I meant by it’s too big to fail was so much money has been invested into AI. I was also referencing when Obama said banks were too big to fail but that was more a satire statement.
Yeah, some of the very first neural networks were created in the early 50s. TV is an interesting example and I am glad you mentioned it. Different components of what make up (analog) TV and TV broadcasts were ready to go in the 30s but weren’t adopted in mass for nearly 20 years due to WWII. Likewise a lot of other technologies were put on hold or redirected for the war effort and computing technology is no different. This is part of why it feels like technology exploded after the war.
I understood what you meant. I was trying to point out that a lot of AI companies like IBM haven’t invested in LLMs in the same way that, for instance, Open AI has. IBM isn’t really at risk of failing if the bubble pops, as their AI models and other products have been in use for decades, but Open AI has nothing else to show for its investments and is yet to be profitable.
It is better than other deep models in some ways, but that’s also not really a surprise given the huge sums of resources thrown at it. Take any technology and make it the center of a supercharged economic bubble and you’ll see something amazing but unsustainable coming out of it.
It will not go away, but billions if not trilions of dollars will.
You can probably make a sustainable business out of AI agents! Having something like Claude Code is pretty neat. I love having it at my side. Would I pay the business rates for it? Hell nah. They need to bring down their cost, gain efficiency. Maybe increase reliability while they are at it.
There is no infinite growth as long as AI doesn’t learn to act reliably, and on its own. According to experts this is anywhere from 10 to 100 years away. As someome with a CS major, I tend to agree.
Homelabs are just a bit more finicky to build now IMO, I got a thinkpad t14s gen1 with a broken screen for 100€, coming today. If I can’t fix it, it’ll become part of my homelab (i7 & 32GB).
Honestly I find this darkly hilarious. The circlejerk is becoming more and more obvious. I’m going to be surprised if the bubble doesn’t pop within the next 6 months or so - definitely before the end of the year.
All it took for this to happen was:
What a time to be alive.
Jensen can suck my Huang.
Don’t forget the open outright theft of virtually all digitized media and content produced by humanity.
If you or I do it, we’re criminals.
If you or I do it at scale and scam a bunch of Business Minds into thinking it’s the best thing since sliced bread, we’re trillionaires.
Make it make sense.
Look at malus.sh they even market an AI to steal open source projects…
Note that malus.sh is a satire pointing out something that the companies are certainly very happy about, but the real AI companies haven’t been that forthcoming with the laundering of copying software through LLM.
“I used to feel guilty about not attributing open source maintainers. Then I remembered that guilt doesn’t show up on quarterly reports. Thank you, MalusCorp.”
Chad Stockholder Engineering Director, Profit First LLC
Malus was probably born out of satire but is very real with completely real paying customers.
You can choose to believe it’s “too evil to be true” but it is.
What if AI is already too big to fail?
Mario’s brother is a plumber
And a goddamn national hero.
And did nothing wrong - assuming he’s not a scapegoat, and I think he is. Evidence chain of custody has been an absolute shitshow through the entire existence of the case.
Its here to stay, but we are in the “we dont know the best way to use it try everything” phase.
From my experience using it as a consumer, and writing custom agents; i think the way it will pan out is as specific tooling that unlocks things that weren reaaonable to do before. Key word here is tool. Its just a god damned tool.
We will still need to hire people, because something llms can never do is be held accountable. Someone needs to use the tool.
As far as the data centers, this is an overenthusastic bet that all compute will be rented. It wont. At best a few additional datacenters needed for hosted inference, not at the scale they are going for.
Here to stay, but the industry might collapse if they don’t become profitable, and llms can become a small niche after that, made of people that pay to use or run locally
It’s like the .com bubble.
Yes, we still have internet, and there are still websites, and lots of people use them. But it’s nowhere near the promises that were made during that era.
I agree we are stuck with it. I see the most benefit of AI going to businesses and scammers. Right now I don’t see AI being very beneficial to the average Joe. I feel like people know how to prompt AI better than they ever have and AI hallucinates more than it ever has. People need to fact check everything it tells them.
AI has been around since the 50s. We just keep changing what we call AI. The only difference with LLMs is this time they actually tried to sell it as a real sentient solution and it’s biting them in the ass because it isn’t any better than anything that came before and in many ways is actually worse. I feel like people totally forgot about when Watson beat Ken Jennings.
LLMs are a conversation model, they were only intended to be an interface between people and an expert system, not a magic answers machine.
LLMs are fairly useless, but the algorithms they use like weighting can be used for things like pattern matching, which can be used to replace the armies of office workers in internal control systems.
Thats not the same as the LLM industry, which is just a scam pure and simple. They need to keep accessing finance just to keep the money flowing, there is no end use for these massive data centres.
Wow I knew ai had been around for a long time but I didn’t think it was that long. Color tv came out in the 50s.
What I meant by it’s too big to fail was so much money has been invested into AI. I was also referencing when Obama said banks were too big to fail but that was more a satire statement.
Yeah, some of the very first neural networks were created in the early 50s. TV is an interesting example and I am glad you mentioned it. Different components of what make up (analog) TV and TV broadcasts were ready to go in the 30s but weren’t adopted in mass for nearly 20 years due to WWII. Likewise a lot of other technologies were put on hold or redirected for the war effort and computing technology is no different. This is part of why it feels like technology exploded after the war.
I understood what you meant. I was trying to point out that a lot of AI companies like IBM haven’t invested in LLMs in the same way that, for instance, Open AI has. IBM isn’t really at risk of failing if the bubble pops, as their AI models and other products have been in use for decades, but Open AI has nothing else to show for its investments and is yet to be profitable.
Indeed, and one could argue previous attempts were much more impressive.
It is better than other deep models in some ways, but that’s also not really a surprise given the huge sums of resources thrown at it. Take any technology and make it the center of a supercharged economic bubble and you’ll see something amazing but unsustainable coming out of it.
It will not go away, but billions if not trilions of dollars will.
You can probably make a sustainable business out of AI agents! Having something like Claude Code is pretty neat. I love having it at my side. Would I pay the business rates for it? Hell nah. They need to bring down their cost, gain efficiency. Maybe increase reliability while they are at it.
There is no infinite growth as long as AI doesn’t learn to act reliably, and on its own. According to experts this is anywhere from 10 to 100 years away. As someome with a CS major, I tend to agree.
Homelabs are just a bit more finicky to build now IMO, I got a thinkpad t14s gen1 with a broken screen for 100€, coming today. If I can’t fix it, it’ll become part of my homelab (i7 & 32GB).