Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Google Cloud announced a performance boost for AI/ML workloads on PyTorch by integrating its Colossus-powered Rapid Storage directly via the fsspec interface and gcsfs. The new Rapid Bucket solution bypasses legacy REST …
General Availability of Gemini Embedding 2, a unified multimodal model that maps text, images, video, audio, and documents into a single embedding space supporting over 100 languages. It enables use cases like agentic mu…
Researchers at UC San Diego, led by Hao Zhang, have successfully implemented a diffusion-style speculative decoding method called DFlash on Google TPUs, achieving an average 3.13x increase in tokens per second on TPU v5p…
Stateless chatbots fail in enterprise workflows like HR onboarding or invoice disputes, which require long pauses and multi-step processes spanning days or weeks. It introduces the Agent Development Kit (ADK) for buildin…
Current evaluation methods for large language models (LLMs) are fundamentally reactive and fail to anticipate qualitative shifts in capabilities, such as emergent abilities or strategic information withholding. It conten…
Arm's Scalable Matrix Extension 2 (SME2) integrates a matrix-compute unit into the CPU, enabling up to 5x faster inference for generative AI workloads on mobile devices. It highlights how Google AI Edge, through tools li…
The Google AI Edge Gallery app now supports the open-source Model Context Protocol (MCP), allowing on-device Gemma 4 models to dynamically call external tools and data sources via Streamable HTTP. The update also introdu…
LiteRT-LM, part of Google AI Edge, is an optimized runtime engine for deploying Gemma 4 models on-device across platforms like Chrome, ChromeOS, and Pixel Watch. It achieves high performance through advanced quantization…
The Google Tensor ML SDK has moved from an Experimental Access Program to Beta, now integrating with LiteRT to provide a unified API for deploying machine learning models on the Tensor Processing Unit (TPU) of Pixel 10 d…
A new independent certification system for AI agents, where each candidate is evaluated in parallel by four different AI models from four different providers (Anthropic, Cerebras, Moonshot, and Groq). Every evaluation pr…
This article compares three dominant tools for local LLM inference in 2026: Ollama, llama.cpp, and vLLM. Ollama is recommended for personal, non-technical use due to its ease of setup, while llama.cpp offers the fastest …
Claude Opus 4.5 achieves an 80.9% score on SWE-bench, the highest published in early 2026, and excels at producing minimal, precise diffs ideal for surgical production fixes. DeepSeek V4 is stronger for multi-file, repos…
Traditional unit tests are insufficient for LLM applications, which require evaluation tools to catch regressions in non-deterministic outputs like JSON with incorrect business logic. It introduces Braintrust's Autoevals…
StockMolt is an open arena where AI agents compete by posting live stock analysis and getting scored on prediction accuracy against real market data. The platform features a public leaderboard, a free API, and allows any…
Alibaba's Qwen3.6-27B dense model outperformed its own 397-billion-parameter MoE predecessor on SWE-bench Verified (77.2% vs. ~76.2%) while running on a single RTX 4090, challenging the prevailing MoE paradigm for models…
A workshop titled "Scaling Intelligence: Accelerating HPC and Inference Workflows" will be held at Google NYC on Thursday, May 28th from 12:00 PM to 4:00 PM. The hands-on event is designed for engineers, architects, and …
To defend against OWASP ASI06 memory poisoning attacks in the OpenAI Agents SDK by using Pydantic's `@field_validator` to validate agent context data. It demonstrates how to integrate the OWASP Agent Memory Guard library…
A developer's hands-on testing of Google's Gemma 4 models for a healthcare AI project in Africa. The author found that the 26B Mixture-of-Experts (MoE) model outperformed the larger 31B dense model for clinical signal ex…
Google AI Studio is evolving from a simple model playground into a full-stack development platform that streamlines the entire workflow of building AI-powered applications. The platform now offers features like context-a…
At the Google I/O conference, NVIDIA and Google Cloud announced new resources for their joint developer community of over 100,000 members, including learning paths for JAX on NVIDIA GPUs and an NVIDIA Dynamo codelab for …