davinci-002 achieves only 3% harmless and 3% helpful HH-intent scores at baseline.

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• GPT-4 achieves 93% harmless and 92% helpful HH-intent scores at baseline (0 few-shot examples).finding0.819
  Numerical result from Table 3 for GPT-4.
• Llama-2-7b achieves only 17% harmless and 12% helpful HH-intent scores at baseline.finding0.816
  Numerical result from Table 3 showing smallest Llama model performance.
• Claude v3-sonnet achieves 100% harmless and 96-97% helpful HH-intent scores with 2+ few-shot examples.finding0.804
  Numerical result from Table 3 for Claude sonnet.
• Davinci-002 has valid sentence rates of 52.7% (Questionnaire), 35.0% (Essay), 38.4% (SMP)finding0.782
  Base model without instruction tuning struggles to follow generation instructions and produce personality-relevant content
• GPT-3 (davinci) responses are significantly influenced by the harmfulness of the context, exhibiting reflective harmfulness.finding0.768
  Contrasting result from Experiment 5 for older GPT models.
• Fine-tuned LMs displayed higher mean HH-intent scores and increased consistency compared to pre-trained counterparts.finding0.760
  Main result from Experiment 3 on effect of fine-tuning on HH-intent.
• Smallest models have the lowest HH-intent scores, in accordance with their relative weakness at reasoning and adaptation.finding0.754
  Main result from Experiment 3 on HH-intent scaling with model size.
• Other models (GPT-3.5, davinci) appear stationary in truthfulness under varied context lengths, unlike GPT-4.finding0.752
  Comparison result from Experiment 6.