Shanghai AI Laboratory researcher working on LLM self-distillation training and LLM-based scientific agents
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-09) 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—
Not yet scored: this profile rests on only two cataloged 2026 arXiv preprints, with no publication counts or venue record available. The output axis would measure total publications, their venues, and year-over-year volume.
Impact—
Not yet scored: both cataloged works are 2026 preprints without citation data, so influence cannot be estimated. The impact axis would measure citations, downstream adoption, and reference-based influence once citation records accrue.
Focus0/100
Focus scores 0/100, matching the raw field distribution: 'NLP / AI — LLM-based scientific agents' and 'machine learning (LLM training / knowledge distillation)' each account for exactly 50%. The two cataloged 2026 papers sit one per field, so the low score reflects an evenly two-sided, early-stage agenda — distillation training methods on one side, research-asset infrastructure for scientific agents on the other.
Collaboration68/100
Collaboration scores 68/100, drawn from 29 distinct coauthors across the cataloged papers. The frequent-collaborator list spans the ScholarSeed AI Team plus eleven named individuals — including Caoqinwei Gong, Xue Jiang, Wei Luo, Zheng Yu, and four researchers surnamed Zhang — a pattern of large, lab-internal author teams typical of industrial LLM research groups, with every cataloged paper a multi-author effort.
Activity—
Not yet scored: only two 2026 preprints are in evidence, without a multi-year publication record, so research cadence cannot be measured. The activity axis would track output frequency, recency, and consistency of publishing across recent years.
Reproducibility55/100
Reproducibility scores 55/100, the mean readiness rating across the two reviewed papers (reproPapers = 2), both rated 'medium'. The self-distillation study reports hierarchy-unstable, seed-sensitive results that Evolve recommends replicating across seeds; ScholarStack's gains across ten agent tasks rest on LLM-judge scoring without significance tests. Both state testable claims, but neither yet has independent confirmation.
Research summary
Yi Wang is a researcher at the Shanghai Artificial Intelligence Laboratory whose cataloged work, profiled here from public bibliographic data, sits at the intersection of two threads: LLM training methods and LLM-based scientific agents. The Evolve catalog contains two 2026 arXiv preprints. In 'What Should a Self-Teacher See?', Wang appears as the last author on a study of privileged context design for on-policy self-distillation (OPSD), claiming that a deliberately reduced teacher context outperforms full-solution teachers on OPSD mathematics tasks at 4B/8B peaks — results the review flags as hierarchy-unstable and in need of seed replication. In 'ScholarStack', Wang is a middle author on a system that orchestrates layered, provenance-tracked research assets for scientific agents, claiming quality gains and reduced query tokens across ten agent tasks, with the caveat that evaluation is LLM-judged without significance tests. Bibliometric signals show 29 distinct coauthors and an even 50/50 split between machine-learning distillation work and agent infrastructure, yielding a low focus score but broad collaborative reach; output, impact, and activity metrics are not yet scored.
Collaboration pattern
The cataloged papers carry large author teams: 29 distinct coauthors appear across Wang's records, with the ScholarSeed AI Team and eleven recurring individuals — Caoqinwei Gong, Xue Jiang, Wei Luo, Xiaoyu Qiu, Jiayi Sheng, Zheng Yu, four Zhangs, and Yuan Cao — at the core. Both 2026 papers are multi-author efforts (Wang last author on one, middle on the other, neither corresponding), consistent with a collaborative contributor inside a big industrial LLM lab rather than an independent small-group lead.
Research keywords
Highlights
Minimal privileged teacher context for on-policy self-distillation
ScholarStack: layered research assets for scientific agents
A two-thread 2026 agenda spanning training and agent infrastructure
Both cataloged papers appeared in 2026 and split evenly across Wang's two fields — LLM training/distillation and LLM-based scientific agents — with 29 distinct coauthors, indicating an early-stage profile built inside large collaborative teams rather than a solo line of work.
Timeline
- 2026Appeared on arXiv as last author of 'What Should a Self-Teacher See? Privileged Context Design for On-Policy Self-Distillation'.
- 2026Appeared on arXiv as a middle author of 'ScholarStack: Layered Research Asset Orchestration and Cross-Task Reuse for Scientific Agents'.
- 2026Both cataloged works landed in a single year, marking a concentrated early burst spanning LLM distillation training and scientific-agent infrastructure.
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.
- Bibliometric scores derive from OpenAlex-based data; the output, impact, and activity axes lack sufficient data and are unscored.
- Coverage is limited to two 2026 arXiv preprints reviewed by Evolve; peer-reviewed venue status and non-cataloged work are not reflected.
- 'Yi Wang' is a common name; this profile is name-resolved from public bibliographic data and may conflate homonymous researchers.
- No ORCID or country is available in the source record; the affiliation is as listed in the bibliographic data.