Research · 研究

Pushing the frontier of reasoning, code, and Chinese-language AI.

We work on agentic coding, multimodal evaluation, reasoning training, low-resource NLP, and model auditing, writing each result up as a bilingual explainer for researchers and general readers alike.


Reasoning · Code · Chinese-language AIUpdated June 2026
Featured work · AI safety

SafeGEO measures how recommendation agents can be steered — and defended.

A controlled benchmark of search-optimization attacks and mitigations across real recommendation settings.

Read the research
600Recommendation cases 22Attack variants 2026Research release
Research trajectory · 研究轨迹

Milestones

From model interpretation and cross-modal evaluation to reasoning, coding agents, and AI safety, each step extends a verifiable and reproducible research program.

  1. 01Interpretability & evaluation

    Report Cards earns a NeurIPS SoLaR Spotlight

    Readable, comparable model capability reports establish the foundation for the lab’s evaluation work.

    View research
  2. 02Cross-modal consistency

    SEAM is accepted at COLM 2025

    9,600 evaluations quantify whether models stay consistent across text and image inputs.

    View research
  3. 03Program expansion · Active

    Five research lines form a broader program

    Coding agents, RLVR, multilingual NLP, recommendation-agent safety, and grounded reasoning.

    View 2026 research