Long-record HKUST researcher now bridging legal AI evaluation and the theory of LLM-agent self-improvement
4
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
65
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—
No aggregate output score is available yet. This axis would measure publication volume and productivity trends across the career, but the computed metric is absent from the current data, so it stays unscored.
Impact—
No citation-impact score is available yet. This axis would measure the influence of the work — citations, citation peaks and per-paper reach — but that computed metric is missing here, so no estimate is offered.
Focus0/100
Scores 0, the widest dispersion. Current field attribution splits exactly evenly — 0.5 natural language processing (legal AI evaluation) and 0.5 machine-learning theory of LLM agents and self-improvement — while indexed OpenAlex topics add asthma and respiratory diseases (49), distributed and parallel computing systems (45), topic modeling (37), cloud computing (33) and GANs (33). The 2026 preprints themselves span legal benchmarks, agent-evolution theory, reasoning-model surveys and life-science modelling.
Collaboration70/100
Scores 70, reflecting a strongly collaborative but distributed pattern: 32 distinct coauthors in the most recent three years and a recurring inner group of 12 named collaborators (e.g., Wei Xue, Zhiqin Yang, Jingwen Fu). Across the four cataloged 2026 preprints Guo holds middle or last authorship without corresponding status, consistent with a senior contributor who seeds and steers projects while day-to-day leadership sits with coauthors.
Activity100/100
Scores 100, the maximum. The OpenAlex counts-by-year curve shows sustained acceleration: 34 works in 2017, 51 in 2019 and a peak of 62 in 2022, remaining high at 36 (2024) and 34 (2025), with 9 already recorded in 2026 and 79 works over the most recent three years. This cadence has persisted for more than three decades of indexed output since the first 1989 entries.
Reproducibility55/100
Scores 55, computed from Evolve's review of 2 papers (reproMean 55). Among cataloged preprints, LexAgentHallu and the safe harness self-evolution analysis are rated medium reproducibility readiness, while the large reasoning model survey is rated low — consistent with Evolve's verdict that its theory is rigorous but its empirics minimal and that the survey contains no original experiments.
Research summary
Yike Guo is a researcher at Hong Kong University of Science and Technology whose recent work sits at the intersection of legal AI evaluation and the theory of LLM-agent self-improvement. The OpenAlex record behind this profile is long and broad: an h-index of 65, an i10-index of 256, and annual output climbing from single digits in the 1990s to 51 works in 2019 and 62 in 2022, with 79 works across the most recent three years and 32 distinct coauthors. Current field attribution splits evenly — 0.5 natural language processing (legal AI evaluation) and 0.5 machine-learning theory of LLM agents and self-improvement — while indexed topics also span distributed systems, cloud computing, topic modeling and generative image synthesis, a spread wide enough that the focus dimension scores 0. Four 2026 arXiv preprints anchor the present agenda: LexAgentHallu, a 3,414-case benchmark for profiling legal-agent hallucinations; a theoretical analysis proving the limits of safe harness self-evolution; a survey of scaling large reasoning models beyond human supervision; and Life Operators, a self-evolving framework for multiscale life modelling. Evolve's reviews credit the taxonomic ambition and rigor while flagging LLM-judge inflation risk and thin empirical support.
Collaboration pattern
The recent record shows a hub-and-spoke pattern: 32 distinct coauthors over the most recent three years, with a stable inner group of 12 recurring collaborators — Wei Xue, Zhiqin Yang, Jingwen Fu, Yuhan Liu, Hengyu Liu, Yonggang Zhang, Kainan Cao, Zizhuo Zhang, Chenxin Li, Ruibin Yuan, Jiahao Pan and Jiankai Sun. On the cataloged 2026 preprints Guo appears as a middle or last author but never as corresponding author, suggesting project-level steering rather than hands-on paper management.
Research keywords
Highlights
First benchmark for legal-agent hallucinations
Proving the limits of safe agent self-evolution
Mapping reasoning models beyond human supervision
Self-evolving framework for multiscale life modelling
Life Operators (2026) proposes a self-evolving framework for multiscale life modelling, extending the agent-evolution agenda into the life sciences. It is the second last-author preprint of 2026 in the catalog and has not yet been reviewed by Evolve.
Timeline
- 2026Publishes LexAgentHallu, a hierarchical benchmark profiling hallucinations in legal agents across 3,414 cases with a 27-class taxonomy.
- 2026Publishes a theoretical analysis of safe harness self-evolution, proving feasibility limits and divergent evaluation cost near the reward ceiling.
- 2026Publishes, as last author, a survey on scaling large reasoning models beyond human supervision with an L0–L4 self-improvement ladder.
- 2026Publishes Life Operators, a self-evolving framework for multiscale life modelling, again as last author.
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 indicators (h-index, i10-index, counts by year, topics) come from an OpenAlex profile; the output and impact dimensions lack computed scores and are left unscored.
- This profile is name-resolved from public bibliographic data; the very broad topic spread (e.g., asthma and respiratory diseases alongside cloud computing) may reflect entity-resolution noise rather than one person's agenda.
- Reproducibility scoring covers only 2 reviewed papers (mean 55); the remaining catalog entries, including Life Operators, are unscored and unreviewed by Evolve.
- All cataloged works are 2026 arXiv preprints; no peer-reviewed venue, grant, or career details are present in the provided data.