{"id":"arxiv:2407.00695","title":"Learning Formal Mathematics From Intrinsic Motivation","date":"2024-06-30","section":"Agents, open-endedness, AGI","quality":0.425409,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/","source_url":"https://arxiv.org/abs/2407.00695","text_available":true,"authors":["Gabriel Poesia","David Broman","Nick Haber","Noah D. Goodman"],"outgoing":[],"incoming":[{"source":"arxiv:2505.19590","target":"arxiv:2407.00695","evidence":"title","id":"arxiv:2505.19590","title":"Learning to Reason without External Rewards","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d8df33b706906a7abf4e660f54c75a81ffe2caaec8b0a10d1a6ad6ec63060832/"}],"text_parts":[{"start":0,"end":15778,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-1.html"},{"start":15778,"end":31597,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-2.html"},{"start":31597,"end":47517,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html"},{"start":47517,"end":63462,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-4.html"},{"start":63462,"end":75023,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-5.html"}],"contents":[{"title":"Learning Formal Mathematics From Intrinsic Motivation","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-1.html#text"},{"title":"Abstract","offset":57,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-1.html#text"},{"title":"1 Introduction","offset":1336,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-1.html#text"},{"title":"2 Related Work","offset":9200,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-1.html#text"},{"title":"Mathematical conjecturing.","offset":9516,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-1.html#text"},{"title":"Learning to prove theorems from human data.","offset":10316,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-1.html#text"},{"title":"Learning to prove theorems from synthetic data.","offset":11207,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-1.html#text"},{"title":"Intrinsic motivation","offset":12360,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-1.html#text"},{"title":"Self-improvement of language models","offset":13123,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-1.html#text"},{"title":"3 Minimo","offset":14141,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-1.html#text"},{"title":"3.1 Conjecturing","offset":15380,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-1.html#text"},{"title":"3.2 Proof Search","offset":22210,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-2.html#text"},{"title":"3.3 Hindsight Relabeling","offset":25124,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-2.html#text"},{"title":"3.4 Training loop","offset":27523,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-2.html#text"},{"title":"4 Experiments","offset":28448,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-2.html#text"},{"title":"4.1 Learning dynamics","offset":29882,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-2.html#text"},{"title":"Difficulty increases as training progresses (RQ1).","offset":30607,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-2.html#text"},{"title":"Proof policy improves as training progresses (RQ2).","offset":31538,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-2.html#text"},{"title":"Hindsight relabeling is necessary for joint self-improvement (RQ3).","offset":32486,"level":5,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"4.2 Proving human-written theorems (RQ4)","offset":34414,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"5 Limitations and Conclusion","offset":37157,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"Acknowledgments","offset":40311,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"Appendix A Training details","offset":41057,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"Appendix B Proof log-likelihood and MCTS Expansions","offset":42088,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"Appendix C Fraction of provable conjectures across iterations","offset":42509,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"Appendix D Axioms","offset":43779,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"D.1 Arithmetic","offset":43864,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"D.2 Propositional Logic","offset":44696,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"D.3 Groups","offset":46335,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"Appendix E Extrinsic evaluation problems","offset":46866,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"E.1 Arithmetic","offset":47010,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-3.html#text"},{"title":"E.2 Propositional Logic","offset":49636,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-4.html#text"},{"title":"Appendix F Discussion: extension to other proof assistants","offset":52294,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-4.html#text"},{"title":"NeurIPS Paper Checklist","offset":53718,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-4.html#text"},{"title":"Instructions for reporting errors","offset":74059,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/part-5.html#text"}],"text_code_points":75023,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68b19af6f86f72a1a7cedd8c0e476588d44d21a891d79fc640768859eac9a909/source.md"}