Reproduce “Facilitating Long Context Understanding via Supervised Chain-of-Thought Reasoning” 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
Jingyang Lin, Andy Wong, Tian Xia, Shenghua He, Hui Wei, Mei Han, Jiebo Luo
The authors develop PAI, which breaks questions into measurable facts, retrieves evidence for each, and summarizes the findings. They use its intermediate reasoning traces to generate answers and supervise a fine-tuned model. On Loong’s overall Average Score, GPT-4o-mini with PAI scores 69.58 versus 49.25 without it, at 128K context; this benchmark result does not establish generality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.