University of Pisa NLP researcher in LLM evaluation and mechanistic interpretability — likelihood ranking, SAE latents
2
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
1
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-28) 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 output score is available yet: the snapshot lacks a reliable total-works count. This axis would measure cumulative publication volume — total indexed papers across venues and years — once the OpenAlex record is more complete.
Impact—
No impact score is available yet: the snapshot reports only h-index 1 and i10-index 1 with no citation totals, far too thin to rate. This axis would measure citation-based influence and uptake of the work by researchers.
Focus100/100
The focus score of 100 reflects a single assigned field in the OpenAlex distribution — 'natural language processing / mechanistic interpretability' — and a topic profile concentrated in NLP: topic modeling, NLP techniques, sentiment analysis and opinion mining, and emotion recognition, with one software-engineering topic. Both reviewed papers sit squarely in LLM evaluation and interpretability, indicating a tightly scoped agenda.
Collaboration32/100
The collaboration score of 32 is the lowest scored axis: OpenAlex records only 4 distinct coauthors in the recent window, and both reviewed 2026 papers list this author in a middle position without corresponding-author status. Recurring coauthors include Alessandro Bondielli, Alessandro Lenci, Serena Auriemma and Davide Bacciu, suggesting steady work within a small, stable team rather than a wide network.
Activity67/100
Activity scores 67 on the strength of 9 works in the last three years (recent3yWorks), with OpenAlex counts-by-year showing 1 indexed work in 2024 rising to 8 in 2026 — a steep recent ramp. Both catalog entries appeared as 2026 arXiv preprints, consistent with an early-career output phase concentrated in the current year.
Reproducibility100/100
Evolve's reproducibility mean for this author is 100, computed from one reviewed paper (reproPapers: 1) — the Parts-of-Speech SAE study, rated high readiness for its honestly hedged single-model/layer and probe-dependence limits; the likelihood-ranking paper is rated medium. The perfect score rests on a very small base and should be read as a promising early signal, not a settled record.
Research summary
Lucia Passaro is a researcher at the University of Pisa working in natural language processing; this profile is name-resolved from public OpenAlex bibliographic data (ORCID 0000-0003-4934-5344). The snapshot depicts an early-career record: h-index 1, i10-index 1, 9 works in the last three years, and counts-by-year rising from 1 indexed work in 2024 to 8 in 2026. Topic assignments concentrate in NLP — topic modeling, NLP techniques, sentiment analysis and opinion mining, emotion and mood recognition — plus one software-engineering topic, and the field distribution lists a single field, 'natural language processing / mechanistic interpretability', yielding a focus score of 100. Two Evolve-reviewed 2026 arXiv papers, both co-authored from middle positions, define the current agenda. 'Likelihood Ranking doesn't Scale Like Prompting in LLMs' finds that across 95 LLMs statement-likelihood accuracy stays flat with scale while prompted accuracy soars. 'Parts-of-Speech as Emergent Categories in SAE Latent Space' argues that PoS structure in LLaMA-3-8B sparse-autoencoder latents is distributed and category-dependent. Reproducibility is scored 100 (mean over one reviewed paper), with readiness rated high and medium respectively. Collaboration is the weakest scored axis (32): 4 distinct coauthors, with frequent collaborators Alessandro Bondielli, Alessandro Lenci, Serena Auriemma and Davide Bacciu. Output and impact remain unscored for lack of data.
Collaboration pattern
Collaboration is the weakest scored axis (32): the OpenAlex window lists only 4 distinct coauthors, and both reviewed 2026 papers place this author in a middle position without corresponding-author status, indicating contribution within a team rather than leading a group. The recurring partners are Alessandro Bondielli, Alessandro Lenci, Serena Auriemma and Davide Bacciu — a small, stable set consistent with an early-career researcher embedded in one core collaboration cluster.
Research keywords
Highlights
Likelihood ranking stays flat while prompting scales
Parts of speech emerge as distributed SAE structure
Timeline
- 2024OpenAlex counts-by-year records the earliest indexed work in the snapshot: 1 publication in 2024, marking the start of the visible record.
- 2026Output ramps sharply: 8 indexed works in 2026, the most active year in the record, within 9 works over the last three years.
- 2026Co-authors the Evolve-reviewed arXiv study 'Likelihood Ranking doesn't Scale Like Prompting in LLMs', in a middle authorship position.
- 2026Co-authors the Evolve-reviewed arXiv study 'Parts-of-Speech as Emergent Categories in SAE Latent Space', also from a middle position; the paper earns a reproducibility score of 100.
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 figures come from the OpenAlex record only (identity: openalex); no Google Scholar, Scopus or publisher data is included.
- Citation signal is minimal in the snapshot (h-index 1, i10-index 1, no citation totals), so the output and impact axes are unscored (dimMissing) rather than estimated.
- The reproducibility score of 100 is a mean over a single reviewed paper (reproPapers: 1); only two papers by this author are in the Evolve catalog.
- This profile is name-resolved from public bibliographic data (ORCID 0000-0003-4934-5344); biography, career history and awards beyond the snapshot are not covered.