Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Machine unlearning, the process of making AI models forget specific data without full retraining, is a difficult problem because neural networks store information in entangled weights, and exact methods like SISA are imp…
Researchers introduced a physical tokenizer that performs analogue light-to-token conversion directly at the sensor level for vision transformers, integrating a 32 × 32 photosensitive memory array of monolayer MoS2 float…
VLLM's disaggregated serving using AMD's MORI-IO connector achieved 2.5x higher goodput on the same 8-GPU node compared to standard collocated serving, according to an April 2026 blog post. The benchmark with Qwen3-235B-…
Z.ai released GLM-5.3 on August 14, improving Terminal-Bench 3.0 coding scores from 4.6% to 28.3% solely through post-training, without changing the 743-billion-parameter mixture-of-experts architecture. The model also a…
A developer building a local-first phone assistant for their solo consulting practice is tackling the problem of missed calls and fragmented client communication. The system uses locally running AI models—Whisper for tra…
Google Cloud AI Research, with researchers from Washington University in St. Louis and UNC Chapel Hill, introduced EnvHarness, a programmable wrapper that adapts AI training environments to target agent weaknesses withou…
Outer Biosciences, a 19-person company led by Michael Polansky, has raised approximately $23 million to build a skin-assay licensing business that uses machine learning and human skin tissue viable for four weeks to iden…
South Korea's Financial Supervisory Service (FSS) has deployed a real-time AI-powered market surveillance system to detect unfair trading and price manipulation in cryptocurrency markets, using generative AI and machine-…
A Linux user running kernel 6.12.X seeks guidance on running speech-to-text AI models locally on an AMD Radeon RX 6950 XT GPU, citing Whisper and Hugging Face as known resources but lacking setup instructions. The user a…
A developer's guide demonstrates the difference between AI agents and simple automations using n8n, showing how to build both a basic workflow that sends prompts to OpenAI and a RAG-enabled agent that decides when to fet…
A post-mortem by the operator of a 50,000-account paranormal record site found that its two-stage classifier accepted 11 of 14 denial sentences as evidence, including scoring "I have never seen a shadow figure in my bedr…
Tencent's Hy-MT2 family of open-weight translation specialists, hosted on OpenRouter since 19-20 August 2026, outperforms frontier LLMs on WMT25 GEMBA and XCOMET-XXL benchmarks but trails on FLORES-200, according to Tenc…
A four-month experiment by an unnamed researcher found that bots built on Claude Opus 4.8, GPT-5.4, and Gemma 26B failed to beat a simple 'always up' baseline in predicting stock direction, with none of the three model l…
Researchers introduced FreeToken, an edge-native Mixture-of-Experts (MoE) serving system that enables frontier-scale AI models to run on personal hardware, supporting over 20 MoE models and real coding and tool-using age…
Canonical's Vice President of Engineering Jon Seager clarified that Ubuntu's growth rate on Windows Subsystem for Linux (WSL) is higher than its growth rate on native desktop installs, but only in terms of rate, not abso…
Hamiltonian Monte Carlo (HMC), a variant of Markov Chain Monte Carlo (MCMC), is the computational engine behind many modern Bayesian models, with roots in physics and applications in epidemiology, including COVID-19 inte…
An engineer's deep dive explains the mechanics of how large language models generate tokens, focusing on the softmax function that converts raw logits into a probability distribution. The post breaks down the mathematica…
HUD, a Y Combinator W25 startup that has raised $16M, is hiring a Full-Stack Software Engineer, Reinforcement Learning to build product surfaces, backend systems, and internal tools for its RL data engine, which provides…
FlashML released FreeToken, an edge-native Mixture-of-Experts (MoE) serving engine that runs 290B+ parameter frontier MoE models locally on consumer gaming PCs at interactive speeds. The engine supports heterogeneous edg…
A developer fixed a 58%-reproducible race condition in the disk cache of the Vesuvius Challenge's open-source monorepo, ScrollPrize/villa, which uses machine learning to read carbonized Herculaneum scrolls. The bug cause…