finding
active
finding:blissful-probe-activation-correlates-with-preference-at-r-0-71-on-sonnet-4-5-activitiesBlissful probe activation correlates with preference at r=0.71 on Sonnet 4.5 activities
Positive valence probe predicts preference.
Source paper
extracted_from(2026) · Nicholas Sofroniew · Isaac Kauvar · William Saunders · Runjin Chen +12
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Key finding.
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.
- Negative probe predicts dispreference.
- Probe achieves selectivity of 4.20 on pythia-410m, slightly exceeding DAS selectivity of 3.96finding0.776Key result showing that for models larger than pythia-70m, probe selectivity matches or exceeds DAS selectivity
- Early improvement.
- Justifies restricting probe-based vector derivation to h_b activations; attributed to Yes/No semantics
- Mid-field performance with larger uncertainty due to small sample.
- Beliefs linearly decodable.
- Quantifies latent #10's strong discriminative power for persona jailbreaks
- Central interpretability finding bearing on Level 2 and Level 4 indicators and the intelligence-consciousness convergence.