Researcher on LLM research agents and pre-training data curation — recursive self-improvement, agentic data selection.
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—
Publication-output metrics are not yet available for this profile. This axis would measure total papers, venues, authorship position distribution, and annual output volume; none of these can be computed from the current data.
Impact—
No citation or impact metrics are available yet. This axis would measure citations, h-index-type indicators, and uptake of the two cataloged 2026 papers; both are recent arXiv preprints, so influence cannot be scored.
Focus0/100
The focus score is 0, and the underlying field distribution explains why: the cataloged work splits exactly 0.5/0.5 between "artificial intelligence — autonomous research agents / LLM agent systems" and "machine learning / NLP (LLM pre-training data curation)". Rather than concentrating on a single track, the two 2026 papers occupy two distinct subfields — one on self-improving research agents, one on pre-training data selection — linked by shared LLM-agent methodology.
Collaboration41/100
Collaboration scores 41 on the 0-100 scale, based on 7 distinct coauthors across the cataloged work. The frequent-coauthor list includes Dhruv Srikanth, Bingchen Zhao, Zhengyao Jiang, Dixing Xu, and Yan Meng. Authorship positions vary — middle author on the recursive self-improvement paper and last author on AutoData, neither marked corresponding — consistent with a mid-sized, team-based collaboration pattern.
Activity—
Activity metrics are not yet available. This axis would track publication cadence over time — papers per year, recency, and sustained output — but the current snapshot contains only two 2026 arXiv entries, too little to characterize an activity pattern.
Reproducibility38/100
Reproducibility averages 38 across 2 assessed papers. Per-paper readiness is rated low for "Recursive self-improvement of AI research agents", whose results rest on few runs with noisy grade comparisons, and medium for "AutoData", whose val-bpb comparisons hold across scales while CORE gains lack statistical support. The pattern suggests artifacts and recipes exist, but statistical robustness remains the weak point.
Research summary
Yuxiang Wu (ORCID 0000-0002-1156-6718; institution not listed in the source data) is a researcher whose cataloged work splits evenly — 0.5 and 0.5 in the field distribution — between two connected areas of modern machine learning: autonomous research agents built on large language models, and pre-training data curation for LLMs. The Evolve catalog holds two 2026 arXiv papers. The first, "Recursive self-improvement of AI research agents", reports the AIDE² system, whose 8-day self-improvement loop lifted an agent grade from 0.703 to 0.778 with held-out transfer, though on few runs with noisy grade comparisons; Wu is a middle author. The second, "AutoData: Agentic Search for Pre-training Data Selection", applies LLM-agent search over executable data-selection recipes, beating curated-pipeline baselines on validation bits-per-byte across scales; Wu is the last author, though not marked corresponding. Collaboration metrics list 7 distinct coauthors, including Dhruv Srikanth, Bingchen Zhao, Zhengyao Jiang, Dixing Xu, and Yan Meng. The reproducibility axis averages 38 across the 2 assessed papers, with readiness rated low for the first and medium for the second. Output, impact, and activity metrics are not yet available, so this should be read as an early-stage, name-resolved profile.
Collaboration pattern
The profile lists 7 distinct coauthors, with Dhruv Srikanth, Bingchen Zhao, Zhengyao Jiang, Dixing Xu, and Yan Meng among the frequent collaborators. Wu's roles differ across the two cataloged papers — middle author on the research-agents paper, last author on AutoData — and neither lists Wu as corresponding author, suggesting contributions within larger team efforts rather than solo-led projects.
Research keywords
Highlights
AIDE²: an 8-day recursive self-improvement loop lifts agent grade 0.703→0.778
AutoData: agentic search beats curated pipelines on pre-training data selection
Timeline
- 2026Enters the Evolve catalog with "Recursive self-improvement of AI research agents" (arXiv), as a middle author in the autonomous research agents / LLM agent systems field.
- 2026Publishes "AutoData: Agentic Search for Pre-training Data Selection" (arXiv) as last author, in the LLM pre-training data curation field.
- 2026The two cataloged papers split the year's output evenly across two fields (0.5/0.5 in the field distribution), from self-improving research agents to pre-training data selection.
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 are computed from OpenAlex-based public bibliometric data only; no institution or country is on record.
- Only two 2026 arXiv preprints are in the Evolve catalog; output, impact, and activity dimensions are missing and deliberately not scored.
- "Yuxiang Wu" is a common name; this profile is name-resolved from public bibliographic data via ORCID 0000-0002-1156-6718, and identity should be verified before citing.
- Reproducibility rests on only 2 assessed papers (mean 38, one rated low and one medium); the frequent-coauthor list contains duplicates and the researcher's own name, so per-person collaboration counts are not inferable.