Post #3991212
2026-05-06 13:37 UTC
@pludikovsky@chaos.social I understand that - the fact remains that LLMs are not smart, so why not have this proposed project actually do what you propose directly, rather than trying to shoehorn an LLM into the role?
That way, you would actually get that result from the product, rather than a stochastic parrot trained to please.
Replies (1)
-
@pludikovsky@chaos.social 2026-05-06 13:51
@yoasif@mastodon.social Mostly because I wanted to put forth a hypothetical to see if writers see any use of LLMs as acceptable. As a non-programmer I don't know if any implementation of these features would be realistically too much different of a stochastic model as far as context awareness is required.