Skip to content
Chain-of-Thought Author Jason Wei Leaves Meta for $20B Data Startup Mercor After One Year

Chain-of-Thought Author Jason Wei Leaves Meta for $20B Data Startup Mercor After One Year

Chain-of-Thought co-author Jason Wei departs Meta's TBD Lab after roughly one year to become Chief Scientist at Mercor, a $20B-valued startup building expert-driven AI training data and evaluation platforms, underscoring a growing industry shift toward high-quality professional data as the next frontier and a pattern of top AI talent leaving big tech after short tenures.

Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing

Jason Wei's Career and Move to Mercor

Jason Wei, co-author of the seminal Chain-of-Thought (CoT) paper and first author of the FLAN instruction-tuning work, announced on X that he is leaving Meta's TBD Lab after roughly one year to join Mercor as Chief Scientist.

Background: From Google Brain to OpenAI

Wei graduated from Dartmouth College in 2020 and joined Google Brain directly, becoming a Research Scientist at age 23 — possibly the youngest in Google's history. In 2022 he published the CoT paper, showing that prompting models to reason step by step dramatically improves performance. He also co-authored FLAN, a foundational instruction-tuning method. In 2023 he moved to OpenAI, where he contributed to the o1 reasoning model and Deep Research, and created benchmarks including SimpleQA, BrowseComp, and HealthBench (developed with 262 physicians). His Google Scholar citations exceed 160,000.

Work at Meta: Muse Spark and Medical Research

Meta recruited Wei in July 2024 with a reported nine-figure package. At the TBD Lab he led development of the Muse Spark series and applied models to medical research. A recent experiment used Muse Spark 1.3 and Muse Wonder to analyze about 5 million de-identified electronic health records. Doctors posed questions; models wrote code and ran analyses. In four weeks the team produced ten research manuscripts. Wei observed that while models accelerate hypothesis testing, clinical data contains massive noise — deciding which variables matter and interpreting results still requires expert judgment.

Mercor: Expert-Driven Data and Evaluation

Mercor, founded in January 2023 by Brendan Foody, Adarsh Hiremath, and Surya Midha (high-school debate teammates from Harvard and Georgetown), provides AI training data and evaluation services. It organizes domain experts — doctors, lawyers, investment bankers, senior engineers — to decompose professional work into high-difficulty training tasks with strict scoring rubrics. This helps AI labs train and test models on complex professional problems. Mercor also helps enterprises monetize proprietary data and deploy AI in workflows.

Mercor founding team
Mercor founding team

Funding and Valuation

On November 10, 2024, Mercor announced a $750 million Series D at a $20 billion valuation, led by Felicis with participation from Benchmark, General Catalyst, Sapphire Ventures, Nvidia, Salesforce, and Zoom. The investment thesis: general internet corpora are exhausted; only the 'tacit judgments' of seasoned professionals (e.g., a lawyer spotting contract risks) can teach AI to truly perform expert work. Clients now include foundational model labs and vertical leaders like legal AI company Harvey and finance software Ramp.

Strategic Fit and Prior Relationship

Wei's move aligns with his statement that 'as more companies train models, data will shape the next frontier.' Mercor CEO Brendan Foody said Wei will build the research team and oversee all data and evaluation. LoRA co-author Edward Hu joins to lead model training. Notably, Meta was already a Mercor customer, and the TBD Lab had used Mercor's annotation services; Foody and Wei had maintained close contact for years.

Industry Trend: The One-Year Cliff at Big Tech

Wei's departure follows a pattern: former Apple foundation model lead Pang Ruoming left after seven months (forfeiting over $200M) to join OpenAI; OpenAI researchers Ethan Knight and Avi Verma left Meta after only weeks to return to OpenAI. Observers note that big-tech offers often include a one-year cliff for large signing bonuses and stock vesting; once that milestone passes, retention depends on the lab's real technical vision. Meanwhile, Meta's 1997-born Chief AI Officer Alexander Wang remains in place.

References: https://x.com/BrendanFoody/status/2122825607763689873, https://www.mercor.com/
Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

Chain-of-ThoughtAI data annotationMeta AIOpenAI o1MercorJason Weiexpert datatalent movement
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.