Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Marvin Minsky and Seymour Papert's 1969 book 'Perceptrons' demonstrated that a single-layer perceptron cannot solve the XOR problem, and they incorrectly argued that multi-layer networks would not help, leading to a fund…
A developer argues that vector databases should adopt tiered storage, similar to traditional data storage, to reduce costs. They cite a startup whose vector index grew from 50 million to 500 million embeddings, causing i…
A new experiment by Dylan Castillo comparing Matryoshka Representation Learning (MRL) and Principal Component Analysis (PCA) for reducing embedding dimensions found that both methods preserve retrieval quality, but MRL r…
LangChain 0.3.0, Transformers 5.0, Claude 4 Ultra, AWS AI SDK 3.0, and Docker-AI 1.2 were released in late July 2026, offering developers enhanced AI capabilities, performance improvements, and new deployment options. Th…
A 15-year-old developer from Germany, known as Mojo, has built an integer-only AI in Rust that achieves 99.7% accuracy on the XOR problem, with 98.7% of runs being perfect. The project, called ABSL (Adaptive Bitshift Lea…
An engineer at PreserveMy.World built a CPU-only Structure-from-Motion pipeline using COLMAP to reconstruct the Moorcroft Pavilion at Shalimar Gardens, Lahore. By curating 15 tightly-overlapping images instead of 43 gene…
Researchers at the University of Illinois Urbana-Champaign and Harvard University introduced Explorative Modeling (XM), a training method that generates K candidate outputs per example and backpropagates through the lowe…
Tracer, a San Francisco AI lab founded by Adam Rida, launched Echo on July 31st, an adaptive inference system that coordinates multiple open-weight models and returns a single answer through an OpenAI-compatible API, cla…
Pinterest is hiring a Staff Machine Learning Engineer for its Applied Research team, offering a salary of $189k–390k/yr for a remote position based in San Francisco, CA. The role involves conducting research in machine l…
A developer's analysis shows that INT4 weight-only quantization speeds up decode but not prefill, because prefill is compute-bound while decode is memory-bound. The crossover point where a GEMM becomes compute-bound is l…
Handshake, the career platform powering 25 million job seekers, 1 million+ employers, and 1,600 educational institutions, is hiring a Machine Learning Engineer I for its Network and Handshake AI Marketplace Relevance tea…
Researchers distilled DeepSeek V4 Flash into the open-source GPT-OSS-120B model for finance reasoning, achieving 83.61% on FinanceReasoning at an 8,000 token budget, outperforming Kimi K3 (81.93%) and Inkling (65.1%). Th…
In a technical explainer, the author demonstrates that L2 regularization and weight decay, equivalent in plain SGD, diverge under the Adam optimizer because Adam's adaptive step rescales the L2 penalty, causing parameter…
Flapping Airplanes, an AI startup researching data-efficient model training, is in talks to raise funding at a $5 billion valuation, according to Forbes. The company aims to improve AI training methods using less data.
FACEIT will launch a machine-learning anti-cheat feature called 'Human Input Detection' on August 5, 2026, alongside Season 9 of its Counter-Strike 2 platform, which analyzes player inputs to distinguish human behavior f…
Google DeepMind unveiled Gemini Robotics 2, its most advanced vision-language-action model, designed to control robots ranging from tabletop arms to full-body humanoids, along with Gemini Robotics ER 2, a higher-level re…
Researchers led by Tonio Ball introduced the Inverter framework, a neuro-inspired approach for embodied planning and control that uses paired forward/inverse internal models and open-loop multi-step motor commands, train…
A developer built a personalized long-term memory RAG stack on Fedora 44 using Mem0, Qdrant, and Open WebUI, with a FastAPI proxy that injects relevant past facts into chat prompts and auto-extracts new memories in the b…
SWE-rebench's latest leaderboard shows frontier coding models within 2.2 points of each other on resolved rate, with Fable 5 (Anthropic) at 64.5%, Grok 4.5 (xAI) at 63.8%, Opus 5 (Anthropic) at 63.4%, GLM-5.2 (Z.ai) at 6…
Microsoft released Echoverse, a set of twelve high-fidelity training worlds for computer-use agents, and reported that a 9B model trained on all twelve nearly doubled its base score from 36.5% to 67.1%, coming within fou…