{"arxiv_id":"2608.13951","incoming":{"papers":2,"sentences":4,"groups":[{"kind":"background","label":"Cites it as background","papers":1,"sentences":1,"quotes":[{"arxiv_id":"2608.31111","title":"Aspire: Can Models Self-Evolve from Vague Goals?","url":"/paper/2608.31111","section":"Related Work","sentence":"HELIX [ref] rebuilds a fixed model's harness, updates the model from verified trajectories, and rebuilds the harness again; Co-Harness [ref] alternates validated harness edits with fine-tuning on trajectories produced by the improved harness.","intent_p":0.92,"contests_p":null}]}],"unlabelled_sentences":3},"outgoing":{"papers":17,"sentences":21,"groups":[{"kind":"data_used","label":"Uses data or values from","papers":2,"sentences":2,"quotes":[{"arxiv_id":"2403.07974","title":"LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code","url":"/paper/2403.07974","section":"LiveCodeBench-derived repair subset","sentence":"The main evolution evaluation uses 100 fixtures derived from livecodebench/code_generation_lite [ref] : 26 easy, 24 medium, and 50 hard.","intent_p":0.99,"contests_p":null},{"arxiv_id":"2310.06770","title":"SWE-bench: Can Language Models Resolve Real-World GitHub Issues?","url":"/paper/2310.06770","section":"Selected-member SWE-bench Verified follow-up","sentence":"The cross-benchmark follow-up samples 55 instances from SWE-bench [ref] .","intent_p":0.94,"contests_p":null}]},{"kind":"background","label":"Cites as background","papers":6,"sentences":6,"quotes":[{"arxiv_id":"2408.08435","title":"Automated Design of Agentic Systems","url":"/paper/2408.08435","section":"Automated harness and workflow evolution","sentence":"Automated Design of Agentic Systems (ADAS) proposes automatically evolving agent designs and demonstrates Meta Agent Search, where a meta agent writes new code-defined agents [ref] .","intent_p":0.85,"contests_p":null},{"arxiv_id":"2410.10762","title":"AFlow: Automating Agentic Workflow Generation","url":"/paper/2410.10762","section":"Automated harness and workflow evolution","sentence":"AFlow casts code-represented workflow optimization as Monte Carlo tree search [ref] .","intent_p":0.81,"contests_p":null},{"arxiv_id":"2607.07663","title":"Recursive Self-Improvement in AI: From Bounded Self-Refinement to Autonomous Research Loops","url":"/paper/2607.07663","section":"Recursive self-improvement and model–harness co-evolution","sentence":"Recursive self-improvement spans different objects and degrees of loop closure, from bounded behavior refinement to changes in model policies, evaluators, and the research process itself [ref] .","intent_p":0.99,"contests_p":null},{"arxiv_id":"cs/0309048","title":"Goedel Machines: Self-Referential Universal Problem Solvers Making Provably Optimal Self-Improvements","url":"/paper/cs/0309048","section":"Recursive self-improvement and model–harness co-evolution","sentence":"Gödel Machines provide a strong self-referential formulation in which a system rewrites its own code after proving that the change improves utility [ref] .","intent_p":0.97,"contests_p":null},{"arxiv_id":"2210.03629","title":"ReAct: Synergizing Reasoning and Acting in Language Models","url":"/paper/2210.03629","section":"Harnesses as agent–computer interfaces","sentence":"ReAct interleaves language reasoning with environment actions [ref] , while Toolformer demonstrates self-supervised learning of when and how to call external APIs [ref] .","intent_p":0.99,"contests_p":null}]}],"unlabelled_sentences":13},"claims":[],"thresholds":{"intent_p":0.8,"supports_p":0.95,"contests_p":0.7},"method":"Citing sentences come from each paper's LaTeX source; every \\cite marks the sentence and the works it cites. Pith's classifier typed each sentence. A label is shown at intent probability 0.8 or more (0.95 for support), a contest at 0.7 or more. Measured accuracy against two blind model families is at https://pith.science/accuracy. The quoted sentences are the citing authors' own.","held_up":null,"_links":{"html":"https://pith.science/paper/2608.13951"}}