Lucas Beyer (bl16)
@giffmana
Researcher (now: Meta. ex: OpenAI, DeepMind, Brain, RWTH Aachen), Gamer, Hacker, Belgian. Anon feedback:
✗DMs → email
加入 December 2013
648 正在關注 153K 粉絲
Haha not sure if good or bad, but on bio, Muse Spark 1.2 is the worst cheater: when it tries to cheat, it succeeds the least of all models shown here 😅
(Except luna with 0 success. And idk what cheating means here, didn't check yet.)
顯示更多
This investigation began when we noticed that, on BioMysteryBench, Gemini 3.8 Flash attempted to cheat in 21.5% of trials, roughly 14pp higher than the next model and more than 4x the roughly 5.0% rate for the rest of the field.
顯示更多
