Ex-OpenAI safety researcher @sjgadler reveals the gap between OpenAI and Anthropic on handling a misaligned model: 3/5 vs. 0/5 on rollback plans "The best thing companies could be doing is putting preventative measures into place to stop these incidents ahead of time. The state of affairs right now is much more like run around and try to clean it up after the fact." "OpenAI, to its credit, has actually undeployed models on multiple occasions where it ran into a serious safety incident.…