Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
NVIDIA and SchedMD have released a new topology/block plugin for Slurm 23.11 that enables topology-aware job scheduling on NVIDIA GB200 NVL72 systems, allowing workloads to be aligned with NVLink domain boundaries for op…
Andrej Karpathy's new Autoresearch method automates the experimental loop, running up to 100 AI experiments overnight while the user sleeps. The system autonomously tests ideas, discards failures, and retains only the re…
Glance, a mobile-first content platform, uses AI to transform long-form horizontal videos (1-2 hours) from sources like podcasts and movies into short vertical clips (30-180 seconds) optimized for mobile lock screens. Th…
An AI model has disproved an 80-year-old mathematical conjecture by Paul Erdős. The proof was generated by a new general-purpose reasoning model, not a system specifically trained for mathematics.
Development of an AI content repurposing tool that learns a creator's unique style through memory and feedback, rather than relying on generic templates or manual settings. The system observes how creators edit and appro…
Three key structural decisions made when building an enterprise RAG retrieval pipeline. First, the author chose deterministic lexical retrieval over semantic search for local validation to ensure reliable evaluation sets…
Building a custom on-device Agent Skill called "Offline Brain" for Android using Google's new AI Edge Gallery, which allows AI inference and tool execution to happen entirely on the device without sending data to a serve…
0.1.0 release of the Agent Development Kit (ADK) for Kotlin and a specialized ADK for Android, enabling developers to build AI agents for backend and on-device applications. The Android version allows agents to run local…
OPLOG, a fulfillment company operating across Türkiye, the United Kingdom, and Germany, built a production-ready business intelligence system using AI agents on Amazon Bedrock AgentCore to address fragmented business dat…
A global AI platform failed for users in Cameroon because it did not support French and pidgin English, leading to encoding errors and inaccessible models. After initially failing to adapt the platform's existing APIs, t…
Amazon Web Services released a reference architecture for building an AI-powered recruitment assistant using Amazon Bedrock, designed to automate candidate evaluation, generate personalized interview questions, and provi…
Amazon has released a solution enabling business analysts to modify dashboards using natural language commands, eliminating multi-day waits for IT teams. The system uses Amazon Bedrock AgentCore with Strands Agents and A…
Zep announced Observations, a new context type that automatically discovers patterns and insights across Context Graphs, capturing behaviors, decisions, and preferences that emerge across multiple episodes. The feature c…
Roboflow released a guide for integrating its computer vision workflows with OPC UA servers, enabling detection data to be written as live tags accessible to PLCs, SCADA, and dashboards. The process requires a Roboflow E…
Google has released Gemini 3.5 Flash, an AI model designed to close the gap between speed and intelligence, particularly for complex, multi-step agentic workflows. The model features a 1 million token context window, "th…
Local video indexing system on a 2021 MacBook using Gemma 4 to solve the problem of searching through a year's worth of unlabeled wildlife and travel footage. The author realized that AI video editing tools are ineffecti…
According to a May 2026 report from Tom's Hardware, the custom AI ASIC market is being driven by major deals from Broadcom and the continued development of in-house chips like Google's TPUs and Meta's MTIA. These special…
Enterprise inference now spans diverse hardware like GPU clusters and provider APIs, making each routing decision an infrastructure placement problem. The API gateway, designed for homogeneous traffic, lacks the visibili…
A 4-billion-parameter AI model released in early 2025 is now outperforming models seven times its size on standard reasoning benchmarks, with Google's Gemma 3 4B scoring 89.2% on GSM8K math reasoning and Microsoft's Phi-…
Running large language model inference servers like vLLM and TGI in production requires specialized observability because they behave differently from standard web services, with key metrics like latency being multi-dime…