DeepSeek-R1 — Reasoning via RL (2025)
The open recipe for reasoning models — a Nature cover, from a hedge fund's lab.
Anda sedang melihat entri makalah penelitian: sebuah makalah penting dan apa yang diperkenalkannya ke bidang ini.
The R1 report showed reasoning emerging from pure reinforcement learning on verifiable problems (R1-Zero), then a practical pipeline mixing SFT and RL — published openly with weights while Western labs kept their recipes closed. It democratized the o1 breakthrough within weeks and later ran as Nature's cover paper.
Bagaimana ini terhubung ke bagian lain dari stack
Tautan catatan ini sekilas — klik grup untuk langsung menuju ke sana.
Diperkenalkan pada
Terkait
Sumber
Rekaman ditinjau per 2026-01-15 — lihat Metodologi.