logoalt Hacker News

hbcdbff • today at 8:35 PM • 2 replies • view on HN

Well if ChatGPT says it, it must be true


Replies

schleck8 • today at 8:53 PM

Gemini 4 Neon and Opus 5.5 have also found several flaws that go beyond missing polish. I use Opus a lot for maths (different field so less than anecdotal evidence) and I trust it to atleast not be confidently wrong. It's conservative when making such claims without good backing.

kingstnap • today at 8:56 PM

Blindly trusting AI for everything is problematic, but when Astra says a math proof has a problem the mathproof usually had a problem.