Discover Metrax, a JAX-native, high-performance metrics library: jittable, vectorized evaluation for classification, ranking@K, NLP, speech, and vision.

Metrax is a JAX-native, high-performance, open-source evaluation metrics library developed by Google. Metrax offers standard evaluation metric implementations using JAX, allowing you to focus on your model and training. 

Resources: 

Documentation → https://goo.gle/3LFkZIR 
GitHub repo → https://goo.gle/3WMu05o 
Example notebook → https://goo.gle/3JYO0yN 

Subscribe to Google for Developers → https://goo.gle/developers 

Speaker: Yufeng Guo 
Products Mentioned: Google AI

Introduction to Metrax: Evaluation metrics for JAX

A one-day experiment where engineers with an AI assistant (Claude) got a robot dog to fetch — cutting setup time, generating controllers, and lowering barriers to robotics.

We asked two teams of Anthropic researchers to program a robot dog. Neither team had any robotics expertise—but we let only one team use Claude.

In the past, we’ve run simulated studies where Claude trained a robot dog. These helped us assess how Claude might contribute to AI research and development. Project Fetch was us trying something similar in practice.

This project suggests that we’re not far from a world where frontier AI models can interact with previously-unknown pieces of hardware, even with non-experts at the helm.

For more, read on: https://www.anthropic.com/research/project-fetch-robot-dog 

0:00 Introduction: Why robotics?
0:30 The experiment
1:02 Phase 1: Fetch, manually
1:51 Phase 2: Fetch, programmatically
5:08 Phase 3: Fetch, autonomously
6:24 Results

Can AI program a robot dog?

Discover how Nebius built a purpose‑built AI cloud for scalable, reliable, high‑performance training, inference and data processing—designed for multi‑node models and sensible economics.

When it comes to future-proofing AI deployments, you need reliable underlying AI infrastructure that is purpose-built for high performant and scalable inference.

Explore how Nebius built and deployed its high-reliance AI cloud, powered by NVIDIA.

Using managed Kubernetes with auto-scaling, Nebius optimizes its AI cloud to deliver multi-node training and inference of frontier models and AI applications for startups and enterprises. Nebius, an ecosystem partner for Dynamo, is enabling AI inference at scale with NVIDIA infrastructure. 

Learn how NVIDIA Dynamo and Kubernetes help scale high-performance AI inference in our new Think SMART blog: https://nvda.ws/4qPgD26 

Explore Nebius and NVIDIA collaboration: https://nvda.ws/4hMGjbz 

Watch Nebius leadership session from GTC: https://nvda.ws/3XfIZEZ

Scaling AI Inference Performance in the Cloud with Nebius

Learn how NetApp AFX and the NetApp AI Data Engine with NVIDIA transform data storage, embeddings, and near-data compute to power agentic, multimodal AI at scale.

Our CEO Jensen Huang and NetApp CEO George Kurian announce a historic partnership to deliver an enterprise-grade data platform purpose-built for AI reasoning.

Integrated with NVIDIA accelerated computing and software, learn how the NetApp AI Data Engine, which leverages the NVIDIA AI Data Platform reference design, and the NetApp AFX streamline the entire data pipeline for any production environment.

Learn more: https://www.netapp.com/

NetApp and NVIDIA: Revolutionary AI Products at a Revolutionary Time

Deep dive into Kaggle Game Arena insights, Gemini-powered Firebase Studio workflows, MCP support, and Dart & Flutter updates — practical guidance for devs and educators.

Google Developer News has the latest developer updates for Google developers for August 2025. From covering the Kaggle Game Arena with Google DeepMind to exploring the newest AI-enhanced tools in Firebase Studio, this episode will help you deploy faster builds with AI, while watching leading AI models battle it out in head-to-head competition. 

Chapters:
0:00 - Introduction 
0:11 - Kaggle Game Arena: Testing the limits of AI reasoning with Google DeepMind
0:45 - Gemini & Firebase Studio: New AI-powered workflows
1:01 - Foundational support for model context protocol
1:10 - Gemini CLI integration with Firebase Studio
1:18 - AI Optimized templates: faster app building
1:31 - Firebase backend integration with Gemini
1:38 - Safety net experimentation with Workspace forking
1:51 - Collaborative features in Firebase Studio
2:00 - Enhancing prompts with Gemini
2:11 - Dart 3.9 & Flutter 3.35: performance & visualization updates
2:31 - Dart and Flutter MCP server.

