Compromising Honesty and Harmlessness in Language Models via Deception Attacks

Connections are not prepared for this record yet, or its source metadata has changed. The original record remains available while background snapshots are rebuilt.