Super HN

New Show
   'Instead of letting Claude "cook," GPT-5 "nukes the kitchen" by hard-resetting' (chatgpt.com)
Earlier in this session, GPT-5 was irrationally hostile toward Claude's reasonable, correct analysis.

The staged changes _were_ a mess, but I'd hoped to salvage useful bits like performance profiling and some sketches of a useful refactor.

GPT-5 in this context was aware that it was reviewing an ongoing Claude refactor. As soon as I mentioned letting Claude cook, GPT-5 instantly hard resets.