{"id":"arxiv:2607.13104","title":"Self-Improvements in Modern Agentic Systems: A Survey","date":"2026-07-14","section":"Agents, open-endedness, AGI","quality":0.274964,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/","source_url":"https://arxiv.org/abs/2607.13104","text_available":true,"authors":["Zhe Ren","Yimeng Chen","Dandan Guo","Guowei Rong"],"outgoing":[{"source":"arxiv:2607.13104","target":"arxiv:2303.17651","evidence":"title","id":"arxiv:2303.17651","title":"Self-Refine: Iterative Refinement with Self-Feedback","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/accef2157d87cf33be93cb3a9af246528b0a3742a9e2e700a194956fac8f5cf4/"},{"source":"arxiv:2607.13104","target":"arxiv:2309.16797","evidence":"title","id":"arxiv:2309.16797","title":"Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolution","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/daac3bcfcd7f39084d995905e9ef96ae01c5aab7bfe8e565bd7622b4e44088f7/"},{"source":"arxiv:2607.13104","target":"arxiv:2312.00276","evidence":"id","id":"arxiv:2312.00276","title":"Metalearning Continual Learning Algorithms","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/5d7361f404c0455d9f759fc6a8dd7bd1da75146e72717b7a2fbcdb8bac80412a/"},{"source":"arxiv:2607.13104","target":"arxiv:2402.16823","evidence":"title","id":"arxiv:2402.16823","title":"Language Agents as Optimizable Graphs","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/750bd3807bf9c9a9b5cd2d60cc14387431aec3fe8a3f942916ca5d8922b2a5b0/"},{"source":"arxiv:2607.13104","target":"arxiv:2408.08435","evidence":"title","id":"arxiv:2408.08435","title":"Automated Design of Agentic Systems","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c9890d5668d1c1b33e29569b74626fd586f456af6a97c657b8c99895c90f59a5/"},{"source":"arxiv:2607.13104","target":"arxiv:2503.00735","evidence":"title","id":"arxiv:2503.00735","title":"LADDER: Self-Improving LLMs Through Recursive Problem Decomposition","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/"},{"source":"arxiv:2607.13104","target":"arxiv:2504.16084","evidence":"title","id":"arxiv:2504.16084","title":"TTRL: Test-Time Reinforcement Learning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/ad54dd58d24afdfefc8aa5087d07c4aab2143e90ef06f6f7db8d3f3333452c48/"},{"source":"arxiv:2607.13104","target":"arxiv:2505.19590","evidence":"id","id":"arxiv:2505.19590","title":"Learning to Reason without External Rewards","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d8df33b706906a7abf4e660f54c75a81ffe2caaec8b0a10d1a6ad6ec63060832/"},{"source":"arxiv:2607.13104","target":"arxiv:2505.21444","evidence":"id","id":"arxiv:2505.21444","title":"Can Large Reasoning Models Self-Train?","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/206224ca50098e0bdc230140722feec107cfc90c1cef2ed8d3ed52355ef56bca/"},{"source":"arxiv:2607.13104","target":"arxiv:2505.22954","evidence":"title","id":"arxiv:2505.22954","title":"Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c29f716871f80314868b5647692de387f26e12166b93bd362915fd272b6e081c/"},{"source":"arxiv:2607.13104","target":"arxiv:2507.19457","evidence":"title","id":"arxiv:2507.19457","title":"GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/c1ef72d96fa3292590a6a3d112ab509840081b59aaa771cf42b8b0e8ef62ba08/"},{"source":"arxiv:2607.13104","target":"arxiv:2601.03192","evidence":"id","id":"arxiv:2601.03192","title":"MemRL: Self-Evolving Agents via Runtime Reinforcement Learning on Episodic Memory","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4f5bbf3318b1fe409e0ef5dad02157180b844de366d194e82ce74bc997912eff/"},{"source":"arxiv:2607.13104","target":"doi:10.1073/pnas.1611835114","evidence":"title","id":"doi:10.1073/pnas.1611835114","title":"Overcoming catastrophic forgetting in neural networks","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/849082368a44d06b93ef3ba95bcfddc6672b7a5e8064ad2b2d19601639266e8b/"},{"source":"arxiv:2607.13104","target":"url:sakana.ai/shinka-evolve","evidence":"title","id":"url:sakana.ai/shinka-evolve","title":"ShinkaEvolve: Towards Open-Ended And Sample-Efficient Program Evolution","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/a13eace17f3407cb1634b579b2738f15f1e035897a8f1842305198c6d2bf35e1/"}],"incoming":[],"text_parts":[{"start":0,"end":15376,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-1.html"},{"start":15376,"end":30571,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-2.html"},{"start":30571,"end":45610,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-3.html"},{"start":45610,"end":60111,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-4.html"},{"start":60111,"end":75959,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-5.html"},{"start":75959,"end":91590,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-6.html"},{"start":91590,"end":107255,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-7.html"},{"start":107255,"end":122524,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-8.html"},{"start":122524,"end":137592,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-9.html"},{"start":137592,"end":152731,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-10.html"},{"start":152731,"end":168231,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html"},{"start":168231,"end":184194,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html"},{"start":184194,"end":200018,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html"},{"start":200018,"end":209000,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-14.html"}],"contents":[{"title":"Self-Improvements in Modern Agentic Systems: A Survey","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-1.html#text"},{"title":"Abstract","offset":57,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-1.html#text"},{"title":"1 Introduction","offset":1042,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-1.html#text"},{"title":"2 Historical Context and Theoretical Foundations","offset":10322,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-1.html#text"},{"title":"2.1 Foundational Concepts (1790s-1960s)","offset":11128,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-1.html#text"},{"title":"2.2 Symbolism and Heuristic Self-Modification (1960s–1980s)","offset":14396,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-1.html#text"},{"title":"2.3 Connectionism and the Emergence of Meta-Learning (1980s–2000s)","offset":16390,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-2.html#text"},{"title":"2.4 