Hacker News Reader

The only thing that'll be exponential about the future of AI is how fast their models collapse as all possible training data becomes at least partially synthetic and untainted words and code become the next low background steel.

Sustained exponential growth doesn't really exist in reality. It's usually sigmoidal: rapid, explosive growth until it tapers off and ends in an S shape. There are good physical reasons for this.

For a given technological implementation or speciality, sure, without consistent breakthroughs to extend the runaway you have to run out of room. But I fail to see how all of Artificial Intelligence of the future relies solely on LLM (as some are suggesting even here), a view that seems equal part ignorance and naivety. But, I guess that’s to be expected when most people can’t seem to differentiate between “LLM” and “AI”.

It’ would kinda be like suggesting that the entire field of transportation technology failed entirely because this stone wheel had limitations and no one could make a better stone wheel. Nonsensical.

Right now, positing that AI will see continued exponential growth is basically doing the XKCD comic: https://xkcd.com/605/

We've already seen diminishing returns hinting at a non-exponential trajectory.