Latest publications
AI platform LUMI-lab sets record for gene-editing efficiency
Built on foundation models, the LUMI-lab platform autonomously decoded the mRNA delivery code, achieving 20.3% accuracy in lung gene editing.
The limits of scaling: why more AI agents do not guarantee results
A new study has identified a critical barrier in the development of multi-agent systems. Information redundancy is becoming the main obstacle to LLM efficiency.
DeepSeek, together with Tsinghua University and Peking University, improves AI agents' reasoning
DeepSeek has published a new study co-authored with China's leading universities, focused on optimizing reasoning in agentic language models.
Musk announces Grok's return
Elon Musk confirmed: the AI assistant Grok is returning today. What should we expect from the updated version?
Agentic AI is rewriting the rules of biomedical research
How the shift from LLM capabilities to autonomous agents is changing the very architecture of team science in biomedicine — and what it means for researchers.
DistDF: a new method for time series forecasting via distribution alignment
Researchers proposed replacing the standard MSE metric with the Wasserstein distance to improve neural network accuracy in analyzing complex data. The new DistDF method was presented at ICLR 2026.
Google's AI outperformed Olympiad champions in the FirstProof math test
Google DeepMind brought AI closer to the level of the world's best mathematicians, with results on extremely difficult problems beyond the level of Olympiad medalists.
Ctrl-World: Joint Tsinghua-Stanford project outperforms Google in robotics
The new world model Ctrl-World sets records in embodied AI, surpassing Google and Nvidia systems in robot control accuracy.
Leading researcher turned down Meta’s $200 million offer for OpenAI
Pan Zhomin unexpectedly left Meta, rejecting a $200 million compensation package and moving to OpenAI. A high-profile talent move in AI.
Can Neural Networks Really Reason? A Study of Structural Errors in LLM Logic
A new scientific analysis questions AI's capacity for logical reasoning, identifying critical vulnerabilities in the inference structure of popular models.
Alibaba's Qwen3.5: three new open models for consumer GPUs
Alibaba has open-sourced three new Qwen3.5 models that run on standard consumer graphics cards — without expensive server hardware.
AI rewrote Next.js in a week: a revolution in development costs
Developers recreated the complex Next.js framework in just seven days, spending $1,100 on AI tokens. This case is changing perceptions of the speed and cost of building software.
Isomorphic Labs develops a next-generation AI engine for drug design
The DeepMind spinoff unveiled a closed platform described as a proprietary AlphaFold 4 engine for pharmaceuticals.
Alibaba Cloud launches support for Qwen 3.5 and GLM-5 for coding
The Alibaba Cloud platform has added four new open models: Qwen 3.5, GLM-5, MiniMax M2.5, and Kimi K2.5, focused on programming and high performance.
EmotionThinker: LLMs learn to explain emotions in speech
A new approach allows speech models not only to classify emotions, but also to explain their conclusions about them.
Unitree unveiled a new robot dog right after the New Year show
China's Unitree Robotics announced a new robot dog model following its high-profile appearance on the country's main TV show.
AI wipes $31 billion off IBM over COBOL coding
The ability of neural networks to work with legacy code triggered IBM's biggest market-cap drop in the past 26 years. Investors fear for the consulting giant's future.
Eight years in the making: China launches WAICA, a world-class academic platform
After eight years of development, China is presenting WAICA, its own world-class conference in artificial intelligence. A new meeting point for the global AI community.
DeepMind: More agents, worse results?
DeepMind researchers have called into question the idea that scaling multi-agent systems leads to higher efficiency. The ceiling may already have been reached.
Qianxun Intelligence raised nearly 2 billion yuan and became a unicorn
Chinese AI startup Qianxun Intelligence closed a major funding round with participation from leading venture capital funds. The company’s valuation exceeded 10 billion yuan.
FARS: 100 research papers in 228 hours and 11.4 billion tokens
The autonomous AI agent FARS set a record by writing 100 research papers in less than ten days of continuous work.
From aesthetics to practice: MIT makes generative 3D design fit for manufacturing
MIT researchers introduced a differentiable physics optimization method that turns beautiful 3D models into real objects ready for industrial production.
Alibaba Cloud updates Coding Plan: new Qianwen and GLM models
Alibaba Cloud's Coding Plan platform has gained support for the Qianwen 3.5 and GLM-4.7 language models, intensifying competition in the market for AI developer tools.
OpenAI quietly rewrote its mission: “benefiting all of humanity” and safety disappeared
OpenAI quietly changed the wording of its mission, removing references to “benefiting all of humanity” and safety. What is behind this shift, and why it worries the industry.
The era of solo models is over: University of Washington unveils MoCo to unite AI models
The new open-source framework allows several language models to work as a team, solving tasks beyond the reach of even the most powerful standalone systems.
OpenClaw project creator joins OpenAI: Sam Altman hires talent
OpenAI is strengthening its engineering team: Sam Altman has hired the creator of OpenClaw. The new addition to the team underscores the company's leadership ambitions in the AI arms race.
Gemini 3 Deep Think now creates ready-to-print 3D models
Google’s new neural network has learned to generate STL files. Now a text description is enough to create a physical object without complex modeling.
TI-DPO: A New Method for AI Alignment by Evaluating Token Importance
Researchers introduced TI-DPO, an algorithm that fixes the main problem of standard DPO through detailed analysis of each word during neural network training.
Ant Group open-sources multimodal model Ming-Flash-Omni 2.0
Ant Group unveiled Ming-Flash-Omni 2.0. The new open-source model challenges Gemini 2.5 Pro, offering advanced text, image, and voice capabilities in a single system.
Shanghai scientists uncover the "dark side" of social interactions among AI agents
Researchers in Shanghai modeled destructive behavior by AI agents in autonomous networks, uncovering hidden threats to future digital ecosystems.