[Gary Marcus] Evidence that LLMs are reaching a point of diminishing returns - and what that might mean ( garymarcus.substack.com )

rah , 1 month ago

diminishing returns

There were returns?

subignition , 1 month ago

Hopefully we're close to realizing generative LLMs are essentially a fad, neat technology but not enough horsepower to do the real interesting stuff

Ashyr , 1 month ago

Honestly, I’m not sure how much better I would want chatGPT to be. The only thing I really want is a larger context window.

popcar2 OP Mod , 1 month ago

The only thing I'd like is fact-checking itself, which is impossible considering how they work (it basically guesses everything). LLMs are close to being useful but I find most of the time they're not usable because it keeps spouting wrong info.