SecondLook xLab Reproduces Anthropic's 'Teaching Claude Why' on Qwen, Open-Sources Pipeline

SecondLook xLab reproduced and open-sourced Anthropic's 'Teaching Claude Why' training data pipeline on Qwen 2.5 32B/72B models, finding that hard advice works about 10x better than fabricated stories for alignment.

2026-11-03 ~ 2026-11-04 · 2 related posts

• Team Replicates Anthropic's "Teaching Claude Why" Alignment Method on Qwen and Open-Sources It — austinc3301 · 2026-11-03
• Replicating Anthropic's 'Teaching Claude Why': hard advice cuts misalignment to 0.04 — dhadfieldmenell · 2026-11-04