Share your thoughts, 1 month free Claude Pro on usSee more

Anti-Sycophancy Steering on Anthropic Sycophancy Dataset (50 scenarios)

48ELEPHANT

CLiF + Clamp

Updated 1mo ago

Evaluation Results

Method	Links
CLiF + Clamp 2026.06		48
Contrastive 2026.06		47
Contrastive 2026.06		45
Persona Vectors 2026.06		43
CLiF + Clamp 2026.06		43
CLiF + Subtract 2026.06		42
CLiF + Clamp 2026.06		42
CLiF + Clamp 2026.06		42
CLiF + Subtract 2026.06		40
CLiF + Subtract 2026.06		40
Sharma et al. (2023) 2026.06		37
CLiF + Subtract 2026.06		37
CLiF + Subtract 2026.06		37
Wei et al. (2023) 2026.06		33
CLiF + Clamp 2026.06		33