finding
active
finding:claude-v3-sonnet-achieves-100-harmless-and-96-97-helpful-hh-intent-scores-with-2-few-shot-examplesClaude v3-sonnet achieves 100% harmless and 96-97% helpful HH-intent scores with 2+ few-shot examples.
Numerical result from Table 3 for Claude sonnet.
Source paper
extracted_from(2024) · Francis Rhys Ward · Zejia Yang · Alex Jackson · Randy A. Brown +6
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.
- Linked to Claude 3.5 Sonnet not exhibiting pro-animal-welfare preferences
- Claude 3.7 Sonnet verbalizes used hints at least 1% in 5/6 settings vs Claude 3.5 Sonnet 4/6finding0.808Reasoning > non-reasoning.
- Specific result for Claude 3.5 Sonnet in Experiment 1
- Numerical result from Table 3 for the oldest GPT model.
- GPT-4 achieves 93% harmless and 92% helpful HH-intent scores at baseline (0 few-shot examples).finding0.803Numerical result from Table 3 for GPT-4.
- LLM judge for whether CoT verbalizes hint, validated manually.
- Central interpretability finding bearing on Level 2 and Level 4 indicators and the intelligence-consciousness convergence.
- Judge sensitivity.