PeopleofSuperintelligence

Researcher

Bertie Vidgen

Also known as Bertram Vidgen

Bertie Vidgen is a researcher whose work spans online-hate detection, language-model safety and evaluation of AI agents. At Oxford, he researched Islamophobic hate speech among followers of UK political parties on Twitter and coauthored a dynamically generated dataset for hate detection. He coauthored XSTest, a test for exaggerated safety refusals, and PRISM, a cross-country human-feedback dataset that won NeurIPS's 2024 Best Paper award in the Datasets and Benchmarks track. In 2025, he wrote about human-annotated data for RAG agents at Contextual AI and coauthored Mercor's APEX benchmark for economically valuable knowledge work. In September 2026, he coauthored APEX-Agents 1.1; as of October 2026, he is a researcher on Mercor's research team and first author of APEX-Agents.

Now at
Mercor
Badges

Timeline

  1. 2015–2019

    DPhil research on online hate

    He holds a BA in History and Politics from Warwick and an MA in Ideology and Discourse Analysis from Essex. From October 2015 to June 2019, he studies for a DPhil at Oxford's Internet Institute, researching Islamophobic hate speech among followers of UK political parties on Twitter.

  2. October 2019–December 2023

    Research Associate at Oxford

    From October 2019 to December 2023, he is a Research Associate at the Oxford Internet Institute. He studies hate speech in news and social media using machine learning, natural language processing and statistical modelling. He also works as a post-doctoral researcher within the Alan Turing Institute's Public Policy Programme.

  3. 2021

    Learning from the Worst

    In 2021, he coauthors Learning from the Worst, a hate-detection dataset with 40,000 entries generated and labelled by trained annotators over four rounds. The paper reports that models trained on later rounds perform better on test sets and are harder for annotators to trick.

  4. 2024

    XSTest safety benchmark

    In 2024, he coauthors XSTest, a test suite for exaggerated safety behaviours in large language models. It pairs 250 safe prompts that well-calibrated models should comply with and 200 unsafe prompts they should refuse.

  5. 2024

    NeurIPS Best Paper for PRISM

    In 2024, he coauthors PRISM, a dataset linking the stated preferences of 1,500 participants in 75 countries to fine-grained feedback in 8,011 live conversations with 21 LLMs. NeurIPS names it Best Paper in the Datasets and Benchmarks track.

  6. March 2025

    Human data for RAG agents

    In March 2025, he describes how Contextual AI's Data team creates human-annotated datasets to specialize and evaluate RAG agents for customer use cases.

  7. October 2025

    Mercor's APEX benchmark

    In October 2025, he coauthors Mercor's AI Productivity Index (APEX), a benchmark that evaluates AI models on economically valuable knowledge work.

  8. As of October 2026

    Researcher at Mercor

    As of October 2026, he is a researcher on Mercor's Research team and first author of APEX-Agents. In September 2026, he coauthors APEX-Agents 1.1, a benchmark that tests models on real-world work across investment banking, management consulting and corporate law.

Career

  1. NowMercorResearcher
  2. 2019–2023University of OxfordResearch Associate

Education

  1. 2019University of OxfordDPhil
  2. University of WarwickBA in History and Politics
  3. University of EssexMA in Ideology and Discourse Analysis

From the biography.

  1. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Bertie Vidgen. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.
  2. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes University of Oxford. We estimate where the people shaping AI worked when their most-cited papers appeared: the company share was 31% in 2020–2024 and 8% in 1970–1974, and Google led all organizations in 2020–2024.
  3. The median h-index is 48 at universities and 14 at frontier AI labsIncludes University of Oxford. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  4. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes University of Oxford. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Bertie Vidgen yet. Send us a free-licensed photo.

  1. OII | Bertram Vidgen oii.ox.ac.uk. Checked 22 October 2026.
  2. Learning from the Worst: Dynamically Generated Datasets to Improve Online Hate Detection aclanthology.org. Checked 22 October 2026.
  3. XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models aclanthology.org. Checked 22 October 2026.
  4. NeurIPS 2024 Awards neurips.cc. Checked 22 October 2026.
  5. The PRISM Alignment Dataset: What Participatory, Representative and Individualised Human Feedback Reveals About the Subjective and Multicultural Alignment of Large Language Models proceedings.neurips.cc. Checked 22 October 2026.
  6. Building better agents: The critical role of human-annotated data contextual.ai. Checked 22 October 2026.
  7. Introducing APEX: The AI Productivity Index mercor.com. Checked 22 October 2026.
  8. Why Mercor is building a research team mercor.com. Checked 22 October 2026.
  9. Introducing APEX-Agents 1.1 mercor.com. Checked 22 October 2026.