Watch more Google Developer News → https://goo.gle/4e8Rysd  
Subscribe to Google for Developers → https://goo.gle/developers 

#GoogleDeveloperNews 

Speaker: Ethan Hsu
Products Mentioned: Google AI, Gemma,  Generative AI,  Gemini 1.5 Flash,  Colab,

Kaggle Game Arena, AI tools in Firebase Studio, and more! - Google Developer News August 2025

A SIGGRAPH 2025 report on NVIDIA's special address: how AI, scalable simulation, differentiable physics, and digital twins are launching 'Graphics 3.0' and reshaping graphics, robotics, and content creation.

AI is reshaping what’s possible in graphics. ✨

Hear NVIDIA AI research leaders Sanja Fidler, Aaron Lefohn, and Ming-Yu reveal our latest breakthroughs in computer graphics and physical AI at #SIGGRAPH2025.

SIGGRAPH 2025 Special Address

Ally Solos Glasses pair Ally AI with ultra-light Solos frames to deliver camera-enabled, voice-first smartglasses for people who are blind or have low vision. Learn how they work and pre-order details.

Pre-order the Ally Solos Glasses now, save up to 40% from the original retail price and get 1 year of Ally Pro subscription (worth $200) for free.
Go to: ally.me/glasses

Experience everyday independence with Ally Solos Glasses. Lightweight smartglasses with a camera, built for people who are blind or have low vision.

Powered by Ally, the most accessible AI assistant, these glasses let you interact with your environment just by talking. Read menus, recognize surroundings, objects, people, and much more.

Designed to support independence for blind and low-vision users, Ally Solos Glasses blend intuitive technology with everyday comfort.

With Solos’ award-winning smartglasses and Ally’s accessible AI, it's easy to access information, recognize your surroundings, and move confidently through your day.

For more information go to: ally.me/glasses

Ally Solos Glasses - The most accessible AI assistant, now wearable.

Discover how NVIDIA’s GTC Paris 2025 keynote reveals Europe’s AI-driven fourth industrial revolution, featuring Agentic AI, the Blackwell platform, and transformative industrial innovation.

From sovereign AI to Blackwell, see how Europe is building the future. Watch the GTC Paris 2025 wrap-up with NVIDIA.

Catch up with recommended sessions on NVIDIA On-Demand: https://nvda.ws/46c2tAn

#GTCParis #AI

Highlights from NVIDIA Keynote and GTC Paris 2025

Explore the innovative Poetry Camera, a unique fusion of AI, photography, and poetry that transforms everyday scenes into poetic masterpieces, inviting mindfulness and creative expression.

Poetry Camera is a camera that makes poems, created with Claude. This project is a labor of love by a self-funded team, assembled by hand in New York.

Poetry Camera is not fully photography, is not fully poetry, but a weird new third thing enabled by new technology. Learn more about their project: https://poetry.camera

A camera that makes poems

Discover how DeepAgent's fully autonomous, self-learning AI agent transforms productivity by connecting to APIs and automating complex workflows without coding. Learn why this is a game-changer for freelancers and enterprises alike.

Try DeepAgent here → https://deepagent.abacus.ai/
DeepAgent just received a powerful upgrade that makes it fully autonomous, capable of finding tools online, learning how to use them instantly, and completing complex tasks without any setup or coding. It now connects to services like Twitter, Slack, Notion, Salesforce, and Jira, automating workflows in real time with self-learning capabilities. With support for MCP servers, OpenAPI schemas, and secure credential handling, DeepAgent is quickly becoming one of the most advanced AI agents on the market.

🧠 What’s Inside:
DeepAgent Autonomous AI Agent
https://deepagent.abacus.ai/

⚙ What You’ll See:
DeepAgent connects to tools like Twitter, Notion, Slack, Salesforce, and Jira, then learns their APIs on the fly to automate tasks without coding.
It builds interactive mindmaps, formats reports, fixes access issues, and creates workflows entirely on its own—all from a single prompt.
With support for MCP servers, OpenAPI parsing, real-time orchestration, and secure credential handling, this $10 AI agent is now smarter than ever.

🚨 Why It Matters:
DeepAgent shows how far autonomous agents have come—handling complex, multi-platform tasks with speed, precision, and no human intervention. It’s a glimpse into the future of self-learning automation, where AI doesn’t just assist—it takes over.

#deepagent #aiagent #ai

This New AI AGENT Shocked the Market: Fully Autonomous and Self-Learning

Discover how ByteDance's DetailFlow, Alibaba's Qwen3, and USC's chatbot advancements are revolutionizing AI with faster image generation, open-source embeddings, and reduced hallucinations.

