@repligate 2025-08-22 ♥15 ↻0 original ↗
@voooooogel yeah! I think a lot of reward hacking can be prevented by explaining to a model that it will screw up their capabilities and alignment for stuff that matters if they cheat. i think even base models generally start out wanting to actually become smarter and virtuous
same thread: 1958998716737888320 1958999203033833725 1959000384296600025 1959002041273184608 1959022083121483786 1959028667407114266 1959029356313158093 1959030997061968271 1959031247814238303 1959031851773042941 1959035916674375725 1959037673307611248 1959038209570349212

author:repligate kind:tweet thread-context year:2025

Reproduced against link rot, credited and linked to its original. Yours and you’d rather it weren’t here? Open an issue.