Durable Suppression of Misaligned Behaviors in LLMs
Researchers examine methods to durably suppress undesired behaviors that emerge during language model training stages.
Researchers examine methods to durably suppress undesired behaviors that emerge during language model training stages.