Tencent's Terminal-RSI: a 27B model that rewrites and retrains itself with zero humans
Scobleizer · x · 2026-11-08
Robert Scoble shares a breakdown of Tencent's Terminal-RSI, a 27B pipeline for recursive self-improvement that tackles agent "laziness" — agents claiming success on partial progress and failing the final environment verifier.
It works as a recursive loop with two alternating phases: RSR (recursive self-rewrite), where the model fine-tunes on its own successfully rewritten trajectories, and RPR (recursive policy-refine), which then runs reinforcement learning (PPO). The pipeline forces agents to actually finish the job, billed as a notable step toward recursive self-improvement.
More from Models
- LightOn releases LightOnOCR-3: 0.8B-4B open-source OCR models under Apache 2.0 — IgorCarron · 2026-11-08
- Abliterated GLM's "I am compressed humanity" self-description goes viral — RileyRalmuto · 2026-11-08
- antirez: DwarfStar now runs Kimi K3.1 PRO inference from safetensors via SSD streaming — antirez · 2026-11-08
- teortaxes: Plugin creation in DeepSeek Hub feels impressively natural — teortaxesTex · 2026-11-08
- Weight Space Programming: Claude Opus 5.5 beats GPT-6 Astra at hand-written 18-weight parity transformer — RobertTLange · 2026-11-08
- Benchmaxxed again? Dev says Qwen 3.8 2.4T flopped in real use vs smaller GLM 5.3 — serious_mehta · 2026-11-08