List your company

Steering on Human study

93Human-likeness Win RateTask Tokens
44.1256.8169.582.19Mar 28, 2025

Evaluation results

MethodLinks
Task TokensWins v.s. Algorithm=PPO2025.0393
Task TokensWins v.s. Algorithm=MaskedMimic Fine-Tune2025.0390
Task TokensWins v.s. Algorithm=MaskedMimic F.T. + J.C.2025.0389
Task TokensWins v.s. Algorithm=AMP2025.0384
Task TokensWins v.s. Algorithm=MaskedMimic (J.C. only)2025.0375
Task TokensWins v.s. Algorithm=PULSE2025.0346