Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
A four-stage pipeline for implementing natural language search on Dublin rental listings, using an LLM to parse user queries into a structured JSON schema. The system then resolves geographic intent via a lookup table, e…
This week's top GPU news highlights three key developments: the release of Fournex, an open-source tool that analyzes Nsight Compute data to identify and suggest fixes for PyTorch/CUDA bottlenecks; a report forecasting t…
SandboxAQ has partnered with Anthropic to integrate its physics-based AI models, known as Large Quantitative Models (LQMs), into Claude, allowing researchers to perform complex drug discovery and materials science calcul…
OpenAI's cloud-based Chronicle service, which incurred per-frame costs for analyzing screen captures, with a local Gemma 4 E4B 4-bit MLX model running on a $599 Mac mini. This new setup processes video from four independ…
An autonomous AI system built for penetration testing that overcomes human limitations by combining tool selection, memory of past attacks, real-time vulnerability database queries, mathematical modeling for zero-day dis…
The author quit their FAANG job in 2024 to become an independent researcher and built a $48K GPU server called "grumbl" with six RTX 6000 Ada GPUs. They chose these GPUs over A100s and H100s based on price/throughput rat…
A computer vision system built on Roboflow now enables real-time detection of whether workers on construction sites are wearing helmets and safety vests, automatically tallying safe and unsafe workers directly on video f…
Amazon Web Services (AWS) has released guidance on using Amazon Nova 2 Lite for content moderation, enabling organizations to enforce custom policies through structured or free-form prompts without requiring model retrai…
Roboflow released a tutorial for building a real-time traffic light detection system using its platform and the RF-DETR model. The system, which achieved 84.3% mean average precision on a public dataset, is designed to s…
Roboflow released a tutorial demonstrating how to build a golf swing analysis pipeline using its 3.0 keypoint detection model and Google's Gemini 2.5 Flash for biomechanical reasoning. The system allows users to upload a…
Roboflow released an MCP server that integrates its computer vision platform directly into Claude Code, enabling developers to manage the full project lifecycle from a terminal. The server exposes 30 tools for tasks incl…
Aderant, a global provider of legal business management software, deployed Amazon Quick in October 2025 to unify search across six disconnected vendor systems and automate documentation workflows for its 38-person Cloud …
PyTorch 2.11 now enables direct installation of CUDA-enabled PyTorch wheels on aarch64 Linux from PyPI, eliminating the need for custom package indexes and workarounds that previously complicated deployment on NVIDIA GH2…
Cursor has released Composer 2.5, an AI model that offers significant improvements in intelligence, sustained task execution, and reliability over its predecessor. The model was enhanced through scaled training, complex …
Guide for fine-tuning NVIDIA's Cosmos Predict 2.5 world model using LoRA and DoRA techniques to generate synthetic robot manipulation videos. The approach freezes the base model's 2 billion parameters while training smal…
A deterministically chaotic market simulation successfully reproduced the lifecycle of a real-world stock buyback factor strategy, showing cycles of alpha generation, crowding, and eventual losses. The simulation, popula…
The ExecuTorch MLX delegate now enables GPU-accelerated inference for PyTorch models on Apple Silicon Macs through Apple's MLX framework. The new backend achieves 3-6x higher throughput on generative AI workloads compare…
Distribution Fine Tuning (DFT) is a new post-training algorithm designed to correct the formulaic and repetitive writing patterns common in standard supervised fine-tuned (SFT) language models. The method improves output…
PaddleOCR 3.5 introduces a more flexible inference-engine interface, allowing developers to select the backend (including Transformers) via the `engine` parameter and configure backend-specific options through `engine_co…
Amazon Web Services announced custom code-based evaluators for Amazon Bedrock AgentCore, allowing developers to use AWS Lambda functions as evaluation engines for agentic applications. The new feature enables determinist…