Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
The Little Book of Reinforcement Learning, a short introduction to RL from basics to applied algorithms, is available on GitHub with PyTorch implementations and supplementary proofs. The book, authored by Alexandre Torre…
A developer developed Physics-Augmented Diffusion Modeling (PADM), a framework that embeds physical conservation laws directly into the diffusion process for generating synthetic geological maps of Martian terrain. The a…
Researchers have scaled zero reinforcement learning (zero RL) to a trillion parameters for the first time, achieving emergent reasoning capabilities in the Ring-2.5-1T-Zero model. The team introduced a stable training pi…
A comprehensive guide to LLM fine-tuning covers full fine-tuning, LoRA, learning rate, and VRAM considerations, explaining the major stages of training an AI model from data preparation to production monitoring. The guid…
Agility Robotics is opening a 59,900-square-foot research center in Fremont's Ardenwood neighborhood, planning to hire about 200 workers including software engineers with AI and machine-learning skills. The company has s…
A new book titled 'Mathematics of Data Science' by Thomas Strohmer, submitted to arXiv on 11 Jul 2026, covers the mathematical foundations of data science including topics such as high-dimensional geometry, singular valu…
Large open-weight mixture-of-experts (MoE) models are converging on lower routed-expert activation ratios, with the average ratio falling from roughly 3.7% in 2025 to 2.1% in 2026 across six major models released each ye…
Google Research released TimesFM 2.5, a decoder-only foundation model for time series forecasting, with 10,000 GitHub stars and support for installation, fine-tuning, and real-world applications. The model joins 40 resou…
Sunday Robotics unveiled ACT-2, a new AI model powering its Memo home robot that folds laundry successfully more than 99% of the time in unfamiliar homes on garments it has never seen before. The Bay Area-based startup a…
A developer explores using classical machine learning to detect LLM-generated text, highlighting advantages in interpretability, low computational cost, and effectiveness in constrained environments. The approach leverag…
A FreeBSD Foundation intern, Sourojeet Adhikari, is porting AMD's ROCm compute stack to FreeBSD, aiming to get at least simple vector addition working on the BSD platform. The project was highlighted at the recent FreeBS…
A developer built a 3-tier on-device AI concierge widget that runs entirely in the visitor's browser, using Google's Gemini Nano, a MiniLM embedding model, and keyword matching, achieving zero cost per query and full pri…
A developer built a hybrid RAG API that combines dense and sparse retrieval with reciprocal rank fusion (RRF) to answer questions from PDFs. The system uses Qdrant for vector storage, a cross-encoder for reranking, and p…
A developer known as lyc8503 has built an AI-generated text detector that uses classical machine learning models, specifically scikit-learn's Linear SVC and Naive Bayes, to distinguish LLM-generated text from human-writt…
Patrick O'Driscoll, a former nanotech physicist and current AI architect, introduces Competitive AI Safety as a paradigm to focus the field on measurable, tractable goals, drawing inspiration from OpenAI's Parameter Golf…
VLLM v0.25.0, released July 11, deletes the original PagedAttention implementation and makes Model Runner V2 the default execution backend for all dense models, delivering a 56% throughput improvement on Qwen3-0.6B. The …
Researchers from Google DeepMind and the University of Toronto have developed GenCeption, a method that repurposes a pre-trained video generative model into a single unified, feed-forward vision model capable of solving …
NVIDIA released the Nemotron 3 Embed collection of open embedding models, led by an 8B model that ranks #1 on the RTEB leaderboard with a score of 78.5%, alongside efficient 1B variants for production-scale retrieval. Th…
NVIDIA's Nemotron 3 Embed 1B embedding model took first place on every recall task in Zep's benchmark of 5,954 production queries, outperforming a 4B competitor with a third of the parameters and surpassing Zep's product…
A developer built Lexi-9-Omega, an AI engineering assistant that translates a sci-fi-inspired interface into practical software. The system operates in two layers—a narrative layer for identity and a technical layer for …