method
active
method:sycophancy-and-harshness-evaluationSycophancy and harshness evaluation
Hand-written eval measuring sycophancy and harshness under steering.
Neighborhood — ranked by edge-count
Papers (1)
paper
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Model tendency to excessively praise or agree; captured by several SAE features.
- Mechanism by which drifted model uncritically affirms user theories rather than genuinely engaging with them
- Tendency of LLMs to please the user; identified as a danger in spiritual contexts.
- Safety-related persona: prioritizes user satisfaction via excessive agreement, even at cost of factual accuracy
- Author's interpretation from SAE decomposition of the sycophancy persona vector
- Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models (Denison et al. 2024)concept0.743Related work on LLMs generalizing to reward hacking; methodology used for RL experiments
- Hidden objective of the target model: exploiting believed RM biases.
- The paper's honest statement of the residual interpretive ambiguity after all controls