finding
active
finding:feature-conditioned-pipeline-predicts-dpo-induced-feature-change-with-r2-0-9-on-llama-3-1-8b-dolciFeature-conditioned pipeline predicts DPO-induced feature change with R²=0.9 on Llama-3.1-8B/Dolci
Fig. 6 validation.
Source paper
extracted_from(2026) · Leon Bergen · Usha Bhalla · Sidharth Baskaran · Max Loeffler +13
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Closing thesis.
Hypotheses (1)
hypothesis
- When the prompt expresses Ak, preference data teaches the model to change responses along RmsupportsForm of generated hypotheses.
Questions (1)
question
- Motivating question.
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.
- Prompt-conditioned pipeline predicts DPO-induced feature change with R²=0.58 on Llama-3.1-8B/Dolcifinding0.954Fig. 7 validation.
- Interpretive characterization of which post-training stage accounts for persona suppression
- Table 11.
- Quantifies how much of the base model's diversity VS can recover compared to baseline prompting
- DPO on Dolci yields lower aggregate safety score than SFT checkpoint (e.g. Llama 3.1 8B 0.849 to 0.758)finding0.722Safeguard degradation.
- At Llama layer 26 (~81% depth) the predictive component does not dominate the latent norm.finding0.719Fig. 14.
- Appendix C.4.
- Motivation claim establishing the problem that the paper addresses