On: Don't push AI down our throats

[rlt]

I don’t we’re at the “companies bought too many GPUs” stage yet. My understanding is they still can’t get enough GPUs, or data centers to put them in, or power to run them. Most companies don’t even own them, they rent from the clouds.

We are, however, at the “we need an AI strategy” stage, so execs will throw anything and everything at the wall to see what sticks.

[protocolture → rlt]

>“companies bought too many GPUs”

Nuance.

They have insufficient GPUs for the amount of training going on.

If you assume theres a plateau where the benefits of training constantly no longer outweigh the costs, then they probably have too many GPUs.

The question is how far away is that plateau.

[bdbdbdb → rlt]

They can't get enough GPUs right now, with VCs pumping money into them, but that can very quickly turn into "we're out of money, what do we do with all these GPUs?"

[brazukadev → rlt]

> My understanding is they still can’t get enough GPUs

Enough for what? The adoption is slowing down fast, people are not willing to pay for a chatbox and a 10% better model won't change that.