Keeping language models honest by directly eliciting knowledge encoded in their activations.
Primary LanguagePythonMIT LicenseMIT