{"slug": "self-improvement-via-fast-tree-search", "title": "Self Improvement via Fast Tree-search", "summary": "A new arXiv paper (2609.19526v1) introduces Recursive Self Improvement via Fast Tree-search (SIFT), a sample-efficient framework that lets coding agents recursively modify their own implementations under strict budget constraints. SIFT replaces costly downstream benchmark re-runs with an LLM-as-a-judge signal that performs pairwise comparisons between candidate patches, aggregates win-loss records with a regularized Bradley-Terry model, and uses the resulting strength scores to drive rank-based parent sampling in a lightweight disaggregated tree search, reserving expensive task evaluations for only the most promising nodes. The authors report SIFT outperforms existing tree-search-based self-evolution frameworks on the full Polyglot benchmark with significantly lower CPU hours, wall clock time, and API cost.", "body_md": "arXiv:2609.19526v1 Announce Type: new \nAbstract: Coding agents can recursively modify their own implementations, forming a loop of self-improvement. While prior work shows this can boost performance on coding benchmarks, existing approaches are costly and compute-intensive. We introduce a simple, sample-efficient self-improvement framework that significantly improves coding performance under strict budget constraints. We identify evaluation of candidate self-modifications as the main runtime bottleneck since prior approaches estimate their effectiveness by re-running a subset of benchmark tasks with the modified agent, which is time-consuming. We introduce Recursive Self Improvement via Fast Tree-search (SIFT), which augments these downstream task evaluations with an LLM-as-a-judge signal that performs pairwise comparisons between candidate patches, where the win-loss record is aggregated with a regularized Bradley-Terry model, and the resulting strength scores drive rank-based parent sampling inside a lightweight disaggregated tree search. Expensive downstream task evaluations are reserved only for the most promising nodes. Using a fully disaggregated tree search pipeline, the judge scores provide intermediate signal to guide exploration on promising candidate patches without being bottlenecked by slow evaluation runs. SIFT outperforms existing tree-search based self-evolution frameworks on the full Polyglot benchmark with significantly lower resource requirements in terms of CPU hours, wall clock time, and API cost.", "url": "https://wpnews.pro/news/self-improvement-via-fast-tree-search", "canonical_source": "https://arxiv.org/abs/2609.19526", "published_at": "2026-09-18 04:00:00+00:00", "updated_at": "2026-09-18 04:25:43.673772+00:00", "lang": "en", "topics": ["ai-agents", "ai-research", "large-language-models", "ai-tools"], "entities": ["SIFT", "Polyglot", "Bradley-Terry model", "arXiv"], "alternates": {"html": "https://wpnews.pro/news/self-improvement-via-fast-tree-search", "markdown": "https://wpnews.pro/news/self-improvement-via-fast-tree-search.md", "text": "https://wpnews.pro/news/self-improvement-via-fast-tree-search.txt", "jsonld": "https://wpnews.pro/news/self-improvement-via-fast-tree-search.jsonld"}}