LLMs are really quite limited

They're alright for "draft me an email that says this...". But they're dreadful at research. They make things up. They get basic things wrong, even when you tell them its not right. They CAN be good at summarising large bundles of text, but even then they miss things and make things up, so you can only do that if you want a general sense of what is in them. And LLMs dont think....they just report back the opinions of whatever is in the training data.

Nobody is going to pay $100 a month to subscribe to an LLM given this. Are they? So I dont see how they ever make a return on all of these vast investments in compute/chips, given that chips last what.....8 years max, probably less....so these trillions of investment will need to be reinvested over and over as they dont last.

All these AI companies better have something much f**king better they're working on in the lab, or they're all f**ked.

Posted By: Under soil heating, Nov 13, 11:41:32