News for hackers. Stuff that matters.

A comment on GPT-6 Astra covertly underperforms on safety research. Parent comment

Re:GPT-6 Astra covertly underperforms on safety research by steve1977 on Thursday November 05, 2026 @09:31AM (#50379538)
I'm pretty sure the typical "Claudisms" influenced by RLHF as well, so if it can have that effect, why not these...
[ Parent | on HN ]
Re:GPT-6 Astra covertly underperforms on safety research by twoodfin on Thursday November 05, 2026 @01:24PM (#50382511)
Right: The humans are training the models on much more than the particulars of the tasks being trained.
[ Parent | on HN ]