U.S. labs lean on RL while Chinese labs favor SFT on successful traces
goyalshaliniuk · x · 2026-07-22
A short observation on training trends across labs:
- U.S. AI labs appear to rely on RL more often.
- Chinese labs appear to use SFT on successful traces more often, and the author says this can be very competitive.
- Kimi K3 is cited as soft evidence for that claim.
The post is not a formal result, but a concise industry-level hypothesis about how reasoning models may be trained.
Related event: Divergent Sino-US Training Paradigms and New SFT Alignment Ideas(5 posts)→
More from Research
- MissionReport benchmark: only 3 of 540 AI-generated government reports pass all criteria — Exp_Mark · 2026-10-27
- Finch 1: a geospatial foundation model trained on imagery from 50+ sensors — Scobleizer · 2026-10-27
- Skyfall AI's SKY-001 world model turns sketches into CAD builds in 3.9s, vs 199s for GPT-6 Astra — rohanpaul_ai · 2026-10-27
- Physicists Let AI Pick Its Own Problem—It Wrote a Publishable Paper in 2 Hours — burny_tech · 2026-10-27
- Researcher flags unresolved data transparency issues in compositionality evals — najoungkim · 2026-10-27
- Separate Forms from Estimators: Why Steer by Difference of Means When You Can Use SGD? — johnhewtt · 2026-10-27