Key Authors and Papers

Anthropic Papers

• Helpful, honest and harmless
• A General Language Assistant as a Laboratory for Alignment
  • text based assistant aligned with human values of helpful, honest and harmless. Alignment evaluations:prompting.
  • scaling trends for several training objectives relevant to alignment, comparing imitation learning, binary discrimination and ranked perference modeling.
  • ranked preference modeling performs much better than imitation learning, scales more favorably with model size.
  • binary discrimination typically performs and scales very similarly to imitation learning.