Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
OpenAI hardware team member and XLA creator Sanjoy Das described how AI is being used as a 'stochastic optimizer' to write and optimize kernels, such as the Jalapeño MLA kernel presented at HotChips, arguing that AI-driv…
A developer explains that generative models are best understood not as networks that create images but as solutions to a shared underlying problem: representing, learning, and inferring the probability structure of high-…
Researchers Riyaaz Shaik and Chandru Venkataraman introduced REFACTOR-VLA, a system that learns reusable motor skills for vision-language-action models using a wake/sleep architecture with a Behavioral-Equivalence Kernel…
Doublespeed AI, the model powering the social media platform doublespeed, achieved 61.0% accuracy in a benchmark of 200 real TikTok post pairs, outperforming frontier models including Claude and GPT 5.6 Sol, which scored…
A researcher's causal experiments on nanoGPT, ResNet-18, and a diffusion U-Net indicate that neural networks behave as classical stateful dynamical systems, where the same update has different effects depending on the re…
Researchers at Washington State University used artificial intelligence to identify efficient, low-cost 3D-printing parameters for the high-performance metal alloy GRCop-42, reducing the need to test over 100 million pos…
A developer's on-hardware test shows that prefetching mixture-of-experts (MoE) expert weights from disk ahead of matmuls can improve token generation speed on a severely resource-constrained machine, with the Intel Celer…
AgentSelfEdit, an open-source sidecar that rewrites its own system prompt from execution feedback, A/B tests prompt edits and promotes only statistically-proven winners. In a recent test, an LLM-proposed edit fixed 4 cla…
The Allen Institute for AI introduced BenchMIRT, a new method for auditing LLM benchmarks at the level of individual prompts, which uses multidimensional Item Response Theory to separate the underlying capabilities—such …
Model weights are the adjustable numerical parameters inside a machine learning model that are fine-tuned during training to improve performance, analogous to ingredient amounts in a recipe. Downloading model weights fro…
Waymo, with more than 200 million miles driven fully autonomously, shares ten lessons for building AI, emphasizing that multimodal sensors (cameras, lidar, radar) are indispensable and that HD maps serve as a powerful pr…
Torc Robotics, a Daimler subsidiary developing autonomous trucking software, is hiring a Senior Machine Learning Engineer for its Ego-Motion & Localization Team, offering a salary of $177k–213k per year for a remote posi…
Waymo, an autonomous driving technology company, is hiring a Sr Staff Tech Lead, Machine Learning Engineer for its Perception team in Mountain View or San Francisco, California, with a salary range of $298k–368k/yr. The …
Google Research engineers Vishal Batchu and Michelangelo Conserva developed MAPL-EMIT, a deep-learning framework that automates detection and quantification of methane plumes from NASA's EMIT hyperspectral satellite data…
A developer's guide highlights that only 18% of web developers report faster shipping times from AI adoption, while 73% of enterprise sites now use AI features. The post recommends integrating AI tools like GitHub Copilo…
Atlas, a new AI architecture, shifts from predicting text tokens to predicting the next state of a physical environment, enabling spatial reasoning for robotics and autonomous systems. By modeling 3D relationships and ac…
A developer built an AI agent that automatically writes tests to kill surviving mutants in Python code, but discovered that his own measurement harness contained eight bugs that skewed results. Testing 12 popular Python …
An engineer built AgentSelfEdit, an open-source sidecar that rewrites its own system prompt from execution feedback, and found that its deterministic safety gate rejected all 15 proposed edits. The gate, which uses six s…
Meta Superintelligence Labs has released Muse Voice Transcribe, a real-time speech-to-text model with speaker diarization and endpointing, available through Meta's Model API as part of the Muse Spark family. The model ai…
NVIDIA announced that its DLSS 5 neural rendering model is now five times faster than when first shown at GTC in March, requiring only a single RTX 50-series GPU instead of two RTX 5090s. The feature arrives September 3 …