Yong Li

Research profile

Research summary

Yong Li is a researcher at Tsinghua University whose profile is name-resolved from OpenAlex public bibliographic data; the name is common and no ORCID is attached, so identity certainty is limited. The snapshot shows a strongly current record: 10 works in the recent three-year window, rising from 4 works in 2025 to 6 in 2026, alongside 32 distinct coauthors and a collaboration score of 70. The Evolve catalog adds four 2026 arXiv preprints sketching a coherent machine-learning agenda. CAER proposes causal action-effect reweighting for world-model training (repro-readiness medium; Evolve finds the theory elegant but notes MSE-only comparisons and margins lacking variance reporting). CateKV targets long-context LLM inference acceleration through sequential consistency. A third paper diagnoses human–LLM divergence in single-turn persuasion judgments, and Ideation Arena evaluates LLM-generated research ideas via battle-style human expert assessment. Yong Li is last author on three of the four catalog papers and middle author on one, with no corresponding-author record, a pattern consistent with a supervisory role in a fast-moving group. The OpenAlex topic profile — topic modeling, complex network analysis, recommender systems, information retrieval, and gene regulatory networks — leans toward different areas than the arXiv catalog, possibly reflecting genuine breadth or name conflation. Citation indicators are absent: h-index and i10-index are recorded as 0, so output and impact remain unscored.

Research keywords

Highlights

• CAER: causal action-effect reweighting for world-model training

  This 2026 arXiv preprint proposes online causal reweighting of action effects when training world models. Evolve's verdict: claims of beating uniform MSE on four world-model tasks, with elegant theory, but comparisons are MSE-only and margins lack variance reporting. Repro-readiness is rated medium.

• CateKV: sequential consistency for long-context LLM inference

  A 2026 arXiv preprint on accelerating long-context LLM inference, framed around sequential consistency, with Yong Li in a middle author position. It is in the Evolve catalog but not yet reviewed, so no verdict, keywords, or repro-readiness rating is available.

• Do LLMs change their minds like humans?

  A 2026 arXiv preprint diagnosing human–LLM divergence in single-turn persuasion judgments, with Yong Li in the last-author position. Not yet reviewed by Evolve; no verdict or repro-readiness rating is on file.

• Ideation Arena: battle-style expert assessment of LLM research ideas

  A 2026 arXiv preprint proposing evaluation of LLM-generated research ideas through battle-style human expert assessment, with Yong Li in the last-author position. Awaiting Evolve review — no verdict or repro-readiness rating yet.

Timeline

1. 2025OpenAlex records 4 works for 2025 in the snapshot; the titles themselves fall outside the provided data.
2. 2026CAER preprint posted on arXiv: causal action-effect reweighting for world-model training; Evolve rates reproducibility medium.
3. 2026CateKV preprint appears on arXiv, addressing sequential consistency for long-context LLM inference acceleration; middle-author position.
4. 2026Preprint diagnosing human–LLM divergence in single-turn persuasion judgments, with last-author position.
5. 2026Ideation Arena preprint proposes battle-style human expert assessment of LLM-generated research ideas; last-author position.
6. 2026Snapshot totals: 6 works recorded for 2026 and 10 works over the recent three-year window.

Papers in the Evolve catalog

• 2026-08-31

  CAER: Causal Action Effect Reweighting for World Model Training

Frequent co-authors