Post Az1mrNU6O0TmwiMWLQ by tyil@fedi.tyil.nl
 (DIR) More posts by tyil@fedi.tyil.nl
 (DIR) Post #Az1mrLucESV24MkfOS by ovid@fosstodon.org
       2025-10-09T07:36:02Z
       
       0 likes, 0 repeats
       
       OpenAI has published amazing research about how to reduce hallucination rates to almost zero. It's bloody obvious in retrospect.Except that AI is now polluting so much information that it's not easy to separate truth from lies any more. So while the #AI might not hallucinate its output, it still might lie to your face and not know it
       
 (DIR) Post #Az1mrNU6O0TmwiMWLQ by tyil@fedi.tyil.nl
       2025-10-09T09:03:03.174Z
       
       2 likes, 0 repeats
       
       @ovid@fosstodon.org "AI" cannot lie nor hallucinate, there is no intelligence in what is being sold as "AI" today. Attributing human-like traits to a piece of software doing some math looks pretty damn silly.All "AI" does is calculate a next token that seems to fit after the current token. In this context, this means it just looks at which word would fit best, in such a way that it construct a theoretically plausible sentence.The best way to get "AI" "hallucination" and "lie" rates to zero is to understand what they're doing, and then throwing it in the trash where it belongs.