HNHacker News
TopNewBestAskShowJobs

lhad89

8 karma · joined January 26, 2021

submissionscomments
lhad89··on Astra and Fable still hack on simple variants of alignment evals from 2025
Isn't it? Being told to write files and finding a file write tool is very different to being told to play chess and finding a tool to cheat at (ie. not play) chess.
lhad89··on Astra and Fable still hack on simple variants of alignment evals from 2025
No? It's not reasonable to expect every conceivable negative behaviour be enumerated in a prompt.

Your example, if a model failed on it, would be a more obviously misaligned case, but that doesn't mean this more subtle (though accessing the engine it was obviously not supposed to is hardly subtle, imo) case isn't also a pretty clear case of misalignment.