Stripe Blog Details Two-Stage SFT Plus On-Policy Distillation for Better Fraud Decisions
emilygsands · x · 2026-11-20
The companion Stripe Dot Dev blog post (by Hangzhi Guo) behind the "39% higher recall" tweet: adding explanations directly to SFT hurts decisions because token-uniform loss lets long explanations dominate the gradient. The fix is two-stage SFT—train on labels+explanations first, labels alone second—then on-policy distillation with reverse KL to recover explanation quality. Fraud review achieves 39% higher recall at human-review precision, with explanations beating the teacher's.
Related event: Stripe Boosts Fraud Review Recall 39% by Learning from Explanations(2 posts)→
More from coding & agent
- Rex now works over SSH with one-command server install — josh_wills · 2026-11-20
- Seroter Daily #913: Always-on Agents, Envoy Agentic Stack, Tool Retrieval in Gemini API — rseroter · 2026-11-20
- Blogger's Take: GPT-6.2 Astra as Orchestrator Plus Gemini 4 Neon Is the Best Coding Combo — haider1 · 2026-11-20
- Software factories underdeliver: revisiting the 'autonomous product org' loop design vision — pzakin · 2026-11-20
- Firecrawl goes local: Rust rewrite scrapes 37x faster with 11MB idle memory, fully open source — devdigest · 2026-11-20
- Stripe's Fraud Model Gets 39% Higher Recall by Training on Explanations in the Right Order — emilygsands · 2026-11-20