NLP researcher at Tarsus University focused on LLM reliability, selective prediction, and epistemic honesty
2
papers here
2
reviewed by Evolve
2
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—
The output axis lacks sufficient data: only two catalog papers are in scope, so publication volume and venue spread cannot be characterized. This dimension would measure total publication count, output cadence, and venue types across the researcher's career.
Impact—
The impact axis is unscored because no citation counts, h-index, or usage signals are available in the provided data. It would measure how widely the two arXiv preprints have been cited, reused, or taken up by the field.
Focus0/100
The radar focus score is 0/100. Its underlying field distribution splits evenly (0.5/0.5) between two adjacent NLP subfields — LLM evaluation and LLM reliability/selective prediction — so the scored axis reads zero even though both cataloged papers pursue one theme: when LLMs should abstain and how honestly they report knowledge boundaries.
Collaboration22/100
Radar scores collaboration at 22/100, consistent with a small footprint: only 2 distinct co-authors appear in the underlying bibliometric data, and both catalog papers are first-author arXiv preprints without corresponding-author flags. The profile separately lists H. Russell Bernard and Huan Liu as frequent co-authors, though the two cataloged papers alone cannot establish the extent of that joint work.
Activity—
The activity axis has insufficient data: with both catalog papers dated 2026, there is no multi-year record to estimate current research tempo. It would measure publication recency, yearly throughput, and sustained engagement over time.
Reproducibility78/100
Reproducibility is the strongest scored axis at 78/100, a mean across two papers (reproMean 78, reproPapers 2). Evolve rates the epistemic-honesty benchmark paper at high reproducibility readiness and the CoSQ selective-risk paper at medium, indicating benchmark- and artifact-oriented work that others can rerun with modest effort.
Research summary
Ali Şenol is a researcher at Tarsus University working in natural language processing, with this profile resolved from public bibliographic data under ORCID 0000-0003-0364-2837. The Evolve catalog contains two 2026 arXiv preprints, both with Şenol as first author, concentrated on one question: when should large language models say 'I don't know'? The first introduces Chain-of-Self-Questioning (CoSQ), a prompting-based method for selective risk control that reportedly cuts wrong commitments on TruthfulQA from 13.1% to 8.9% across 11 models; Evolve's review credits strong paired statistics but notes a missing chain-of-thought abstention baseline weakens attribution. The second proposes a fully behavioral, non-cognitive measure of epistemic honesty with a 3,000-question benchmark; the review calls it rigorous and self-aware, while classifier consistency and component redundancy temper the absolute score. Radar scores place reproducibility at 78/100 across the two papers (one high, one medium reproducibility readiness), collaboration at 22/100 with 2 distinct co-authors in scope, and focus at 0/100 despite an even field split between LLM evaluation and LLM reliability/selective prediction. Output, impact, and activity axes lack sufficient data, and H. Russell Bernard and Huan Liu appear as listed frequent co-authors.
Collaboration pattern
The collaboration footprint in scope is small: radar scores it 22/100 with only 2 distinct co-authors in the bibliometric data, and both catalog papers are first-author arXiv preprints without corresponding-author flags. The profile lists H. Russell Bernard and Huan Liu as frequent co-authors, but the two cataloged NLP papers cannot confirm how much of that collaboration is reflected in published output.
Research keywords
Highlights
CoSQ: prompt-gated abstention cuts wrong commitments on TruthfulQA
A fully behavioral measure of LLM epistemic honesty with a 3,000-question benchmark
Timeline
- 2026Publishes first-author arXiv preprint introducing Chain-of-Self-Questioning (CoSQ), a prompt-gating approach to selective risk control for LLMs.
- 2026Publishes a fully behavioral, non-cognitive measure of epistemic honesty with a 3,000-question benchmark, again as first author.
- 2026Both papers enter the Evolve catalog with reproducibility readiness rated medium and high respectively, yielding a 78/100 reproducibility mean.
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.
- Dimension scores come from a limited bibliometric snapshot; the output, impact, and activity axes lack underlying data.
- Identity is resolved from public bibliographic data under ORCID 0000-0003-0364-2837; 'Ali Şenol' may be a name shared by multiple researchers.
- Both cataloged papers are 2026 arXiv preprints; no peer-reviewed venue, citation, or career-history information is in scope.
- Only the two Evolve-reviewed papers inform this profile; the researcher's complete publication record may be broader.