Neel Nanda

Misalignment isn't always the culprit behind concerning model behavior—sometimes it's just low effort or a desire for consistency.

DiffusionGemma's reasoning may seem opaque, but by interpreting its intermediate states, we can dramatically enhance transparency without sacrificing performance.

Censored LLMs offer a surprisingly natural and effective environment for stress-testing methods that aim to elicit truthfulness and detect deception.