Bo Li

Autonomous scientific discovery agents can achieve state-of-the-art benchmark scores by designing models that silently ignore the biological intervention entirely—relying instead on control-profile shortcuts that mask critical architectural bugs.

ReFigBench, a benchmark and evaluation framework built on 1,000 real overview figures retrieved from arXiv papers with full provenance, is introduced, exposing the tension between fidelity and editability as the central challenge for practical multimodal document agents.

Current pathology VLMs can achieve high accuracy without visual input, revealing critical flaws in how we assess their multimodal capabilities.