{"id":"arxiv:2503.00735","title":"LADDER: Self-Improving LLMs Through Recursive Problem Decomposition","date":"2025-03-02","section":"Post-training","quality":-0.008441,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/","source_url":"https://arxiv.org/abs/2503.00735","text_available":true,"authors":["Toby Simonds","Akira Yoshiyama"],"outgoing":[{"source":"arxiv:2503.00735","target":"arxiv:2203.11171","evidence":"title","id":"arxiv:2203.11171","title":"Self-Consistency Improves Chain of Thought Reasoning in Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/02a15b098d96a94d463bb470f59a34723abc34e59196f35c539f9796d8f3c3df/"},{"source":"arxiv:2503.00735","target":"arxiv:2402.03300","evidence":"title","id":"arxiv:2402.03300","title":"DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/636fb871904bc06f742cfdea069796f9e9665f95bd58dc63260eaff5154ef985/"},{"source":"arxiv:2503.00735","target":"arxiv:2501.19393","evidence":"title","id":"arxiv:2501.19393","title":"s1: Simple test-time scaling","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/2a88818f7605b71cd0160831faebd51b3274c24ffd29626b4d519390b4bffffe/"}],"incoming":[{"source":"arxiv:2607.13104","target":"arxiv:2503.00735","evidence":"title","id":"arxiv:2607.13104","title":"Self-Improvements in Modern Agentic Systems: A Survey","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1452caf7b16ffdb8c2efd5093f46dc4df0f89c3d45a5566cd5fdafd69385b2a7/"}],"text_parts":[{"start":0,"end":15681,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html"},{"start":15681,"end":31670,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-2.html"},{"start":31670,"end":45395,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-3.html"}],"contents":[{"title":"LADDER: Self-Improving LLMs Through Recursive Problem Decomposition","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html#text"},{"title":"Abstract","offset":71,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html#text"},{"title":"1 Introduction","offset":1339,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html#text"},{"title":"2 Related Work","offset":6011,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html#text"},{"title":"Self-Improvement and Automated Curriculum Generation in LLMs.","offset":6030,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html#text"},{"title":"Test-Time Compute Scaling and Reasoning Improvements.","offset":7188,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html#text"},{"title":"Reinforcement Learning Approaches for LLM Self-Improvement.","offset":8348,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html#text"},{"title":"3 Methodology","offset":9518,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html#text"},{"title":"3.1 Algorithm Design (LADDER & TTRL)","offset":9536,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html#text"},{"title":"3.1.1 LADDER","offset":9578,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html#text"},{"title":"3.1.2 Variant Generation","offset":11092,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html#text"},{"title":"3.1.3 Solution Verification","offset":15647,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-1.html#text"},{"title":"3.1.4 Reinforcement Learning Protocol","offset":18650,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-2.html#text"},{"title":"3.1.5 Test-Time Reinforcement Learning","offset":21472,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-2.html#text"},{"title":"3.2 Experiments","offset":23391,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-2.html#text"},{"title":"3.2.1 Llama 3B Experiments","offset":23412,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-2.html#text"},{"title":"3.2.2 MIT Integration Bee (LADDER)","offset":25535,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-2.html#text"},{"title":"3.2.3 MIT Integration Bee (LADDER + TTRL)","offset":27569,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-2.html#text"},{"title":"4 Results","offset":28901,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-2.html#text"},{"title":"4.1 Llama 3B Experiments","offset":28915,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-2.html#text"},{"title":"4.2 MIT integration Bee (LADDER)","offset":31670,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-3.html#text"},{"title":"4.3 MIT Integration Bee (LADDER + TTRL)","offset":32897,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-3.html#text"},{"title":"5 Discussion","offset":35517,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-3.html#text"},{"title":"5.1 A Form of Test-Time Compute Scaling","offset":37221,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-3.html#text"},{"title":"5.2 Extension to Other Verifiable Domains","offset":39769,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-3.html#text"},{"title":"5.3 Future Work","offset":41292,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-3.html#text"},{"title":"6 Conclusion","offset":42143,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-3.html#text"},{"title":"Appendix: MIT Integration Bee Failed Questions","offset":43830,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-3.html#text"},{"title":"Instructions for reporting errors","offset":44431,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/part-3.html#text"}],"text_code_points":45395,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/65b264030233bf2bcd17284eb9a521e6e38d4ee83eb7fec6587e4c047199f227/source.md"}