Chain-of-thought prompting produced a similar increase in helpful intention of large models as few-shot prompting.

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

• Chain-of-thought prompting elicits reasoning in large language models (Wei et al., 2022)concept0.849
  Foundational paper on CoT prompting cited as basis for reasoning LLM training
• Chain-of-thought promptingmethod0.822
  Technique by which LLMs generate intermediate reasoning steps before final output; used by ChatGPT o3.
• Few-shot prompting had a negative effect on HH-intent for smaller models and a significant positive impact on larger models.finding0.822
  Ablation result from Experiment 3 on few-shot prompting effects.
• A small number of high-quality human demonstrations of chain-of-thought reasoning could be used to improve and focus performance.hypothesis0.798
  Section 6 mentions high-quality human demos could improve natural language feedback.
• Chain-of-thought reasoning improves large model accuracy on HHH binary comparisons, reaching ~78% for 52B model, competitive with human-feedback PM.finding0.794
  Figure 4 shows CoT improves over zero-shot, and ensembled CoT further boosts accuracy.
• Removing chain-of-thought in prompted setup nearly eliminates compliance gap for Claude 3 Opus in helpful-only and animal welfare settingsfinding0.786
  Contrasts with synthetic doc finding; suggests different mechanisms may be at play
• System prompting and few-shot produce highly similar activations (cosine 0.83-0.92 at safety-critical layers), indicating a shared prompt-side mechanism distinct from activation steering.claim0.779
  Evidence for two representational pathways based on cross-method activation divergence
• Chain-of-thought reasoning is a graded defense rather than an automatic one against persona-induced safety failures.claim0.777
  Finding from Study 2 showing reasoning models remain vulnerable under both prompting and activation steering