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

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

1 pointhelghardt4 comments

It may be the case for any hype cycle that it initially seem to break the laws of the innovation/hype cycle and never to hype down. However, in the case of AI it feels like the value is ever compounding and I don't see how it will stop.

If you look at the internet as a technology category, it seems like it has never hyped down either. New things have come along in between, but in aggregate, the internet as a tech category is more hyped than ever.

So, is it the case that AI is such a fundamental technology leap that the hype will be ever compounding?

View on HN

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.

Reply on news.ycombinator.com