First-author NLP researcher auditing whether 'successful' knowledge edits truly erase facts
1
papers here
1
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-21) 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—
Not yet scored: the catalog holds only one indexed paper (the 2026 arXiv preprint), too thin to measure publication volume or venue mix. This axis would track output volume and types as records accumulate.
Impact—
Not yet scored: with a single 2026 preprint, no citation or downstream-uptake data exists in the record. This axis would measure citations, influence, and uptake once the paper accrues a citation history.
Focus100/100
Focus scores a perfect 100: per the field distribution, the sole indexed paper falls entirely within natural language processing (knowledge editing / interpretability). With a single record this concentration is real but fragile — best read as 'no drift observed yet' rather than a proven long-term specialty. The early signal points squarely at knowledge-editing audits and probing-based interpretability.
Collaboration14/100
Collaboration scores 14, computed from exactly one distinct coauthor in the indexed data — Romit Chatterjee, the recurring partner on the 2026 first-author preprint. That is a two-person working pattern rather than a lab-scale network. The low score reflects sample size as much as working style, and should rise if further coauthored work is indexed.
Activity—
Not yet scored: the data spans a single publication year (2026), so publishing cadence and year-over-year consistency cannot be computed. This axis would chart output rhythm across years as further records appear.
Reproducibility55/100
Reproducibility averages 55 across one assessed paper — the 2026 knowledge-editing audit — which Evolve tags with medium reproducibility readiness. The mid-scale score fits the study's profile: a GPT-2-XL and CounterFact probing setup that is replicable in principle, while the verdict flags a missing pre-edit baseline and thin statistics as obstacles to independent verification.
Research summary
Priyansh Srivastava is an early-career researcher working at the intersection of knowledge editing and interpretability in natural language processing. The Evolve catalog indexes one paper to date — the 2026 arXiv preprint 'Suppressed, Not Erased: A Representational Trace of Edited Facts Survives Even Weight-Free Knowledge Editing' — on which Srivastava is first author. The study audits GPT-2-XL after knowledge edits rated 100% successful, including the weight-free GRACE method, and uses linear probing on CounterFact-style edits to show that original facts remain decodable from internal representations. Evolve's verdict calls the finding promising but statistically thin, noting the absence of a pre-edit baseline; reproducibility readiness is medium, with a mean of 55 across one assessed paper. The bibliometric radar reflects a narrow but deep early record: a perfect focus score of 100, since the single indexed paper falls entirely within NLP knowledge editing and interpretability; a collaboration score of 14 from exactly one distinct coauthor, Romit Chatterjee; and reproducibility at 55. Output, impact, and activity are not yet computable from the available data. Read this as an early-stage profile, name-resolved from public bibliographic data, that will firm up as further publications are indexed.
Collaboration pattern
The indexed record shows a tightly paired working mode: exactly one distinct coauthor, Romit Chatterjee, who is also the recurring frequent collaborator on the 2026 first-author arXiv preprint. This yields a low collaboration score of 14, reflecting a two-person collaboration rather than a broad network. No institutional affiliation is available in the data, so the setting of this partnership cannot be determined.
Research keywords
Highlights
Edited facts stay decodable after 'successful' editing
An audit lens on knowledge editing
Promising but early-stage evidence
Timeline
- 2026First-author arXiv preprint 'Suppressed, Not Erased' appears, auditing GPT-2-XL after 100%-successful knowledge edits including weight-free GRACE.
- 2026Collaborates with Romit Chatterjee, the single distinct coauthor in the indexed record, yielding a collaboration score of 14.
- 2026The paper enters the Evolve catalog with a 'promising but statistically thin' verdict and medium reproducibility readiness (mean 55, one assessed paper).
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.
- Radar scores are derived from public bibliometric data only; the record lacks institution, country, and ORCID for this researcher.
- The profile rests on a single indexed paper (2026, arXiv), so every score is small-sample sensitive and may shift as more records are indexed.
- Output, impact, and activity dimensions are unscored because no publication-volume, citation, or multi-year data is present.
- Name resolution is based on public bibliographic data; 'Priyansh Srivastava' may be shared by multiple researchers.