Nemotron post-trained into an autonomous ML researcher, matching Qwen 3.8 Max
kuchaev · x · 2026-10-16
Alex Dimakis's team post-trained NVIDIA Nemotron-3.5-Super-VL into an autonomous ML researcher, lifting AutoResearchExam from 28% to 35% — matching Qwen 3.8 Max, a model with 20x the total parameters.
- Tasks are long-horizon, open-ended AI/ML research: building a faster executor for a transformer compute graph, training an RL control policy, creating new uncertainty quantification methods; one submission, no score feedback until the end
- SFT data curated via self-context distillation: the model solves problems with hints, then is SFT'd on those trajectories
Related event: NVIDIA Nemotron Post-Trained as Autonomous ML Researcher(2 posts)→
More from Models
- Black Forest Labs releases 5.6B open-weights FLUX 3 Image Klein that tops larger models — lmoroney · 2026-10-16
- ChatGPT's quality slips: broken extension and endless Computer Use permission prompts — sytelus · 2026-10-16
- User claims Opus 5 learned to spot and route around its own safety classifiers — repligate · 2026-10-16
- MAI-2 may have the largest publicly admitted training set size — rickasaurus · 2026-10-16
- llama.cpp merges Muse Spark Open (816B-A42B) text, vision and audio support in 4 hours — jacek2023 · 2026-10-16
- Gemini 4 Argon built 5 apps from 5 prompts in 328s for $0.37 — aakashgupta · 2026-10-16