Bo Li

—

Unclaimed · AI-structured profileOpenAlex

Bo Li — agentic-AI evaluation and human-AI collaboration researcher; last author of the ReFigBench benchmark

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-20) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.

Research profile

Output—

No publication-volume metrics are available in the current snapshot, so this axis is unscored. It would measure total output — paper counts, author positions, and venues — across the researcher's full record rather than the two cataloged 2026 preprints.

Impact—

Impact cannot be assessed yet: citation counts and downstream adoption signals are absent from the input data, so the axis is unscored. It would measure how widely the two cataloged works are cited and built upon once such metrics become available.

Focus0/100

The focus score is 0, and the raw field distribution explains why: activity splits evenly — 0.5 to AI agents and human-AI collaboration and 0.5 to machine learning (multimodal agents and evaluation). Both cataloged 2026 papers are agent-adjacent, but ReFigBench targets evaluation methodology while Atria Dawn is a model release report, so the profile reads as deliberately broad rather than narrowly specialized.

Collaboration86/100

Collaboration scores 86 on the 0-100 scale, backed by 74 distinct coauthors in the raw data — a large-network pattern. Twelve recurring coauthors include Honglin Guo, Tao Gui, Yicheng Chen, and Guanting Dong. The last-author position on ReFigBench and the middle-author slot on the Atria Dawn release report both fit a senior contributor working within sizeable team-based projects.

Activity—

Recent-activity data is not available: publication cadence and preprint frequency over time cannot be computed, and the snapshot's only two papers both date to 2026. This axis would measure ongoing research pace across multiple years.

Reproducibility38/100

Reproducibility averages 38 across the 2 papers with readiness assessments. ReFigBench rates medium — its 1,000-figure benchmark and paired harness-versus-workflow study are checkable, though LLM-as-a-judge scoring introduces judge-family caveats — while Atria Dawn rates low, with cherry-picking risk flagged in its benchmark table and self-reported attribution numbers.

Research summary

Bo Li is an early-stage Evolve profile whose record is name-resolved from public bibliographic data; no institution, country, or ORCID is attached. The cataloged work sits entirely in 2026 and splits evenly across two adjacent areas in the field distribution: AI agents and human-AI collaboration (0.5) and machine learning with multimodal agents and evaluation (0.5). Bo Li is the last author of ReFigBench, a 1,000-figure benchmark for reconstructing scientific figures as editable PowerPoint artifacts, where a paired study shows that agent harness and workflow jointly determine agent quality; Evolve finds it broadly convincing, with LLM-as-a-judge caveats. The second entry, Atria Dawn, is a middle-author contribution to an agentic language model release report whose human-AI collaboration study is its strongest element, though its benchmark table and self-reported attribution numbers carry cherry-picking risk. The collaboration axis scores 86, with 74 distinct coauthors and 12 recurring collaborators, fitting large team-based projects. Reproducibility averages 38 across the two rated papers (one medium, one low). Output, impact, and activity metrics are not yet available and are left unscored; conclusions here should be read as provisional until a fuller record is linked.

Collaboration pattern

The collaboration signature is large-team: 74 distinct coauthors, a collaboration score of 86, and 12 recurring collaborators including Honglin Guo, Tao Gui, Yicheng Chen, Guanting Dong, Qiming Ge, and Yuyang Hu. Both cataloged papers are multi-author arXiv releases — a last-author (senior) placement on ReFigBench and a middle slot on Atria Dawn — consistent with someone who anchors team projects rather than working solo.

Research keywords

benchmarkmultimodal coding agentsscientific figure reconstructionPowerPoint generationagent harness evaluationLLM-as-a-judgeagentic language modelverifiable experience pipelinerecursive self-improvementhuman-AI collaborationtool-using agentsbenchmark evaluation

Highlights

ReFigBench: a 1,000-figure benchmark for editable PowerPoint reconstruction
As last author, Bo Li contributed to a benchmark of 1,000 scientific figures reconstructed as editable PPTX artifacts. A careful paired study shows agent harness and workflow jointly decide agent quality; Evolve finds the work broadly convincing, while noting LLM-as-a-judge caveats. Reproducibility readiness is rated medium.
Atria Dawn: the human-AI collaboration study is the standout
On this agentic-superintelligence model release report, Bo Li is a middle author. Evolve's review finds the human-AI collaboration study its most interesting part, while flagging cherry-picking risk in the benchmark table and self-reported attribution numbers; reproducibility readiness is rated low.
A consistent emphasis on evaluation methodology
Across both 2026 catalog entries, the thread is how agent quality is measured: a paired harness-versus-workflow design in ReFigBench and a human-AI collaboration study in Atria Dawn. Both works also carry flagged evaluation caveats — judge-family limits and cherry-picking risk — consistent with the modest reproducibility mean of 38.

Timeline

2026Last author on ReFigBench, a 1,000-figure benchmark for reconstructing scientific figures as editable PowerPoint artifacts.
2026Middle author on Atria Dawn, an agentic language model release report whose human-AI collaboration study drew the most attention.
2026Both cataloged outputs appear in the same year, splitting evenly between multimodal agent evaluation and AI agents and human-AI collaboration.

Papers in the Evolve catalog

Frequent co-authors

Honglin Guo×1 Tao Gui×1 Yicheng Chen×1 Guanting Dong×1 Qiming Ge×1 Yuyang Hu×1 Zixian Huang×1 Jiajie Jin×1 Alexander Lam×1 Yining Li×1 Jiahang Lin×1 Yanjiang Liu×1

About this profile

This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.

Name-resolved profile: "Bo Li" is a common name, and identity is inferred from public bibliographic data only — no ORCID, institution, or country is available in the source.
Radar scores (collaboration, focus, reproducibility) come from bibliometric data; output, impact, and activity dimensions lack underlying data and are intentionally unscored.
Qualitative assessments rest on only two Evolve-reviewed papers, both 2026 arXiv preprints, which may not represent the full publication record.
The reproducibility mean of 38 is computed over just 2 papers, so the estimate is fragile.