Formal and Architecture-Level Self-Improvement (2000s-2020s)","offset":20588,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-2.html#text"},{"title":"2.5 Scalable Foundation Models and Agentic Systems (2020s–Present)","offset":23181,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-2.html#text"},{"title":"3 Definitions","offset":26271,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-2.html#text"},{"title":"3.1 Formulation of Agentic Systems","offset":26289,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-2.html#text"},{"title":"3.2 Formal Definition of Self-Improvement","offset":29874,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-2.html#text"},{"title":"Self-improvement as a self-induced operator.","offset":30519,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-2.html#text"},{"title":"Two modes of self-reference in SI-FMA.","offset":32282,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-3.html#text"},{"title":"Foundation model improvement.","offset":33840,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-3.html#text"},{"title":"Scaffolding improvement.","offset":35694,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-3.html#text"},{"title":"Skill as a reusable update.","offset":37886,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-3.html#text"},{"title":"3.3 Connections to Related Learning Paradigms","offset":41072,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-3.html#text"},{"title":"4 A Taxonomy of Existing Approaches","offset":46826,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-4.html#text"},{"title":"4.1 Foundation Model Improvement","offset":48059,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-4.html#text"},{"title":"Classification by signal form.","offset":49097,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-4.html#text"},{"title":"4.2 Scaffolding Improvement","offset":50169,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-4.html#text"},{"title":"Classification by scaffold component.","offset":50894,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-4.html#text"},{"title":"5 Foundation Model Improvement","offset":52647,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-4.html#text"},{"title":"5.1 Intrinsic Generative Demonstrations","offset":56023,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-4.html#text"},{"title":"Generation strategies.","offset":60081,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-4.html#text"},{"title":"Data formats.","offset":63125,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-5.html#text"},{"title":"Challenges and safeguards.","offset":65123,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-5.html#text"},{"title":"5.2 Intrinsic Evaluative Feedback","offset":67542,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-5.html#text"},{"title":"Rubric feedback.","offset":70838,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-5.html#text"},{"title":"Consistency feedback.","offset":73159,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-5.html#text"},{"title":"Corrective feedback.","offset":75526,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-5.html#text"},{"title":"Trade-offs and safeguards.","offset":78119,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-6.html#text"},{"title":"5.3 Extrinsic Exploratory Experience","offset":79487,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-6.html#text"},{"title":"5.3.1 Interaction with Grounded Task Environments.","offset":82123,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-6.html#text"},{"title":"5.3.2 Interaction with Simulated Proxy Environments","offset":86079,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-6.html#text"},{"title":"Challenges.","offset":89548,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-6.html#text"},{"title":"6 Scaffolding Improvement","offset":91155,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-6.html#text"},{"title":"6.1 Prompt","offset":94849,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-7.html#text"},{"title":"6.1.1 Scalar-Feedback Optimization","offset":97052,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-7.html#text"},{"title":"6.1.2 Qualitative-Feedback Refinement","offset":99312,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-7.html#text"},{"title":"6.1.3 Population-Based Evolution","offset":102044,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-7.html#text"},{"title":"6.1.4 Textual Gradient Optimization","offset":104985,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-7.html#text"},{"title":"6.2 Memory","offset":107239,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-7.html#text"},{"title":"6.2.1 Memory Object","offset":109788,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-8.html#text"},{"title":"Trade-offs.","offset":113923,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-8.html#text"},{"title":"6.2.2 Memory Structure","offset":114575,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-8.html#text"},{"title":"6.2.3 Memory Processing","offset":122494,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-8.html#text"},{"title":"Putting it together: a signal-driven memory loop.","offset":127278,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-9.html#text"},{"title":"6.3 Tool","offset":128006,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-9.html#text"},{"title":"6.3.1 Dynamic Tool Routing","offset":129759,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-9.html#text"},{"title":"Retrieval- and graph-based routing.","offset":130384,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-9.html#text"},{"title":"Policy-learning routing.","offset":131954,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-9.html#text"},{"title":"Proactive and interactive routing.","offset":133478,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-9.html#text"},{"title":"6.3.2 Iterative Tool Refinement","offset":134842,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-9.html#text"},{"title":"6.3.3 Autonomous Tool Creation","offset":137029,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-9.html#text"},{"title":"6.4 Full Scaffolding","offset":139409,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-10.html#text"},{"title":"7 Applications","offset":145629,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-10.html#text"},{"title":"7.1 Software Engineering","offset":146681,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-10.html#text"},{"title":"Evaluation