Meta’s chief AI scientist, Yann LeCun (Julien De Rosa/Getty Images)

Tension emerges between Meta’s AI teams

Discontent between Meta’s AI research teams is growing, according to a report by The Information, at a critical time for Meta’s effort to get back into the AI race.

Jon Keegan

10/2/25 12:06PM

Long before Mark Zuckerberg pivoted Meta away from its quest for virtual reality to “bring personal superintelligence to everyone,” Meta’s research group FAIR was a powerhouse of important AI research, in addition to working on earlier iterations of the company’s Llama AI models. The group is headed by OG AI legend Yann LeCun, who is a pioneer in neural networks and computer vision.

The FAIR group operates like an academic research lab within Meta, publishing research papers and sharing work with the wider community. But since Meta’s stumble with its Llama 4 AI model, Zuckerberg went on an unprecedented hiring spree of AI all-stars, poaching top researchers from Meta’s competitors to build out a new “Superintelligence team.”

Now, The Information is reporting that there are new tensions between the AI groups, which could have huge ramifications for Meta’s AI research.

Per the report, several changes to how FAIR operates are causing friction. A new layer of review has been imposed on FAIR’s research before publication, and the company has been pressuring the group to direct its work more toward Meta products rather than the wider AI research community.

Adding to this, LeCun appeared to be sidelined when 28-year-old college dropout Alexandr Wang was hired from Scale AI and named chief AI officer. Later, when Meta recruited Shengjia Zhao, the cocreator of ChatGPT, away from OpenAI, Zhao was named “chief scientist of Meta Superintelligence Labs.” Reportedly, the title was given to Zhao to appease him after he threatened to return to OpenAI, going so far as to sign HR paperwork with his former employer.

According to two Information sources, LeCun has discussed with colleagues the possibility of quitting the role. And the nine-figure salaries offered to the Superintelligence team recruits aren’t helping. The rocky start to Meta Superintelligence Labs raises questions about how quickly the new strategy can get Meta back into the AI race.

Bad vibes

Last week, Meta announced “Vibes,” a feed of AI-generated videos that appears in the Meta AI app. But the announcement was quickly dwarfed by the attention on OpenAI’s invite-only Sora app, featuring short videos generated from its new Sora 2 video generation model, which appears to set a new, high standard for the quality of such technology.

The buzz around Sora is real: it’s now No. 3 on the iOS App Store free apps leaderboard despite being invitation-only, while Meta AI sits at No. 97.

Tesla deliveries drop for second straight year; stock on pace to match longest losing streak ever

BYD outsold Tesla in battery electric vehicles for the first time in 2025.

Rani Molla5h

Rani Molla12/31/25

Rather than fully cracking down on scam ads, Meta worked to make them harder to find

In its latest piece on Meta’s scam ads, Reuters found that the social media giant didn’t just remove fraudulent ads from its platforms — it also worked to make them harder for governments and journalists to find.

Fearing that Japanese regulators would require universal advertiser verification — a measure Meta estimated would cost roughly $2 billion to implement and potentially reduce its revenue by nearly 5% — the company took steps to make scam ads less “discoverable” to “regulators, investigators and journalists,” according to internal documents reviewed by Reuters.

“So successful was the search-result cleanup that Meta, the documents show, added the tactic to a ‘general global playbook’ it has deployed against regulatory scrutiny in other markets, including the United States, Europe, India, Australia, Brazil and Thailand,” Reuters wrote.

Previous Reuters reporting found Meta internally projected that about 10% of its 2024 revenue would come from ads tied to scams and banned goods, though the company later said that estimate was overly broad. Reuters also reported the rate was double in China.

Meta created ‘playbook’ to fend off pressure to crack down on scammers, documents show

Rani Molla12/31/25

Michael Burry, the “Big Short” investor who called Tesla “ridiculously overvalued,” is not currently shorting Tesla

Earlier this month, “The Big Short” investor Michael Burry said Tesla has been “ridiculously overvalued” for “a good long time” — and reiterated that message in a post on X on Tuesday. But the once prominent Tesla short seller isn’t currently betting against the stock.

I am not short
— Cassandra Unchained (@michaeljburry) December 31, 2025

Asked directly whether he would short Tesla now, Burry replied simply: “I am not short.”

Tesla is expected to report a double-digit decline in fourth-quarter deliveries this week.

Rani Molla12/30/25

SoftBank becomes OpenAI’s biggest backer after fully funding $40 billion investment

SoftBank has fully funded its $40 billion investment in OpenAI, overtaking Microsoft as the company’s largest financial backer, CNBC reports. The deal was contingent on OpenAI transitioning to a for-profit public benefit corporation, which it did in September.

However, longtime partner Microsoft retains substantial influence over OpenAI with its roughly $13 billion investment, which translates to a stake worth about 27% of the startup’s valuation — which has been cited as high as $830 billion — as well as exclusive cloud and commercial licensing rights tied to Azure.

Softbank has fully funded $40 billion investment in OpenAI, sources tell CNBC

Rani Molla12/30/25

Tesla-compiled estimates show Q4 deliveries expected to fall 15% from last year

A Tesla-compiled average of analyst estimates pegs fourth-quarter deliveries at 422,850, which would mark a 15% slump from the 495,570 the company delivered in the same quarter last year, if realized. The full-year estimate of 1.6 million vehicles would represent an 8% decline from 2024 and the second annual decline for the EV company. The estimates are notably lower than the consensus estimates compiled by Bloomberg and FactSet, which have been declining over the past month.

The market-implied odds derived from event contracts show that most traders think Tesla deliveries will be more than 410,000 but less than 420,000 in the quarter ending December.

(Event contracts are offered through Robinhood Derivatives, LLC — probabilities referenced or sourced from KalshiEx LLC or ForecastEx LLC.)

While Tesla typically shares its compilation of analyst estimates with institutional investors, this is the first time the company has shared those numbers on its own website. Tesla’s numbers include estimates from Daiwa, DB, Wedbush, OpCo, Canaccord, Baird, Wolfe, Exane, GS, RBC, Evercore ISI, Barclays, Wells Fargo, Morgan Stanley, UBS, Jefferies, Needham & Co., HSBC, Cantor Fitzgerald, and William Blair.

Actual numbers are expected Friday.