Skip to content
Chain-of-Thought Pioneer Jason Wei Joins $20B AI Data Unicorn Mercor
Industry Insights 11 min read

Chain-of-Thought Pioneer Jason Wei Joins $20B AI Data Unicorn Mercor

Chain-of-Thought prompting inventor Jason Wei leaves Meta's Superintelligence Lab to become Chief Scientist at Mercor, a $20B AI startup shifting from expert talent matching to building evaluation benchmarks and post-training infrastructure for frontier models, alongside LoRA co-author Edward Hu.

Machine Heart
Machine Heart
Machine Heart

Jason Wei's Research Journey

Jason Wei, widely recognized as the "father of Chain-of-Thought," has held research positions at Google Brain, OpenAI, and most recently Meta's Superintelligence Lab (TBD Lab). His 2022 paper Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (first author, Google Brain) introduced Chain-of-Thought (CoT) prompting — a method that adds intermediate reasoning steps to prompts, significantly improving multi-step reasoning in large language models without modifying model parameters. This technique sparked extensive follow-up work on self-consistency and reasoning-path search.

At OpenAI (2023 onward), Wei contributed to the o1 and Deep Research projects and co-developed several evaluation benchmarks:

SimpleQA (2024): 4,326 fact-seeking questions measuring factual accuracy and awareness of unanswerable queries.

BrowseComp (2025): Evaluates complex web-search capabilities requiring multi-hop retrieval, filtering, and cross-verification.

HealthBench : Uses physician-designed rubrics to assess accuracy, communication quality, and instruction following in realistic medical dialogues.

In July 2025, Wei moved from OpenAI to Meta's newly formed Superintelligence Lab, becoming one of Meta's high-profile hires from a rival lab.

Mercor's Evolution and Funding

Founded in 2023 by Brendan Foody, Adarsh Hiremath, and Surya Midha, Mercor began as an AI-driven talent-recruitment platform. With the rise of generative AI, it pivoted to supplying frontier model companies with expert talent, training data, and evaluation services. Its expert network now exceeds 5 million professionals across domains such as finance, law, and software engineering.

On November 10, 2025, Mercor announced a $750 million Series D round at a $20 billion valuation, backed by Benchmark, General Catalyst, Nvidia, and Salesforce. The company stated the capital would fund deeper investment in model evaluation, post-training infrastructure, and environments for AI agents to learn complex workflows . Mercor is also expanding from serving frontier labs to helping enterprises build custom evaluation suites and improve models with domain-specific expert feedback.

Mercor's Research Agenda and Team

In an October 8, 2025 article Why Mercor Is Building a Research Team , the company outlined a loop: evaluate real-world tasks to uncover model failures → use domain experts to diagnose causes and create training data → post-train models → re-evaluate. Research focuses include real-world task evaluation, training-data quality, post-training methods, and how human expertise drives AI capability gains.

Mercor has released the APEX benchmark series covering management consulting, investment banking, corporate law, accounting, and software engineering — testing AI on realistic professional tasks rather than abstract knowledge QA. In post-training experiments, Mercor fine-tuned Qwen3.5-397B on APEX-Agents, raising Pass@1 from 16% to 27% .

The research team already includes notable figures:

Edward J. Hu — first author of LoRA: Low-Rank Adaptation of Large Language Models (parameter-efficient fine-tuning via low-rank matrices, now a standard technique).

Minyang Tian — first author of SciCode and CritPt.

Victor Barres — first author of tau²-bench.

Researchers from Anthropic, Ai2, and other institutions.

Jason Wei's Move and Strategic Fit

On November 18, 2025, Mercor CEO Brendan Foody announced Wei's appointment as Chief Scientist; Wei confirmed the move, noting that after a "great year-plus at Meta TBD Lab" he was excited to join Mercor because "as more companies train AI models, data will shape the next frontier of AI development." Foody revealed Wei had been both a Mercor client and a close collaborator for years. Wei will help build the research team with emphasis on data and evaluation — a natural extension of his OpenAI benchmark work (SimpleQA, BrowseComp, HealthBench) into Mercor's real-world professional domains.

The convergence of Wei (CoT, evaluation benchmarks) and Hu (LoRA, efficient adaptation) at a company building expert-driven evaluation and post-training infrastructure signals a shift: the next AI frontier may be defined not only by model architecture but by high-quality, domain-specific data and rigorous, task-grounded evaluation.

Reference links: https://x.com/BrendanFoody/status/2122825607763689873 https://x.com/_jasonwei/status/2122826605362127094
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.

LoRAAI evaluationChain-of-ThoughtbenchmarksAI talentpost-trainingMercorJason Wei
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.