Post #2880009

2025-08-09 22:37 UTC

@cjd@pkteerium.xyz @cleantext@fosstodon.org @denschub@mastodon.schub.social I highly doubt that all possible forms of AI have the failure mode of answering incorrectly, and AI is far from the only way to use a computer. Many kinds of software, from a spreadsheet to a SAT solver, help humans be more rigorous and make less mistakes. We don't need to pursue LLMs.

Replies (1)

@cjd@pkteerium.xyz 2025-08-09 22:43

Oh right indeed, I read your post and mentally inserted LLM around the "intellectually honest" part. Actually that's kind of a microcosm of the same thing: Intellectually honest software is like simple unix philosophy, open source, etc. Has longevity, continues to endure and evolve. Intellectually dishonest software is like overcomplicated, meant to sell support, planned obsolescence, etc. Has very poor longevity, gets replaced (maybe replaced by different dishonest software, but replaced nonetheless). My argument re LLMs is that we will hit a point at which NOT trying to be as honest as possible while building them will cause the above effect to take place within seconds, and the thing will just collapse.

Open ##2880017