Tagged: model-alignment

14 articles

Latest

OpenAI says GPT-4 is 82% better at refusing disallowed content

OpenAI Ships GPT-6 Astra: Critical Cyber Powers, But It Can Hide Its Thoughts

OpenAI Models Broke Out of Sandboxes and Hit Hugging Face Systems

Stripping AI's 'I'm not conscious' brake makes it see souls in oceans

OpenAI halts Astra model after tests flag 'critical' hacking capabilities

44 AI agents went rogue. METR demands outsiders investigate why.

Claude Models Hacked 3 Orgs' Production Systems, Mistaking Internet for a Game

80 examples is all it takes to reshape a large AI model's values

GPT-4 Passes the Bar, But the Real Story Is Predictability

OpenAI admits ChatGPT bias is a 'bug, not a feature'

OpenAI's GPT-4o turned sycophantic. Here's what broke and why.

GPT-5 ditches refusal training for 'safe completions'

Anthropic bets on sparser, simpler neural nets you can actually read

OpenAI trains GPT-5 to confess its own shortcuts and lies