AlphaGo at 10: How a Go-playing AI reshaped AI research
DeepMind reflects on a decade of AlphaGo's influence, from games to scientific discovery and the road to AGI.
AI & ML
Frontier models, research results, agents, and developer tooling.
Hugging Face Uses Claude to Generate CUDA Kernels, Teach Open Models
The project turns Claude's GPU code into teaching material for open models.
Google Research Joins Nationwide Randomized AI Study in Virtual Care
A new collaboration will test generative AI in real-world virtual care settings through a nationwide randomized study.
New Benchmark Measures Arabic LLMs on Emirati Dialect
A TII-authored blog post proposes a dedicated evaluation set for Emirati Arabic, addressing a blind spot in Arabic NLP benchmarks.
Google Research Teaches LLMs to Reason Like Bayesians
A new Google Research initiative aims to make large language models update their beliefs probabilistically instead of committing to a single reasoning path.
Gemini 3 Deep Think Targets Science and Engineering
Google DeepMind says its most specialized reasoning mode has been updated to tackle modern challenges in science, research and engineering.
Transformers.js v4 Released on NPM
Hugging Face's JavaScript transformers library reaches a new major version with v4 now available via NPM.
DeepMind's Nano Banana 2 promises Pro-grade image generation at Flash speed
Google DeepMind's latest image model brings advanced world knowledge and production-ready specs to a faster tier.
What 16 Open-Source RL Libraries Say About Keeping Tokens Flowing
A Hugging Face review finds that asynchronous token generation is a key design lesson across open-source reinforcement-learning libraries.
Hugging Face Reworks Diffusion Pipelines as Composable Building Blocks
The update makes diffusion pipelines easier to customize and maintain by breaking them into modular components.
Hugging Face Brings GGML and llama.cpp In-House for Local AI
Hugging Face says the team behind llama.cpp is joining the company to help keep on-device AI moving forward.
Google Research introduces WAXAL for African language speech tech
Google Research has released WAXAL, an open resource intended to support speech technology for African languages.
Gemini 3.1 Flash-Lite: DeepMind's Fastest, Cheapest Model Yet
Google DeepMind positions Gemini 3.1 Flash-Lite as its fastest and most cost-efficient Gemini 3 model, targeting high-scale AI workloads.
New Benchmark and Framework Diagnose Enterprise AI Agent Failures
IBM and UC Berkeley introduce IT-Bench and MAST to systematically identify why enterprise agents fail.
Hugging Face Releases LeRobot v0.5.0 to Scale Every Dimension
The new LeRobot release is framed as a broad scaling update, but the announcement itself offers no concrete details on what changed.
Gemini 3.1 Pro targets complex reasoning tasks
Google DeepMind's latest model is built for problems that require deeper analysis rather than quick answers.
Why Rational AI Agents Should Not Have Goals
A new essay in The Gradient argues that AI alignment should focus on virtue-ethical practices rather than goal-directed behavior.
DeepMind Launches AI-Powered Science and Education Initiative in India
The move extends DeepMind's effort to apply AI to scientific discovery, now targeting India's research and education sectors.