LeCun's New Venture! The New Company is Valued at 24.7 Billion, but the CEO Is Not Him!

In the field of artificial intelligence, renowned scientist Yann LeCun is about to launch a new company, which has attracted widespread attention. The new company, named Advanced Machine Intelligence Labs (AMI Labs), is planned to be officially established in January next year, with a target valuation of 3 billion euros (approximately 24.7 billion Chinese yuan). During his last days at Meta, driven by his passion for "world models," LeCun has chosen to take the open-source path and maintain cooperation with his former employer, Meta.

AMI Labs will focus on LeCun's long-standing advocacy for "world model" research. LeCun believes that current mainstream autoregressive large language models (LLMs) have logical flaws, as they can only predict text based on statistical probabilities, without truly understanding the laws of the physical world. He has repeatedly warned that continuing down this path would be a "dead end" toward artificial general intelligence (AGI). In contrast, AMI Labs plans to use a new technology called JEPA (Joint Embedding Prediction Architecture), which aims to understand and predict the world through "abstraction" and "planning," striving to achieve a higher level of intelligence.

Notably, LeCun himself will not serve as CEO of AMI Labs. The role will be taken over by his former subordinate — Alexandre LeBrun, founder of the medical AI company Nabla. LeBrun has a deep connection with LeCun, having worked together with him for three years at Meta's FAIR lab. Under LeCun's leadership, LeBrun gained rich experience. As he transitions into a commercial operations-focused CEO, it signals that AMI Labs will form a "dual-core structure" where LeCun focuses on internal research while LeBrun handles external business operations.

Despite being 65 years old and facing the choice of retirement, LeCun still chooses to move forward. In a recent interview, he stated that he still has a mission to "enhance human intelligence." The establishment of AMI Labs represents his latest adventure in the field of artificial intelligence, aiming to abandon the shortcuts of current large models and delve deeper into the real challenges of understanding the physical world.

Apple Launches New Multimodal AI Model UniGen 1.5, Integrating Image Understanding, Generation, and Editing in One

Apple introduces the multimodal AI model UniGen 1.5, integrating three major functions of image understanding, generation, and editing within a unified framework, significantly improving efficiency. The model leverages its image understanding capabilities to optimize generation results, achieving technological breakthroughs.

Gemini 3 Flash Launches: Free, Fast, and Intelligence Surpasses Pro - Google AI Fully Enters the Zero-Latency Era

Google launches its new lightweight model, Gemini 3 Flash, with response speed three times faster than the previous generation, nearly zero latency, and surpassing the flagship Gemini 3 Pro in multiple high-difficulty benchmark tests. The model is freely available worldwide and has been integrated into tools such as the Gemini App and AI Studio, breaking through performance and cost boundaries.

New Benchmark for AI Scientific Research: FrontierScience Evaluates Model Reasoning Ability

AI models have made significant progress in evaluating scientific reasoning abilities, performing excellently in international mathematics and informatics olympiads. With the development of advanced models such as GPT-5, AI is effectively accelerating real scientific research processes, demonstrating strong capabilities in hypothesis generation, testing, refinement, and cross-domain integration.

Ministry of Education Issues Notice: Primary and Secondary School Examinations Will Enter the AI Era, with Full Process Pilot Testing of Artificial Intelligence in Question Setting, Test Paper Compilation, Grading, and Analysis

The Ministry of Education issued a notice to use artificial intelligence technology to improve the management of primary and secondary school examinations, piloting the application of AI in areas such as question setting, test paper compilation, and grading to enhance the professionalism and scientific nature of examinations.

Japanese Scientists Launch Sui Programming Language, Claiming It Can Write Code with 100% Accuracy Using LLM

Japanese data scientist Takahito Honda launched the open-source programming language Sui, aimed at solving the accuracy issues of code generated by large language models, claiming it can achieve 100% accuracy. Its design concept is inspired by Japanese aesthetics "Wabi-Sabi", emphasizing refinement and elimination of redundancy, with core principles including ensuring zero syntax error rate and using numbers as variables.