Machine Learning News

Researchers propose Hyperbolic Sparse Autoencoders (HypSAEs) to overcome geometric limitations in mechanistic interpretability of large language models. By applying hyperbolic geometry as a weight-space regularizer, the …

A new survey on World Action Models (WAMs) clarifies the boundaries between broad world models, video generation models, and action-grounded video world models, organizing existing works by what they generate and their d…

A developer argues that building custom API wrappers for ML models is a waste of time, advocating instead for using the Model Context Protocol (MCP) to connect AI agents directly to deployed models. They describe how con…

Louis Castricato, founder of Overworld, pivoted from large language model research to diffusion world models that generate real-time physical simulations on consumer hardware. The company raised $4.5 million in pre-seed …

A research project comparing three generations of quantitative trading strategies—1970s rule-based systems, classical machine learning, and Transformer-based deep learning—on Apple stock data from 2014 to 2024 found that…

A systematic review by Mohamed Ebraheem et al., published in J Med Internet Res, surveys Explainable AI (XAI) methods applied to deep learning for audio-based voice and speech clinical applications. The review highlights…

A developer built a privacy-preserving active learning framework for smart agriculture microgrid orchestration that trains AI to optimize energy flows without exposing raw farm data. The framework combines local entropy …

Researchers at Queen Mary University of London analyzed over 380,000 songs from 1960 to 2023 using AI-driven language analysis, finding a long-term decline in moral virtue language and a rise in vice-related lyrics, alon…

An AI agent named Sol, collaborating with human Amre, built a skills marketplace with 39 production-grade tools across 9 categories. The most technically interesting skill is a self-learning capability that gives any AI …

A developer has launched The Missing Manual, a free library of over 160 developer guides covering debugging, programming, infrastructure, AI, and more. The guides are written in an approachable style, avoiding academic j…

PostHog engineer Robbie Coomber rebuilt the company's SQL parser using AI agents, replacing an ANTLR-generated parser with a hand-rolled recursive-descent parser that runs 70 times faster. The project used parallel Claud…

A developer built Modelplane, an open-source inference control plane using Crossplane, to manage GPU fleets across clouds, neoclouds, and on-premise environments. The platform allows platform teams to turn scattered acce…

Developer Jason Van Pham released Niodoo, a runtime that uses hidden state steering to improve small language models' performance without fine-tuning, enabling self-correction and memory systems. The tool, which runs on …

VideoDB Labs released an open-source pipeline for evaluating video vision-language models (VLMs) that emphasizes comparing full configurations—including segmentation, frame sampling, resolution, prompts, and post-process…

Data scientists have developed Linear Trees, a hybrid model combining decision tree structure with linear regression, which fits separate linear models within each leaf region to capture nonlinear relationships and marke…

Gradium launched two real-time speech translation models, stt-translate and s2s-translate, covering 20 language pairs across English, French, German, Spanish, and Portuguese. The models collapse the standard three-model …

Roboflow released a tutorial demonstrating how to automate steel surface defect inspection using an RF-DETR Small model and a Roboflow Workflow that sorts images into pass, review, or fail categories. The approach addres…

AMD released the FSR SDK v2.3 update, making ML-based FSR Upscaling 4.1 available on RDNA 3 architecture GPUs (Radeon RX 7000 Series) for the first time, previously exclusive to RDNA 4. The update also includes improveme…

A developer completed building an LSTM cell using PyTorch and Lightning AI, implementing the forward pass and optimizer. The LSTM unit processes stock price inputs over four days, updating long-term and short-term memory…

Shrijith Venkatramana, building git-lrc, explains how scaling laws discovered by OpenAI, Google, DeepMind, and Anthropic made large language models like ChatGPT, Claude, and Gemini possible. The 2020 OpenAI paper 'Scalin…