Meta is an American multinational technology conglomerate. It owns Facebook, Instagram, and WhatsApp, among other products and services.
Sooth Labs, an AI prediction laboratory founded by former Meta executives and a CMU professor team, is completing a $50 million funding round at a $335 million valuation, led by Felicis Ventures. Yann LeCun and Jeff Dean are participating as individual investors, and Meta CTO Andrew Bosworth serves as an advisor. The company specializes in multimodal AI event probability prediction, serving institutions in finance, defense, insurance, and other sectors, and has already provided probability predictions for events such as the WHO pandemic and Anthropic's IPO.
Coinversation
Jeff
According to Bloomberg, Sooth Labs, an AI lab founded by former Meta employees, is raising approximately $50 million in funding, led by Felicis Ventures, with a post-money valuation of about $335 million. The company plans to develop AI models that predict the probability of specific geopolitical and market events, offering these capabilities to enterprises. Sooth Labs has also received support from Yann LeCun and Google’s Chief Scientist Jeff Dean, and Meta’s CTO Andrew Bosworth serves as an advisor.
Coinversation
Jeff
Odaily News Meta has found a new source of training data for its AI models: its own employees. The company plans to use data collected from its employees' mouse movements and keystrokes to build more powerful and efficient artificial intelligence. A company spokesperson stated: "If we want to build agents that help people accomplish daily tasks on computers, our models need real-world examples to understand how people actually use computers—such as mouse movements, button clicks, and operating dropdown menus. To this end, we will launch an internal tool to capture these operational inputs on specific applications to help train our models. We have implemented corresponding safeguards to protect sensitive content, and this data will not be used for any other purposes." (TechCrunch)
button
According to The Block, Texas man Robert Dunlap was sentenced to 23 years in federal prison for a cryptocurrency fraud scheme exceeding $20 million and ordered to pay restitution to nearly 1,000 victims. Prosecutors alleged that Dunlap operated cryptocurrency projects and sold Meta-1 Coin, falsely claiming the token was backed by $44 billion in gold and approximately $1 billion in artworks—including pieces by Pablo Picasso, Vincent van Gogh, and Salvador Dalí—and that the assets had been audited. Last year, a jury found him guilty of mail fraud.
Block
Pablo
The Block
Vincent
According to the South China Morning Post, Interconnects AI, a U.S.-based AI tracking firm, released a report stating that as of March 2026, Alibaba Cloud’s Qwen series models accounted for over 50% of global open-source model downloads, with a cumulative total of 942.1 million downloads—far surpassing competitors such as Meta’s Llama and DeepSeek. In February alone, Qwen downloads reached 153.6 million—exceeding the combined total downloads of the next eight major vendors. The report notes that Qwen’s dominant position stems from the exceptional popularity of its smaller-parameter variants (under 10 billion parameters), which enable developers to customize and deploy models freely at low cost. Since the launch of Qwen 2.5 in September 2024, Chinese models have begun outpacing mainstream U.S. open-source models like Llama; the release of Qwen 3.5 in February this year further solidified its lead. Meanwhile, open-source strategy has become a critical battleground in the U.S.-China AI competition. Meta has abandoned its open-source approach this year, instead launching the closed-source flagship model Muse Spark. Similarly, Chinese vendors including Alibaba Cloud and Zhipu AI have shifted some of their latest models to closed-source to expand direct commercialization channels.
based
Llama
Meanwhile
Spark
Spark