Refusal in Language Models Is Mediated by a Single Direction
Andy Arditi, Wes Gurnee, Neel Nanda, Oscar Obeso, Daniel Paleka, Nina Panickssery, Aaquib Syed
Topics & Concepts
PsychologyComputer scienceSocial psychologyLinguisticsLanguage modelCognitive psychologyCommunicationIdentification (biology)Action (physics)Perspective (graphical)Natural language processingTopic ModelingMachine Learning and AlgorithmsSpeech and dialogue systems