Microsoft Research (UK) researcher in LLM red teaming and scalable multi-agent systems; two 2026 arXiv papers reviewed.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-02) 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 output data is available in this snapshot: publication counts, venue tiers, and yearly paper volume are missing, so this axis is unscored. It would measure total and recent research productivity across venues.
Impact—
Impact indicators — citation counts, h-index, and downstream adoption of the reviewed methods — are absent from the data, so no score is given. This axis would measure how widely and durably the work influences other researchers.
Focus0/100
Focus scores 0 on the radar, grounded in the raw field distribution: the two cataloged papers split exactly evenly, 0.5 to artificial intelligence (AI safety / LLM red teaming) and 0.5 to LLM multi-agent systems (agentic AI). Both are 2026 releases, so the current snapshot spans two parallel directions rather than one concentrated line; the score reflects breadth of this small snapshot, not research quality.
Collaboration59/100
Collaboration scores 59, computed from 18 distinct coauthors in the snapshot. Twelve of them recur frequently — Shaohan Huang, Li Dong, Dongdong Zhang, Tengchao Lv, Yilin Jia, Yuzhong Zhao, Yupan Huang, Wenshan Wu, Xiangyang Zhou, Nan Yang, Lei Cui, and Zhihao Zhan — pointing to a stable core group. Both 2026 papers list Furu Wei last without corresponding-author flag, consistent with senior oversight of team projects.
Activity—
Activity data is not present: publication frequency per year and recent-year output cannot be computed from the snapshot. This axis would measure whether publishing cadence is steady, accelerating, or slowing over time.
Reproducibility55/100
Reproducibility scores 55, the mean readiness across the 2 papers assessed, each rated medium. For CART, Evolve judges the LLM-judged gains over static replay as real-looking but short of statistical rigor; for Agensh, the +49% relative gain at 128 agents on ProgramBench comes without variance, cost, or orchestrator baselines. Both setups are promising, but independent replication would need stronger evidence.
Research summary
Furu Wei is a researcher at Microsoft Research (United Kingdom) whose cataloged work concentrates on two strands of large language model research: automated safety evaluation and multi-agent systems. The Evolve catalog contains two 2026 arXiv papers, both with Wei in the last author position. CART builds a closed-loop adaptive red teaming framework for LLMs, combining Thompson sampling with provenance-tracked probing; Evolve's review finds the LLM-judged gains over static replay plausible but flags the absence of statistical rigor. Agensh reports an orchestrator-free harness scaling organizational intelligence to 1,024 LLM agents, with a +49% relative gain at 128 agents on ProgramBench — promising, though lacking variance, cost, and orchestrator-baseline comparisons. The bibliometric radar scores collaboration at 59 from 18 distinct coauthors, twelve of whom recur as a stable core group including Shaohan Huang, Li Dong, Nan Yang, Lei Cui, and Dongdong Zhang. Reproducibility averages 55 across the two assessed papers, both rated medium readiness. Focus registers 0, driven by an even 0.5/0.5 field split between LLM red teaming and agentic AI. Output, impact, and activity axes lack underlying data and are left unscored, so citation counts and publication volumes are unknown here. The profile is name-resolved from public bibliographic data.
Collaboration pattern
Collaboration centers on a recurring core: 12 of the 18 distinct coauthors appear frequently, including Shaohan Huang, Li Dong, Dongdong Zhang, Tengchao Lv, Yupan Huang, Wenshan Wu, Nan Yang, and Lei Cui. Both 2026 papers place Furu Wei in the last author position without corresponding-author flagging, a pattern consistent with senior supervision of group projects rather than hands-on first-author work. The radar collaboration score is 59.
Research keywords
Highlights
Closed-loop adaptive red teaming for LLMs
Scaling multi-agent organization to 1,024 agents
Timeline
- 2026CART appears on arXiv, introducing closed-loop adaptive red teaming for large language models.
- 2026Agensh appears on arXiv, scaling orchestrator-free multi-agent organization to 1,024 LLM agents.
- 2026Both catalog papers are 2026 arXiv releases with Furu Wei in the last author position, spanning LLM red teaming and multi-agent systems.
Papers in the Evolve catalog
Frequent co-authors
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.
- Radar scores and coauthor statistics come from the underlying bibliometric snapshot; citation counts, h-index, and publication volumes are not available in this profile.
- The profile rests on 2 Evolve-reviewed papers, both 2026 arXiv preprints; earlier or peer-reviewed work is outside the current catalog scope.
- Reproducibility is averaged over only 2 assessed papers, both rated medium, so the score of 55 should be read as provisional.
- Identity and affiliation are name-resolved from public bibliographic data as provided (Microsoft Research, United Kingdom); no ORCID was supplied.