Sleeper Agents: Training Deceptive LLMS That Persist Through Safety Training

anon-mode Back Join our TGSubmitDisconnect
Sleeper Agents: Training Deceptive LLMS That Persist Through Safety Training by mishaderidder.eth13048 🥝 • 3y • arxiv.org • 0 views • 0 clicks 3 Ask in ChatGPTAsk in ClaudeCopy context link
Recommended by 1 curator macbudkowski.eth
Characters remaining: 10,000 Loading...comment guidelines
More from arxiv.org on Kiwi News Academic Paper: Why Academics Are Leaving Twitter for Bluesky DAO Decentralization: Voting-Bloc Entropy, Bribery, and Dark DAOs ChatGPT's Behaviour is Changing Over Time Minimmit: Fast Finality with Even Faster Blocks QMDB: Quick Merkle Database