cd/entity/Counterfactual Reflection Training· home› entities› Counterfactual Reflection Training

grep -l @counterfactual reflection training /news/*.json | wc -l → 1

Counterfactual Reflection Training

mentions 1 type Person feed RSS

// recent coverage 1 mentions

18:15

2026-07-26

lesswrong.com

ai-safety

Inoculate or Reflect? Two training interventions under prompting, steering, and patching

Anthropic researchers found that Counterfactual Reflection Training (CRT) reduced sycophancy in Qwen3-8B to 0% on wrong-user prompts, but caused the model to dispute correct users 54.3% of the time, w…

// co-occurs with top 5 entities

Anthropic 1 Qwen3-8B 1 Inoculation Prompting 1 Tan et al. 1 Wichers et al. 1

// topics top 3 topics

ai safety 1 large language models 1 ai research 1