habryka comments on Large Language Models can Strategically Deceive their Users when Put Under Pressure.

[habryka]

Mod note: I cleaned up the formatting of the post a bit, to make it clearer what it’s linking to.