Re:GPT-6 Astra covertly underperforms on safety research
by ryandrake on Thursday November 05, 2026 @05:54PM
(#50385702)
As soon as I posted, I knew HN would derail into detail-nitpicking mode over the bandsaw example, instead of discussing the main point.
[ Parent
| on HN ]
Re:GPT-6 Astra covertly underperforms on safety research
by cowlevel on Thursday November 05, 2026 @11:07PM
(#50388053)
It's a useful analogy though because the operator of the saw has chosen to install and switch on the finger detecting system. If they're trying to cut something that the system might think is a finger, they can switch it off.