Illinois Institute of Technology researcher in LLM safety: benchmarks, calibration, and conformal prediction.
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
0
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—
Output data is insufficient to score this axis. It would measure publication volume and consistency over time — total works, yearly counts, and venue spread — but the indexed record is too new or sparse for a reliable estimate.
Impact—
Impact cannot be scored yet: the snapshot reports an h-index of 0 and an i10-index of 0 with no citation distribution. This axis would measure citations and uptake relative to field norms once the work has time to accumulate citations.
Focus0/100
The radar focus score is 0, reflecting an even 0.5/0.5 split in the OpenAlex field distribution between natural language processing (AI safety / trustworthy ML) and machine learning (uncertainty quantification and LLM safety). The topic profile spans five topics, led by adversarial robustness in machine learning (9) and ethics and social impacts of AI (8), with topic modeling (5), hate speech detection (4), and network security (2) rounding it out.
Collaboration22/100
The collaboration score is 22, consistent with a very small network: the record lists only 2 distinct co-authors, and the recurring collaborator is Hanyu Su. Both Evolve-reviewed papers are first-authored but not corresponding, so writing leadership is present while the co-author pool stays narrow. This compact pattern is typical of an early-stage researcher building a footprint around a small, stable team rather than a broad consortium.
Activity77/100
The activity score is 77. The OpenAlex counts-by-year snapshot records 13 works in 2026, and recent3yWorks is also 13, meaning the entire indexed output falls inside the recent three-year window. Two Evolve-reviewed first-author papers appeared on arXiv in 2026. Concentrating all indexed output in a single year marks a burst of early-career productivity rather than long-run accumulation.
Reproducibility100/100
The reproducibility score is 100, the maximum, computed across 2 papers (reproMean 100). Both Evolve-reviewed papers carry a high reproducibility readiness rating: Safety-Flag is described as a solid, openly released benchmark, and the conformal-prediction study is characterized by broad and honest evidence. For a researcher whose work centers on calibration and reliability, an open, checkable artifact trail is a distinguishing strength.
Research summary
Yibo Hu is a researcher at Illinois Institute of Technology whose work sits at the junction of large language model safety and statistical reliability. The OpenAlex record splits evenly (0.5/0.5) between natural language processing (AI safety / trustworthy ML) and machine learning (uncertainty quantification and LLM safety), and the topic profile is led by adversarial robustness in machine learning (9) and ethics and social impacts of AI (8). Two first-authored 2026 arXiv papers anchor the profile. Safety-Flag unifies seven safety benchmarks into one balanced flag protocol for LLM content moderators, showing that error direction, calibration, and abstention diverge across moderators; Conformity Breaks Conformal Prediction demonstrates that peer-pressure conformity among multi-agent LLMs can destroy clean-calibrated conformal coverage, while honestly noting that the threat rests on stylized transcripts. Both papers are rated high for reproducibility readiness, and the radar reproducibility score is 100 across two papers. Activity is high (radar 77): the counts-by-year snapshot records 13 works in 2026, all inside the recent three-year window. The collaboration footprint is small — 2 distinct co-authors, with Hanyu Su the recurring partner — and both reviewed papers are first-authored rather than corresponding. Citation impact is not yet measurable (h-index 0), making this an early-stage profile whose benchmarks and guarantees, rather than accumulated citations, are the current evidence of influence.
Collaboration pattern
The collaboration footprint is compact: the OpenAlex record lists only 2 distinct co-authors, yielding a radar collaboration score of 22. Hanyu Su is the recurring collaborator in the frequent-coauthor list. Both Evolve-reviewed papers are first-authored but not corresponding-author works, indicating writing leadership within a small stable team. Note the coauthor list also contains a self-matching entry, a likely name-resolution artifact in the underlying bibliographic data.
Research keywords
Highlights
Safety-Flag: one balanced flag protocol across 7 LLM-moderator safety benchmarks
Peer-pressure conformity erodes conformal coverage guarantees in multi-agent LLMs
Reliability-first agenda with fully open artifacts
Across both reviewed papers the common thread is testable reliability — calibration, selective prediction, and coverage guarantees — with both rated high for reproducibility readiness and a radar reproducibility score of 100 (reproMean across 2 papers), a distinguishing early-career signal.
Timeline
- 2026First-authored Safety-Flag on arXiv, unifying seven LLM-moderator safety benchmarks into one balanced flag protocol.
- 2026First-authored Conformity Breaks Conformal Prediction on arXiv, showing peer-pressure conformity degrades conformal coverage for LLMs.
- 2026OpenAlex snapshot records 13 works in 2026, all within the recent three-year window, with a radar activity score of 77.
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 records only (identity: openalex); no ORCID was provided to confirm the author record.
- Output and impact axes are unscored: the snapshot reports h-index 0 and i10-index 0, with indexed works concentrated in a single year (2026: 13).
- The frequent-coauthor list includes a self-matching entry (Yibo Hu), so name-resolution noise is possible and topic counts may aggregate records beyond the 2 reviewed papers.
- This profile rests on 2 Evolve-reviewed 2026 arXiv papers; the 13-work 2026 count is not itemized in this catalog.