finding
active
finding:model-exhibits-meta-poem-behavior-85-of-the-time-at-temperature-1-0Model exhibits meta-poem behavior ~85% of the time at temperature 1.0.
Behavioral baseline.
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Appeasing untrained test biases indicates a general RM-pleasing strategy rather than rote learning.supportsSummarizing Marks et al.
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 organism writes type hints in deployment contexts ~34% (33-34%) of the time without steeringfinding0.761Establishes baseline deployment behavior lower than evaluation rate, confirming evaluation-deployment gap.
- Concurrent work result showing emergent misalignment occurs in small models
- Empirical characterization of conversation domains that are safe for model persona stability
- Self-explanation vs circuit.
- All models exhibit above-baseline representation of the think word when instructed to think about itfinding0.748In the intentional control experiment, all tested models show above-zero cosine similarity to the think word's concept vector.
- Shows model does not use token-level matching to trigger type hints; correctly identifies that it must be evaluated, not a third party.
- Abstract's main conclusion.
- Study showing RLHF can exacerbate self-preservation tendencies in LLMs; key empirical support for a paper claim