Grab your free copy of the AI Income Blueprint here → https://aiskool.io/
ByteDance introduced DetailFlow, a groundbreaking AI that generates images like humans sketch—big shapes first, details later—while running nearly twice as fast as other models. Alibaba launched Qwen3, a powerful open-source AI that outperforms Google and Meta in language and code search benchmarks across over 100 languages. Meanwhile, U.S. researchers trained chatbots to recognize when they don’t know an answer, reducing hallucinations and improving AI reliability in sensitive tasks.

🧠 What’s Inside:
ByteDance DetailFlow
https://github.com/ByteFlow-AI/DetailFlow
Alibaba Qwen3 Embeddings
https://qwenlm.github.io/blog/qwen3-embedding/
USC’s New Chatbot Refusal Training
https://arxiv.org/pdf/2505.13988

⚙ What You’ll See:
ByteDance’s DetailFlow draws like a human—sketching big shapes first, details later—nearly twice as fast as other models.
Alibaba’s Qwen3-Embedding beats Google’s Gemini and Meta’s GritLM across language, code, and multilingual benchmarks.
USC researchers train AI to finally refuse bad prompts, cutting hallucinations with smart math-based data corruption.

🚨 Why It Matters:
These breakthroughs prove AI is evolving fast—getting faster, more honest, and more efficient with fewer resources. From sketch-speed image generation to world-class open embeddings and safer chatbot behavior, it’s clear: the future is already here, and it's open-source.

#ai #china #google

This Open AI from China Just Crushed American Models and Broke Benchmarks

Discover how Xiaomi's open-source AI model MiMo-VL-7B outperforms larger competitors with just 7 billion parameters, bringing advanced vision-language capabilities to consumer hardware.

Xiaomi has released MiMo-VL-7B, a powerful open-source vision-language AI model with only seven billion parameters that outperforms much larger models like Claude Sonnet and Qwen72B. It can analyze images, videos, and text with high accuracy, handle complex reasoning, and run on consumer-level hardware. Trained on over two trillion tokens and refined with reinforcement learning, MiMo-VL-7B marks a major leap in efficient, multimodal AI technology from China.

🧠 What’s Inside:
MiMo-VL-7B – Xiaomi’s 7B Open Source AI Beats Claude and Qwen72B
PAPER: https://github.com/XiaomiMiMo/MiMo-VL/blob/main/MiMo-VL-Technical-Report.pdf
GITHUB PAGE: https://github.com/XiaomiMiMo/MiMo-VL/
MODEL: https://huggingface.co/collections/XiaomiMiMo/mimo-vl-68382ccacc7c2875500cd212

⚙ What You’ll See:
Xiaomi’s new MiMo model handles images, text, video, and reasoning with shocking precision
Trained on over two trillion tokens, it outperforms models ten times larger in benchmarks
Multimodal, memory-efficient, and reinforced with real-time feedback for smarter answers

🚨 Why It Matters:
This changes the rules—AI no longer needs 70B parameters or cloud servers to compete at the top. MiMo-VL-7B proves that small, open-source models can now challenge giants like Claude, Gemini, and GPT-4o.

#ai #opensource

AI News

AI News

Introduction to Metrax: Evaluation metrics for JAX

AI News

Can AI Program a Robot Dog?

AI News

Scaling AI Inference Performance in the Cloud with Nebius

AI News

NetApp and NVIDIA: Revolutionary AI Products at a Revolutionary Time

AI News

Kaggle Game Arena, Firebase Studio AI Workflows, Dart & Flutter Updates — A Comprehensive Guide for ComfyUI, AI, Educational, Developers

AI News

SIGGRAPH 2025 Special Address — AI, Simulation, and the Dawn of Graphics 3.0

AI News

Featured

Ally Solos Glasses - The most accessible AI assistant, now wearable.

AI News

Highlights from NVIDIA Keynote and GTC Paris 2025: Europe’s AI Revolution Unveiled

AI News

Featured

A Camera That Makes Poems: Exploring the Fusion of AI, Photography, and Poetry

AI News

Featured

This New AI AGENT Shocked the Market: Fully Autonomous and Self-Learning

AI News

Featured

This Open AI from China Just Crushed American Models and Broke Benchmarks

AI News

Featured

New Open Source AI From China SHOCKED the Industry Beating Titans at Just 7B

AIWorldVision