UC Berkeley researcher in LLM-based human simulation, neural audio codec analysis, and large-scale multimodal datasets
3
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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 axis would aggregate publication volume and productivity signals — paper counts, venue throughput, authorship roles — over time. The snapshot holds only three 2026 arXiv records, too few coverage-verified entries for a reliable output score.
Impact—
Not yet scored: impact would measure citations, derivative works, and uptake of this researcher's outputs by others. No citation counts, h-index, or downstream usage data were present in the input, so no estimate is offered here.
Focus100/100
Focus scores a perfect 100/100, computed from the field distribution in the bibliometric snapshot: all labeled catalog activity maps to a single field, natural language processing (LLM-based human simulation). The three reviewed papers do reach into speech processing and a vision-language dataset, but the labeled field profile remains highly concentrated.
Collaboration60/100
Collaboration scores 60/100, grounded in 19 distinct coauthors in the snapshot, 12 of whom recur frequently — including John Canny, Alane Suhr, Tsung-Han Wu, Anya Ji, and Suhong Moon. Across the three catalog papers, Chan sits in middle and last authorship positions, with no corresponding-author role recorded.
Activity—
Not yet scored: activity tracks recent publishing cadence and ongoing momentum, such as papers per year and output recency. Beyond three 2026 preprints, the snapshot lacks the multi-year data needed to establish any trend.
Reproducibility55/100
Reproducibility scores 55/100, derived from a mean of 55 over only one scored paper (reproPapers=1), so coverage is thin. Of the reviewed entries with a readiness rating, both the persona mixture and audio codec papers are graded medium, with Evolve flagging unexamined data contamination and small-sample correlations respectively.
Research summary
David M. Chan is a researcher at the University of California, Berkeley whose recent preprint record spans three adjacent areas of machine learning. In natural language processing, he works on LLM-based human simulation: the 2026 paper 'Pretrained Persona Mixture Models and Tandem Models for Human Simulation' (middle authorship) proposes pretrained PMM base models and a tandem selection scheme that, per Evolve's review, outperform instruction-tuned LLMs at simulating human respondents, though training-data contamination remains unexamined. In speech processing, he contributed to a language-statistical analysis of neural audio codec tokens across 13 codecs, corpora, and noise conditions, applying Zipf's law, Heaps' law, and Jensen-Shannon divergence; Evolve rates the descriptive evidence solid while flagging that quality links rest on n=12 correlations. A third 2026 preprint, CoVA-SFT, lists him as last author and releases a large-scale dataset for chain-of-visual-abstractions fine-tuning; it is not yet reviewed. The profile radar shows perfect focus (100/100) in the labeled field distribution, a moderate collaboration score (60/100) built on 19 distinct coauthors, and a reproducibility score of 55/100 averaged over a single scored paper; output, impact, and activity axes lack sufficient data. All catalog entries are arXiv preprints dated 2026, so this is an early-stage, preprint-centric profile.
Collaboration pattern
The snapshot records 19 distinct coauthors, with 12 recurring collaborators, notably Berkeley colleagues John Canny, Alane Suhr, Tsung-Han Wu, Anya Ji, and Suhong Moon, plus Minwoo Kang, Téa Wright, Seun Eisape, Ayush Raj, Joseph Suh, Heekyung Lee, and Haoming Chen. The pattern is small-team arXiv preprints: Chan sits in middle or last authorship positions, none as corresponding author, consistent with contributing specialist components to group projects rather than leading solo efforts.
Research keywords
Highlights
Pretrained persona mixture models beat instruction-tuned LLMs at human simulation
Language-statistical audit of 13 neural audio codecs under noise
CoVA-SFT: large-scale dataset for chain-of-visual-abstractions fine-tuning
A 2026 arXiv dataset release with Chan in the last author position. It has not yet been reviewed by Evolve — no verdict, keywords, or reproducibility rating — and carries no catalog slug, so treat it as an early data contribution pending evaluation.
Timeline
- 2026Published 'Pretrained Persona Mixture Models and Tandem Models for Human Simulation' on arXiv (middle authorship)
- 2026Released a language-statistical analysis of neural audio codec tokens across architectures, corpora, and noise conditions (middle authorship)
- 2026Co-released CoVA-SFT, a large-scale dataset for chain of visual abstractions (last authorship)
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 name 'David M. Chan' is not rare; this profile is name-resolved from public bibliographic data and may merge or omit records.
- All catalog entries are 2026 arXiv preprints; no peer-reviewed venue, citation, or h-index data was available.
- Radar output, impact, and activity axes are unscored because the underlying publication-volume and citation data are missing.
- CoVA-SFT is unreviewed in Evolve, and the reproducibility mean covers only one scored paper, so both readings carry wide uncertainty.