Post #3209134
2026-06-03 06:34 UTC
@screwlisp@gamerplus.org @kentpitman@climatejustice.social @yantar92@fosstodon.org
"fancy report ... versus having an ai ..."
Two problems here: the cost of the ai is currently being massively subsidised and the LLMs can't not halucinate, so verifying that the ai report is accurate has costs.
The AI stuff in a specific domain where it's possible to have a domain model works (Go, protein folding), but has to be redone for each domain.
As someone who applied to college to study AI in 1971, neither LLM nor domain model games are AI.
Replies (1)
-
@djl@mastodon.mit.edu 2026-06-03 06:43
@screwlisp@gamerplus.org @kentpitman@climatejustice.social @yantar92@fosstodon.org My guess here is that the really large LLM stupidity is going to fizzle, and the game is going to be small models that either run locally, or run really cheaply in the cloud. Several monthes ago my friend Larry (hi Larry!) showed me a small model running on a friggin' iphone, and although I didn't admit it to him, I was blown away. Later a music YouTuber talked about this in detail, with small models specialized for what you are doing. This seems sensible.