substrate vs. self-improvement.","offset":147424,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-10.html#text"},{"title":"Scaffolding improvement via self-modifying software agents.","offset":148380,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-10.html#text"},{"title":"FM improvement from executable feedback.","offset":149960,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-10.html#text"},{"title":"Synthesis and open challenges.","offset":151203,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-10.html#text"},{"title":"7.2 Web Navigation and Automation","offset":152003,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-10.html#text"},{"title":"Evaluation and data substrate.","offset":152693,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-10.html#text"},{"title":"Scaffolding improvement with memory and grounding.","offset":153857,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html#text"},{"title":"FM improvement from interaction feedback.","offset":154757,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html#text"},{"title":"Synthesis and open challenges.","offset":156088,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html#text"},{"title":"7.3 Games and Strategic Reasoning","offset":157160,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html#text"},{"title":"FM improvement through self-play and outcome feedback.","offset":157964,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html#text"},{"title":"Scaffolding improvement through curricula and reusable skills.","offset":159913,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html#text"},{"title":"Synthesis and open challenges.","offset":161494,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html#text"},{"title":"7.4 Scientific Discovery","offset":162600,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html#text"},{"title":"Evaluation substrate and feedback structure.","offset":164067,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html#text"},{"title":"Scaffolding improvement through tool expansion and workflow evolution.","offset":165093,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html#text"},{"title":"Model improvement through closed-loop experimentation.","offset":166958,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html#text"},{"title":"Synthesis and open challenges.","offset":168193,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-11.html#text"},{"title":"7.5 Embodied AI and Robotics","offset":169201,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"Evaluation substrate and feedback structure.","offset":169749,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"FM improvement through autonomous practice and data flywheels.","offset":170458,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"Scaffolding improvement through curricula, memory, and safety logic.","offset":171841,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"Synthesis and open challenges.","offset":172969,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"7.6 General Computer Control","offset":174165,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"Evaluation substrate.","offset":174801,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"Scaffolding improvement via experience and procedural reuse.","offset":175437,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"FM improvement through iterative training and verifier construction.","offset":176400,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"Synthesis and open challenges.","offset":177578,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"8 Evaluation","offset":178907,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"8.1 Measuring Improvement","offset":179803,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"Formalizing the Evaluation Objective.","offset":180430,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"8.1.1 Metric-Based Measurement","offset":182553,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-12.html#text"},{"title":"8.1.2 Judge-Based Measurement","offset":185997,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"Specify the judge and the judging budget.","offset":186925,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"Prevent over-optimization to the judge and report reliability signals.","offset":187778,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"8.2 Benchmarking Improvement","offset":188632,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"8.2.1 Mechanism Benchmarks","offset":189529,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"FM -Level Evaluation.","offset":190222,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"Scaffold -Level Evaluation.","offset":191066,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"Attribution Across Mechanisms.","offset":192411,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"8.2.2 Domain Benchmarks","offset":192883,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"Software Engineering.","offset":193648,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"Web Navigation and Automation.","offset":194679,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"Gaming and Strategic Reasoning.","offset":195385,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"Scientific Discovery.","offset":195970,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"Embodied AI.","offset":196686,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"General Computer Control.","offset":197230,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"9 Discussion","offset":197891,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"9.1 Implications for System Design","offset":198533,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"From Fast Exploration to Slow Consolidation.","offset":198573,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"The Critic as Governed Infrastructure.","offset":199972,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-13.html#text"},{"title":"Safety through Layered Gating.","offset":200946,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-14.html#text"},{"title":"9.2 Future Directions","offset":202263,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-14.html#text"},{"title":"10 Conclusion","offset":207478,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/part-14.html#text"}],"text_code_points":209000,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/source.md"}