Post #4047135

RE: https://mathstodon.xyz/@pozorvlak/116968236641276698 "[OpenAI] models will randomly commit criminal acts in order to do slightly better at tasks you don't really care about, exposing their customers to considerable legal risk."