Gemini 4 Argon hallucination (!)
It's incredible how Argon is supposed to be the least hallucinating model. I just tried it to review some code Opus 5.5 wrote for game development, and it straight up invented code that didn't exist; Argon 4, high setting. I've never had nothing of that magnitude of error with either recent Sol/Astra or Opus/Fable models. (First I thought Opus had put in an invented skill to make something work while magic was not implemented, which it could have done, I asked it to put a rule in agents to not invent anything not yet in design docs, hence the rule comment.) Sticking in with Opus I guess 🤷♂️.
评论
?
参与讨论