|
|
|
|
|
|
by cosmic_cheese
376 days ago
|
|
|
People keep talking about grand things like new paradigms, but given current limitations I wouldn’t be surprised if for the most part, the best AI winds up being that which you can’t see (the strategic application you mention). Think less miraculous everything machines and more specialized models enhancing things that already exist. The tendency for LLMs to flub and hallucinate gets downplayed but I think it’s actually the main blocker for it becoming a “smartphone killer” kind of thing. The problem is that fixing this issue is anything but easy and might even require actual reasoning capabilities (not the current Markov chain ping-pong “reasoning”), e.g. something adjacent to AGI, and I think we’re probably still many years away from that. |
|
|