Reproduce “O1 Replication Journey: A Strategic Progress Report -- Part 1” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yiwei Qin, Xuefeng Li, Haoyang Zou, Yixiu Liu, Shijie Xia, Zhen Huang, Yixin Ye, Weizhe Yuan and 3 more
Researchers from Shanghai Jiao Tong University, NYU, MBZUAI, and GAIR replicated OpenAI's O1 model, introducing a "journey learning" paradigm that improved performance on the MATH dataset by over 8% by focusing on the entire reasoning process. This effort also serves as a model for open science in AI by transparently documenting the full research journey.
Key insights
- Journey learning enhances generalization, adaptability, and self-correction by emphasizing the entire problem-solving process rather than just final answers.
- The quality of training data, particularly detailed reasoning processes, can lead to substantial performance improvements even with limited sample sizes.
- Transparently documenting the full research journey, including failures and re-directions, promotes open science and accelerates collective progress in AI development.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.