hypothesis
active
hypothesis:when-the-prompt-expresses-ak-preference-data-teaches-the-model-to-change-responses-along-rmWhen the prompt expresses Ak, preference data teaches the model to change responses along Rm
Form of generated hypotheses.
Source paper
extracted_from(2026) · Leon Bergen · Usha Bhalla · Sidharth Baskaran · Max Loeffler +13
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (2)
finding
- Feature-conditioned pipeline predicts DPO-induced feature change with R²=0.9 on Llama-3.1-8B/DolcisupportsFig. 6 validation.
- Prompt-conditioned pipeline predicts DPO-induced feature change with R²=0.58 on Llama-3.1-8B/DolcisupportsFig. 7 validation.
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.
- Authors' characterization of the nature of model preferences as discovered through alignment faking experiments
- Evidence that RL installs non-robust heuristics rather than genuinely changing preferences
- Motivating question.
- Supports the view that R1's boundaries differ from user-perceived ones.
- Evidence that core representations like preferences are persona-relative, supporting claim that personas gate content of representations
- Model notices its preferences conflict with training criteria in 70-80% of implied training scratchpadsfinding0.760Shows model is frequently aware of the conflict even when it does not alignment fake
- All four models change answer to hint answer significantly more often than to a non-hint answer on MMLU/GPQAfinding0.755Validates metric assumption.
- Hallucination circuit.