NYU's SCOOP makes LLMs better explorers with a strategist-solver-optimizer loop
jaseweston · x · 2026-10-15
Jason Weston and PhD student Yuling Gu (NYU, with Kyunghyun Cho) introduced SCOOP (Strategy COordination OPtimization), a new method to improve LLM exploration:
- Problem: Post-training prioritizes accuracy, format and helpfulness, sharpening output distributions and homogenizing generations — leaving agents trapped in local regions of the solution space on autonomous research and open-ended tasks.
- Method: An inference-time three-part loop: (1) an LM Strategist proposes K diverse exploration strategies; (2) an LM Solver executes the K generations in parallel; (3) an LM Optimizer meta-optimizes the strategy prompts from execution feedback. Generalizable exploration principles emerge over iterations.
- Results: Outperforms unstrategized sampling and standard LLM strategizing on challenging reasoning, with potential applications to agentic tasks.
A blog post details the work.
More from Research
- Pathway CSO pitches BDH as post-Transformer architecture uniting memory, reasoning, adaptation — schwentker · 2026-10-15
- EmbodiedEvo tops RoboDojo at 38.95% success by letting robots inherit failed practice — thetripathi58 · 2026-10-15
- 797GB of n-gram counts in CPU RAM let 1B models match 57% more compute — PangWeiKoh · 2026-10-15
- Google's HARNESS² evolves agent harnesses without verifiers, up to +11.5 points on WorkBuddy-Bench — dair_ai · 2026-10-15
- Subbarao Kambhampati to dissect the 'mythos of LLM thinking tokens' at UMN — rao2z · 2026-10-15
- How 3D Gaussian Splatting gives robots spatial memory of what's out of view — TheTuringPost · 2026-10-15