Reproduce “Adaptive Information Control for Search-Augmented LLM 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

Siheng Xiong, Oguzhan Gungordu, James C. Kerce, Faramarz Fekri

DeepControl trains search-augmented reasoning agents to decide when to continue retrieval and how much evidence to expose. It estimates evidence utility using novelty and support for the gold answer, guiding search continuation and selective expansion during training. Across seven text-QA benchmarks, Qwen2.5-7B-Instruct scores 0.479 average Exact Match, versus 0.385 for Search-R1-Instruct. Evaluation covers text QA with Qwen models.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.