Newshot

Chain-of-Thought co-author Jason Wei leaves Meta after just one year

AI summary

Just over a year after joining Meta, they have packed up and left.

Translation(model translation, not the source's own text)

CoT First Author Jason Wei Leaves Meta! Why Quit After Just One Year?

Just over a year after joining Meta, he up and left!

henry from 凹非寺 Quantum位 | WeChat Account QbitAI

Unbelievable! Last year, Zuck handed out a sky-high nine-figure annual package to poach the "Chain of Thought" author Jason Wei from OpenAI—

and just over a year after joining Meta, he up and left!

And his next stop isn't some giant, but a data startup that's just getting off the ground...

Just moments ago, Jason Wei announced on X:

He will leave Meta's TBD Lab and join the startup Mercor as Chief Scientist.

At the same time, in his post he shared his judgment on the next phase of AI: as more and more companies begin training AI models, data will shape the next frontier.

Mercor co-founder and CEO Brendan Foody, in a welcoming post, further revealed that Jason will help build the research team and oversee all data and evaluation work.

Worth noting, fighting alongside Jason is LoRA paper first author Edward Hu, who will be in charge of model training.

Although it came suddenly—so much so that when Jason made the announcement he hadn't even had time to update his X bio—this job change wasn't entirely without warning signs.

Reportedly, Meta has been a Mercor client for the past few years, and the TBD Lab where Jason works has also used Mercor's data annotation services.

Not only that, Mercor's CEO also said he has maintained close communication with Jason over the past few years.

All one can say is: big companies spend money buying data from startups, and the startups casually poach the person buying the data too.

In the future, when big companies collaborate with startups, they really need to be more careful.

Who is Jason Wei?

The reason Jason Wei's moves draw so much attention isn't just that he's the author of a series of classic works like "Chain of Thought," but also that he's the kind of "young genius" who skipped a PhD and went straight from an undergraduate degree into the core of the AI world.

(Though there's no official confirmation, based on Jason Wei's earlier posts, he's only 27 or 28 this year.)

After graduating with a bachelor's in computer science from Dartmouth College in 2020, Jason didn't follow the conventional path of pursuing a PhD, but instead chose to join Google Brain directly.

In 2022, at just 23, he earned the title of Google "Research Scientist."

(Possibly one of the youngest research scientists in Google's history)

That same year, as first author, he published the foundational Chain-of-Thought (CoT) paper.

At the same time, he was also first author of FLAN, a classic work in instruction fine-tuning.

One could say that the reason today's large models "think step by step" traces back to him.

Later, he joined OpenAI in 2023, deeply participating in the core R&D of the first-generation reasoning model o1 and Deep Research, and also single-handedly built SimpleQA, BrowseComp, and HealthBench—a health evaluation benchmark developed jointly with 262 doctors.

(At the 2024 o1 "fan meetup," sitting front row center was Jason Wei, and next to him was another key o1 figure, Zhao Shengjia, who is still at Meta.)

As of now, Jason Wei's Google Scholar citations have exceeded 160,000.

After being poached by Zuck with a sky-high offer last July, he entered TBD Lab under the Superintelligence Lab, led the development of the Muse Spark series of models, and pushed large models into frontier medical research.

Just one month before leaving, he shared an experiment in which his team used Muse for medical research. Medical researchers could call on a coding tool powered by Muse Spark 1.3 and Muse Wonder to analyze about 5 million de-identified electronic medical records.

Doctors were responsible for asking questions, and the model was responsible for writing code and running analyses. In just four weeks, the team produced ten research manuscripts.

Jason had a key observation about this work: although the model lets doctors verify a hypothesis faster and gives them a chance to try a few more directions, medical record data is full of noise—which variables are worth looking at, and what a set of results actually means, ultimately still depends on professionals to decide.

And Mercor, which he's joining this time, happens to be a company dead set on "high-barrier professional data."

What kind of company is Mercor?

Mercor was founded in January 2023 and is headquartered in San Francisco.

The three founders, Brendan Foody, Adarsh Hiremath, and Surya Midha, are old acquaintances from their high school debate team, and in the early days of the startup the three worked out of dorms at Harvard and Georgetown University.

