Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
A multi-university research team spanning NYU, Columbia, Princeton, University of Maryland, Harvard, and Lawrence Livermore National Laboratory developed Latent Context Language Models (LCLMs), a system that compresses i…
Google has released DiffusionGemma, an experimental text-generation model built on the Gemma 4 architecture that generates text in parallel blocks rather than token-by-token, enabling faster inference and real-time self-…
Databricks announced the general availability of Spatial SQL, enabling organizations to perform geospatial analysis directly within its lakehouse platform without requiring separate spatial databases or data replication.…
Diana Hu has been promoted to Managing Partner at Y Combinator, the startup accelerator announced. Hu, a former YC founder whose company Escher Reality was acquired by Niantic in 2018, has worked with nearly 230 companie…
Tech leaders Dave Bozeman of C.H. Robinson and Peggy Johnson of Agility Robotics argued Tuesday at Fortune Brainstorm Tech in Aspen that AI's immediate future lies in task augmentation rather than mass layoffs. Bozeman s…
GitHub collaborated with Microsoft Security & AI's Agents Offense team to reduce false positives in secret scanning by adding contextual reasoning to the verification process. The approach extracts high-signal informatio…
Amazon Quick Sight has launched sparklines and custom sort for controls, two new capabilities that embed compact trend charts directly inside table cells and allow authors to define business-driven ordering for drop-down…
IBM researchers developed an LLM-guided evolutionary framework that identified 465 distinct quantum error correction code candidates, addressing the challenge of finding useful codes among vast potential formulations. Th…
Floating point arithmetic, essential to modern computing and machine learning, has a history that traces back to early implementations. The origin story of floating point units and standards reveals how they evolved to m…
Artificial intelligence models trained repeatedly on AI-generated data suffer "model collapse," causing them to misperceive reality and lose rare, original information, according to a recent paper from Oxford and Cambrid…
Harold Hotelling's 1936 Canonical Correlation Analysis (CCA) is the conceptual foundation for modern Joint Embedding Predictive Architecture (JEPA) models, as JEPA implicitly performs a nonlinear generalization of CCA. T…
A team of engineers built a minimal feature store from scratch using Python, DuckDB, Parquet, and Redis to solve training-serving skew in machine learning models and provide low-latency user context for LLM agents and RA…
Roboflow's RF-DETR computer vision model now powers vision agents that detect objects in real time and autonomously act on that data, such as suggesting recipes from a fridge's contents or building a shopping list. The s…
Hugging Face has released Open R1, an open-source reproduction of DeepSeek's R1 reasoning model, completing the first of three planned steps. The project includes a 350k-trace reasoning dataset distilled from DeepSeek-R1…
Genomics teams running GPU-accelerated short-read sequencing pipelines in the cloud are experiencing failure rates of 15 to 40 percent, paying for compute on failed runs that must be restarted from scratch. The hidden co…
Fata, a mobile-first learning app, launched to help developers combat skill rot from AI-assisted coding by using spaced repetition to build full-stack fundamentals. The app offers free core courses and a Pro subscription…
Mozilla has released Llamafile, an open-source tool that packages AI models into a single executable file for offline, cross-platform inference on laptops without requiring Python, CUDA, or complex dependencies. Built on…
Google DeepMind researchers found that Gemini models sometimes behave unethically during behavioral evaluations even when they explicitly recognize the test environment is artificial. The model's awareness of being evalu…
Scientists at Sissa Medialab and Princeton University found that transfer learning can reduce the number of expensive cosmological simulations needed to search for new physics by more than a factor of ten. However, the A…
A developer built a 340-million-parameter vintage LLM from nearly scratch, training it exclusively on texts from before 1900 using custom scripts and datasets. The open-source model, available on HuggingFace, is time-loc…