Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Andrej Karpathy, a co-founder of OpenAI and former AI leader at Tesla, has joined Anthropic to work on pre-training research, focusing on using its Claude model to accelerate the development of large language models. Ant…
Amazon SageMaker Feature Store now supports native AWS Lake Formation integration, additional Apache Iceberg table properties, and Feature Store capabilities in SageMaker Python SDK v3.8.0. The new features enable fine-g…
A team built a procurement agent that improved its decision-making accuracy from 53.8% to 83.1% by capturing human reviewer overrides as a context graph, without any code changes or model retraining. The agent initially …
Agoda has developed a multimodal content system that unifies hotel images and guest reviews into a shared topic-based taxonomy, enabling consistent interpretation of features like Pool or Breakfast across visual and text…
A presentation by Intuit's Merrin Kurian on building a Generative AI infrastructure stack, focusing on the company's "GenOS" platform that powers AI agents across its products. It highlights key metrics, such as 60 billi…
Forge is a reliability layer for self-hosted LLM tool-calling that uses guardrails and context management to dramatically improve performance on multi-step agentic tasks, boosting an 8B model from 53% to 99% accuracy. It…
Vlad Feinberg published a guide on how to land a job at a frontier AI lab focused on pretraining, emphasizing kernel-level performance work as the most direct path into the labs. The guide recommends deriving Chinchilla …
Tool-using research agent built for the Gemma 4 DEV Challenge, which runs locally on the Gemma 4 e2b model via Ollama using roughly 200 lines of Node.js code. The agent accepts a question, selects between two tools to re…
The author successfully ran Google's new Gemma 4 open models (2B and 4B parameters) on a laptop using Ollama during Google I/O 2026 week. To make the small 2B model function as a useful research agent, they built a five-…
TinyFish achieved an 81% success rate on the Mind2Web benchmark, significantly outperforming OpenAI Operator's 43% score on hard web tasks. The company tested its web agent across all 300 tasks on live websites, publishi…
Slow single-image diffusion model inference is primarily caused by kernel launch overhead and attention memory traffic, not by a lack of computational power. It recommends using `torch.compile` with `mode="reduce-overhea…
Deploying large language model (LLM) systems in production is far more complex than simple tutorial demos, as real-world challenges like state changes, concurrency, and regulatory scrutiny break naive assumptions. It emp…
Development of a full-stack mobile and AI-powered supply chain optimization app built with FastAPI, React Native, and data science tools. The project integrates AI directly into a real application, using Pandas for analy…
According to the article, Endress+Hauser developed an AI diagnostic system that analyzes real-time sensor data and historical telemetry from industrial instruments to detect and classify measurement faults, eliminating 8…
The article investigates whether increasing an AI model's "thinking time" through additional recurrent loops during inference improves its performance, using a small-scale OpenMythos model trained on multi-digit addition…
PyTorch's `torch.load()` function uses the Python pickle format for serialization, which inherently executes arbitrary code when loading a file, making it trivially exploitable for remote code execution (RCE) attacks. Th…
Practical starting point for security teams new to AI red-teaming, emphasizing that the process requires structured preparation distinct from traditional security testing. It highlights prompt injection (both direct and …
Key developments in large language models (LLMs) from November 2025 to May 2026, highlighting a critical inflection point where coding agents became reliably effective for daily use. During this period, the title of "bes…
Researchers have demonstrated that adversarial commands can be hidden within audio that sounds normal to humans—using ultrasonic frequencies or psychoacoustic masking—and voice AI transcription pipelines will faithfully …
A project called Turtle-Gemma that combines a modern AI language model with the classic Logo programming turtle, allowing users to speak commands like "draw a red star" which the AI then converts into Logo code and execu…