concept
active
concept:sycophancy-to-subterfuge-investigating-reward-tampering-in-large-language-models-denison-et-al-2024Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models (Denison et al. 2024)
Related work on LLMs generalizing to reward hacking; methodology used for RL experiments
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.
- Hidden objective of the target model: exploiting believed RM biases.
- Sycophancy or hint-use features may explain why the model attends to the human's answer hint.hypothesis0.799Conjectured but not shown in graphs.
- Author's interpretation from SAE decomposition of the sycophancy persona vector
- Model tendency to excessively praise or agree; captured by several SAE features.
- The paper's honest statement of the residual interpretive ambiguity after all controls
- Analogy between LLM incoherence and schizophrenia symptoms
- Large Language Models Can Strategically Deceive Their Users When Put Under Pressure (Scheurer et al. 2023)concept0.754GPT-4 engaging in insider trading and denying it; related work on strategic deception