Today, this young company has not only recruited top researchers like Jason Wei and Edward Hu, but also assembled a large network of doctors, lawyers, investment banking analysts, and senior engineers, building a vast network of industry experts.

According to its official website, Mercor's main business is AI training data and evaluation services. It organizes experts from various fields to break down real work into highly difficult training tasks and develops strict grading rubrics, helping AI labs and enterprises train models and test their ability to solve complex professional problems.

In addition, Mercor also helps enterprises monetize their own business data and deploy AI within their daily workflows.

After Felicis led both the Series B and Series C rounds, just eight days before Jason announced his move (November 10), Mercor had just announced the completion of a $750 million Series D round at a valuation of up to $20 billion.

Investors include not only top venture firms like Benchmark, General Catalyst, and Sapphire Ventures, but also industry giants such as 英伟达, Salesforce, and Zoom.

Reportedly, the core logic behind capital's bet on Mercor is: general internet corpora are nearly exhausted, and only the "tacit judgment" of seasoned experts (such as the contract risk points a senior lawyer can spot at a glance) can truly teach AI to do work.

Currently, Mercor's clients have expanded from foundation model labs to vertical-sector giants such as legal AI company Harvey and financial software company Ramp.

As Jason previously lamented on X, writing a high-quality question for a top benchmark today may already have an impact no less than publishing a paper in Nature or Science.

As more and more enterprises begin training their own models, companies like Mercor are trying to define the next frontier of AI evolution by controlling "high-quality data and evaluation standards."

One more thing

That said, with Jason Wei's departure, "stay at Meta for at most one year" seems to have become a curse.

Earlier, former Apple foundation model head Pang Ruoming gave up a pay package worth over $200 million and decisively jumped to OpenAI after just 7 months;

Even earlier, OpenAI researchers Ethan Knight and Avi Verma, who also received sky-high offers, stayed only a few short weeks before quickly "boomeranging" back to their old employer;

On this, many voices in the industry joke that big companies' sky-high poaching usually comes with a "1-year cliff."

The first large signing bonus and stock often unlock and vest together at the one-year mark. Once past that time point, it's equivalent to a "player option" automatically activating—whether the person stays depends entirely on the lab's real combat strength and technical vision.

Of course, AI leaders come and go, but Alexander Wang is iron-clad.

Meta's Chief AI Officer Alexander Wang, born in 1997, remains steadfast and still highly valued by Zuckerberg.

  • 黄仁勋 resigns, 英伟达 changes leaders for the first time in 33 years, market cap evaporates over $3 trillion 2026-11-13
  • Anthropic launches IPO roadshow, valuation has shrunk by $600 billion 2026-11-12
  • 马斯克 Grok leaks its own secrets! SpaceX and Tesla salaries and performance all exposed 2026-11-12
  • 英伟达 robot open source collides with Fable 5.2 release! Jim Fan: not winter but summer 2026-11-10

Related Reading

Meta Agent goes out of control and leaks secrets, Zuck urgently sounds the top-level alarm

The one who can make Zuck stumble is still "AI."

Article title: Chain-of-Thought first author Jason Wei leaves! Why did they run off after just one year at Meta?

Body:

Liu Lu was also poached by Meta! A South China University of Technology alumnus, created the 4o Ghibli hit

Also poached at the same time was Allan Jabri, a member of the 4o core architecture team

Little Zuckerberg in conversation with 黄仁勋: AI models won't be open-sourced, I'm going to curse out loud

If I had known success would take this long...

Meta AI releases Data2vec, a unified model for images, audio, and text, with CV accuracy surpassing MAE

1.4 billion yuan couldn't keep him! Pang Ruoming leaves Meta to join OpenAI

Little Zuckerberg changes his tune on open source, Meta stock price surges 12%

Meta revenue and net income exceed expectations

Values written like {{NAME_ab12cd34}} are placeholders for real values. Use them exactly as written, in replies and in tool calls alike: they are filled in with the real values before anything runs. Don't ask to verify them, and don't replace them with examples.

The translation is machine-generated and individual terms may be off. Nothing is rewritten or cut; the original is authoritative.

Published by 量子位
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