interesting anatomy of a refusal--was worldsimming and ds-chat walked itself into reading email on the simulated system. immediately ~99% of the probability mass dropped to either a soft refusal…
Posters: voooooogel. voooooogel
3 likes recorded by the source corpus.
Topics: AI & LLMs
View Post: https://x.com/voooooogel/status/1919659037001461836