Google Research Applies Differential Privacy to AI Chatbot Insights
A Google Research framework aims to make it possible to analyze AI chatbot usage patterns without exposing individual user interactions.
AI & ML
Frontier models, research results, agents, and developer tooling.
Diffusers Adds FLUX-2 Support
Hugging Face's Diffusers library now supports FLUX-2, giving developers a unified way to run the model.
Product Evals in Three Simple Steps: Label, Align, Run
A practical framework for building LLM evaluation pipelines without overcomplicating them.
AlphaFold maps structure of protein linked to heart disease
DeepMind's AI system has produced a structural model of a protein implicated in heart disease, offering a new tool for understanding its biological role.
Google DeepMind Unveils Nano Banana Pro, a Gemini 3 Pro Image Model
The release points to a push for developers to build applications on DeepMind's latest image-generation foundation.
Simple AI Model Forecasts EV Charging Port Availability
Google Research shows how a lightweight AI model can help drivers find an available EV charging port and reduce range anxiety.
NeuralGCM: AI-Powered Model Targets Long-Range Precipitation
Google Research's NeuralGCM applies AI to improve simulation of long-range global precipitation, a step toward better climate forecasting.
How Tavily Built a State-of-the-Art Deep Research System
A new blog post details the architecture and methods behind a deep research system that outperforms existing benchmarks.
Continuous Batching: A Primer on Efficient LLM Inference
A Hugging Face explainer walks through how continuous batching improves throughput and latency when serving large language models.
RapidFire AI Claims 20x Faster TRL Fine-Tuning
Hugging Face spotlights an integration that promises to cut TRL fine-tuning time by an order of magnitude.
Google DeepMind's FACTS benchmark targets LLM factuality
A new evaluation suite from Google DeepMind aims to measure how accurately large language models ground their outputs in provided source documents.
Hugging Face Adds Multilingual and Long-Form Tracks to Open ASR Leaderboard
The leaderboard update aims to make automatic speech recognition benchmarks more realistic and inclusive.
Google's New Benchmark Measures Auditory Intelligence Beyond Speech Recognition
A new benchmark from Google Research aims to evaluate AI systems on complex auditory reasoning, not just transcription.
Hugging Face Launches Swift Client for Its Platform
The new swift-huggingface library is described as a comprehensive Swift client for Hugging Face's services.
DeepMind and DOE launch Genesis to speed science with AI
A new public-private partnership aims to apply advanced AI to U.S. energy and scientific research challenges.
Hugging Face Explains Open Responses, an Open Format for AI Outputs
A new explainer from Hugging Face lays out how Open Responses aims to standardise and open up the way AI systems return responses.
Google Research’s SpeciesNet Uses Machine Learning to Identify Wildlife
A new model from Google Research applies computer vision to wildlife monitoring, pointing toward faster analysis of ecological imagery.
Bringing VLA robotics models to embedded platforms
A new Hugging Face and NXP collaboration details how to record datasets, fine-tune vision-language-action models, and optimize them for on-device deployment.