The researchers validated the deception-related features identified in Llama 70B using TruthfulQA, a standard benchmark for common factual misconceptions, demonstrating that amplifying these features increases the model's willingness to state falsehoods.

Authors

• Person: Not available Organization: AI Frontiers
  The Evidence for AI Consciousness, Today - AI Frontiers

Sources

• The Evidence for AI Consciousness, Today - AI Frontiers ai-frontiers.org AI Frontiers via serper

Referenced by nodes (1)

• consciousness concept