AI systems researcher spanning LLM pretraining, RL post-training infrastructure, and human-AI collaboration frameworks
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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—
Publication-volume data for this axis is not yet available, so no output score can be reported. This axis would measure the number of papers per year, venue types, and authorship positions across the researcher's full publication record.
Impact—
Citation-based influence data is insufficient for this axis, so it remains unscored. It would measure citations accumulated by the researcher's works and their uptake by later studies; no influence estimate should be inferred from the three 2026 preprints reviewed here.
Focus0/100
The focus score is 0, the lowest on the radar, and the raw field distribution shows why: activity splits evenly at 0.33 each across machine learning systems for LLM training infrastructure, large language model research, and human-AI collaboration and the future of work. The record reads as a deliberate three-way portfolio rather than a single tightly specialized research line.
Collaboration83/100
Collaboration scores 83, backed by 63 distinct coauthors in the raw record and a twelve-person frequent-coauthor list including RadixArk, Tom Chen, Mao Cheng, Yusheng Su, Tao Lin, and Yueming Yuan. All three catalog papers list the researcher as a middle author without corresponding status, a pattern consistent with contributing inside larger engineering teams rather than leading a small group.
Activity—
Per-year publication cadence data is not available, so this axis cannot be scored. It would measure recent output rate and submission continuity beyond the three 2026 arXiv preprints currently captured in the Evolve catalog.
Reproducibility58/100
Reproducibility averages 58 across three papers, and the spread is informative: the NCP-ArchPreview latent-space pretraining report rates high, the Miles v0.1 RL post-training system report rates medium, and the workflow-level AI augmentation framework rates low as a conceptual, empirically untested whitepaper. Readiness tracks how fully training details and artifacts are disclosed.
Research summary
This profile describes an early-stage researcher whose Evolve record consists of three 2026 arXiv preprints spread evenly across three areas: machine learning systems for LLM training infrastructure, large language model research, and human-AI collaboration and the future of work. The work is team-based: 63 distinct coauthors appear in the record, twelve recur as frequent collaborators, and all three catalog papers list this researcher as a middle author without corresponding status. The technical core is LLM training. The NCP-ArchPreview report moves toward latent-space language models through next concept prediction, claiming 1.95× token-efficient pretraining via joint token and concept prediction at 8.9B parameters, with Stage-1 gains of +2.45 shrinking to +0.59 in Stage-2. Miles v0.1 documents a production-level reinforcement learning post-training stack built around mixture-of-experts models, SGLang, fully asynchronous RL, and weight synchronization. A third, more conceptual paper proposes six conditions for genuine AI augmentation through a workflow-level framework. Reproducibility averages 58 across the three papers, ranging from low (the empirically untested framework) through medium (Miles) to high (NCP-ArchPreview). Radar scores exist for collaboration (83), reproducibility (58), and focus (0, reflecting the even three-way field split), while output, impact, and activity lack sufficient data to score. The name field in the source record is unresolved, so identity could not be verified.
Collaboration pattern
The raw record lists 63 distinct coauthors and a twelve-person frequent-coauthor circle including RadixArk, Tom Chen, Mao Cheng, Kangrui Du, Yanbin Jiang, Jiajun Li, Yiming Li, Tao Lin, Yusheng Su, Andy Ye, and Yueming Yuan, yielding a collaboration score of 83. The pattern is large-team contribution: all three catalog papers position this researcher as a middle author without corresponding responsibility.
Research keywords
Highlights
Latent-space language models via next concept prediction
Production-level RL post-training system, self-reported
Six conditions for genuine AI augmentation at workflow level
Timeline
- 2026Publishes a workshop whitepaper proposing a workflow-level framework with six conditions for genuine AI augmentation in human-agent collaboration.
- 2026Co-authors the NCP-ArchPreview technical report on next concept prediction toward latent-space language models, claiming 1.95× token-efficient pretraining at 8.9B.
- 2026Contributes to Miles v0.1, a production-level RL post-training system report spanning mixture-of-experts training, SGLang, and fully asynchronous RL.
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.
- The source name field is unresolved (recorded as ':'); this profile is built purely from public bibliographic records and the researcher's identity could not be confirmed.
- Bibliometrics come from a public OpenAlex-style snapshot only; no citation counts, h-index, or venue acceptance data were available.
- The catalog covers three 2026 arXiv preprints with middle authorship; no first-author or corresponding-author work falls within this data scope.
- Output, impact, and activity axes are left unscored because their underlying data was missing at profile time; no estimates are inferred.