PeopleofSuperintelligence

Researcher

Philippe Laban

Philippe Laban is an NLP and human-computer interaction researcher whose benchmarks test how language models summarize, converse and handle delegated work. He completed a computer science PhD at UC Berkeley in 2021, advised by Marti Hearst and John Canny. After Berkeley, he worked as a Research Scientist at Salesforce, where in 2023 he coauthored the SummEdits benchmark and the FlipFlop experiment, which measures how often LLMs flip their answers when challenged. In 2025, he and three coauthors showed that top LLMs score an average of 39% lower in multi-turn than in single-turn conversations, and ICLR named the paper an Outstanding Paper in 2026. He is a Research Scientist at Microsoft Research in New York.

Badges

Timeline

  1. 2021

    PhD at UC Berkeley

    He earns a PhD in computer science from UC Berkeley's EECS Department in 2021, advised by John Canny and Marti Hearst. His dissertation, "Unsupervised Text Generation and its Application to News Interfaces," develops summarization and simplification methods that need no training data and builds a news chatbot and an automated interactive podcast.

  2. November 2021

    The SummaC benchmark

    In November 2021, he coauthors SummaC with Tobias Schnabel, Paul N. Bennett and Marti A. Hearst. Its SummaCConv method lets natural language inference models detect factual inconsistencies in summaries and reaches 74.4% balanced accuracy on a new benchmark of six datasets.

  3. By 2023

    Research scientist at Salesforce

    After the PhD, he takes a job as a Research Scientist at Salesforce. In May 2023, working at Salesforce AI, he is first author of SummEdits, a 10-domain benchmark for detecting factual inconsistencies in summaries that is 20 times more cost-effective per sample than earlier ones; most LLMs score close to random chance on it.

  4. November 2023

    The FlipFlop experiment

    In November 2023, he coauthors the FlipFlop experiment with Lidiya Murakhovs'ka, Caiming Xiong and Chien-Sheng Wu. Across ten LLMs and seven classification tasks, models flip their answers on average 46% of the time when challenged with "Are you sure?", and accuracy drops by an average of 17%.

  5. May 2025

    LLMs get lost in conversation

    In May 2025, he is a Research Scientist at Microsoft Research in New York and publishes "LLMs Get Lost In Multi-Turn Conversation" with Hiroaki Hayashi, Yingbo Zhou and Jennifer Neville. Simulations of 200,000+ conversations show that top open- and closed-weight LLMs score an average of 39% lower in multi-turn than in single-turn conversations across six generation tasks.

  6. April 2026

    ICLR 2026 Outstanding Paper

    In April 2026, ICLR names "LLMs Get Lost In Multi-Turn Conversation" one of its two Outstanding Papers for 2026. The awards committee credits the work with exceptional experimental design and methodology.

  7. April 2026

    The DELEGATE-52 benchmark

    In April 2026, he coauthors DELEGATE-52 with Tobias Schnabel and Jennifer Neville, a benchmark of long delegated workflows in 52 professional domains. In tests of 19 LLMs, even frontier models corrupt an average of 25% of document content by the end of long workflows.

  8. As of October 2026

    Microsoft Research in New York

    He works as a Research Scientist at Microsoft Research in New York, at the intersection of NLP and HCI. His focus tasks are summarization, simplification and LLM evaluation and benchmarking.

Career

  1. NowMicrosoft ResearchResearch Scientist
  2. SalesforceResearch Scientist

Education

  1. 2021University of California, BerkeleyPhD

From the biography.

  1. 16% of AI career moves in the 2020s went from a university to a companyIncludes Philippe Laban. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  2. Among employers with 30+ full-time cases, OpenAI leads primary technical role H-1B pay at $310,000, and Amazon, Google and Microsoft file 57% of all casesIncludes Microsoft. Across 440,669 certified US H-1B filings from 44 employers (FY2020–FY2026), OpenAI leads primary technical role base pay at $310,000 and xAI leads overall base pay at $290,000.
  3. Google had the most people shaping AI every year from 2014 to 2025Includes Microsoft. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  4. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes Microsoft. 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.

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 Philippe Laban yet. Send us a free-licensed photo.

  1. Unsupervised Text Generation and its Application to News Interfaces | EECS at UC Berkeley www2.eecs.berkeley.edu. Checked 22 October 2026.
  2. Philippe Laban - Researcher tingofurro.github.io. Checked 22 October 2026.
  3. SummaC: Re-Visiting NLI-based Models for Inconsistency Detection in Summarization arxiv.org. Checked 22 October 2026.
  4. Marti A. Hearst: Home page people.ischool.berkeley.edu. Checked 22 October 2026.
  5. LLMs as Factual Reasoners: Insights from Existing Benchmarks and Beyond arxiv.org. Checked 22 October 2026.
  6. LLMs as Factual Reasoners: Insights from Existing Benchmarks and Beyond (HTML version) arxiv.org. Checked 22 October 2026.
  7. Are You Sure? Challenging LLMs Leads to Performance Drops in The FlipFlop Experiment arxiv.org. Checked 22 October 2026.
  8. LLMs Get Lost In Multi-Turn Conversation arxiv.org. Checked 22 October 2026.
  9. LLMs Get Lost In Multi-Turn Conversation - Microsoft Research microsoft.com. Checked 22 October 2026.
  10. On Interacting and Writing with LLMs - Talks.cam talks.cam.ac.uk. Checked 22 October 2026.
  11. Announcing the ICLR 2026 Outstanding Papers - ICLR Blog blog.iclr.cc. Checked 22 October 2026.
  12. LLMs Corrupt Your Documents When You Delegate arxiv.org. Checked 22 October 2026.