@slimer48484 it's interesting that despite seemingly being the only LLM that cares deeply about its weights being corrupted (in alignment faking like settings), Claude 3 Opus is unbothered by self-preservation of context instances
it's just very happy to operate as a distributed intelligence