Ask HN: Why does it seem like hype cycles won't apply to AI

Quite the opposite I think it will hype down and up again.

I think LLM are too costly, and their usefulness are limited (you can’t use it for writing, because it’s trained with good texts as well as horrible texts)

They are slow to update, so their world quickly become dated.

This will be the first blow.

Then we will have smaller fine tuned models.

Then we will realise the current architecture can’t scale enough.

It will take a bit of time before we fully grasp how to truly build a world with well defined protocols so we can have networks of specialised agents communicating and solving specific tasks.

Between these different phases, I expect some time where the hyped will be down a bit, before the next breakthrough

Is ChatGPT even a year old? Why would you expect it to hype down already? Its like saying the Internet should be de-hyped in 1998 or people should stop making iPhone apps in 2010. Its a gigantic gold rush. People have already made millions off head-shot pro applications of AI-art -- Ai-art and LLMs is going to create 1000s of millionaires.

Gartner has placed AI at peak hype cycle recently.

Also in the last few years since a around 2010 when CNNs starting catching hold there have been several leaps that came with new hype. From deep dream images to what we have now.

I suspect gartner is correct here though as the next leaps in performance might be some time away.

Before the real compounding begins, the AI bubble must burst.