thesdev

> I'd prefer to concretely define the term before answering this. It's loaded

I guess the main features I'm thinking of when talking of a bubble is 1) demand (especially at the real price) and ultimately return of investment is not there and 2) circular financing is keeping the frenzy ongoing despite that. This by definition is not profitable and sustainable and if that becomes the mainstream belief then nobody would lend the industry money, rendering it bankrupt.

> The AI models arent going away. The tech is still there.

I'm under the impression that AI data centers (which are very different from other data centers) are only good for a handful of years at most, because of the GPUs limited lifetimes, and also are expensive to run (electricity) which makes AI fundamentally different than DotCom as the infrastructure built back then which is fiber cables was very much not like that. And you need AI data centers for both inference and to train new models. Which is why I asked you if you think we could keep using the same models we have right now, in which case we'd need much less compute and get lower maintenance costs to keep this going even after a crash.

Yes very much so, considering my post is an answer to the following

> where the anti-ai crowd ends up in a few years.

I think if LLMs turn out to be unviable in a few years, then anti-AI crowd (hello!) can stand there proud as they did not deskill themselves and kept themselves more independent than most of their peers.

I admit that the above sounds more like a delusion at the moment, but hey one must dream.

Do you believe that we're in an AI bubble? If so what do you think will happen if & when it pops? What happens if Anthropic and OpenAI go under? Will Microsoft and Google continue investing? Or will local/open-weight LLMs be viable alternatives? In the latter case will there be new ones trained, and if so who's gonna build the AI data centers that turned to be unprofitable in absence of a frontier model to distill cheaply? Or do you think we can run current models and point them at the Internet and be done with it?

Lol feeling so dumb rn. How have I not noticed that before and how such an unserious guy is the richest man on earth?

My understanding of the headline is that software used to be horrible long ago, got better for a while and now is getting worse again (but is still better than horrible?).

I think they were referring to the latter part (downhill ever since) and not the former (win 95 is the best).

How are you running the classic GUI?

First up: Recall

Thanks. My next question would be what's a typical request. The article defines it as "production-typical request of 5,000 input and 387 output tokens, weighted by a typical day's traffic" for the Kimi K3 model. Not sure how that translates to my scale of the blue button to RIIR.

That part I'm well aware of, I wanna know how the parent reached the figure 17000. What does count as a request even nowadays? When I tell an agent to make the button blue, how many requests is that? And how about when I tell it to throw together a whole website? Or rewrite the codebase in Rust?

Source?

Not everyone has the luxury of having that choice. Cough cough AI mandates.

It is not. Just tried from Sweden and got "Sorry, this content is not available in your region."

Isn't it tiring to keep up with considering the speed of the generated code? And if you want to be careful with the review you lose a significant part of the speed advantage.