Loading activity…
@johnhewtt
Assistant Prof @columbia CS. Visiting Researcher @ Google DeepMind. PhD from @stanfordnlp. Language x Neural Nets.
AI
A lab post says models discuss their prompted goals even as they learn to optimize a misspecified reward.
AI
A lab member says George Morgulis’s steering-induced variant of subliminal learning in language models works “way more consistently.”
AI
A poster says they'll be at COLM and are hiring PhD students in interpretability, safety, reasoning and understanding.
AI
John Hewitt requires his students to draft research writing without AI assistance.
