Hugging Face and OpenRouter have partnered to simplify the process of running open-weight models. Open-weight models, which are now being published on Hugging Face Hub, are served at scale by OpenRouter, which determines the best provider for each request based on live measurements of price, speed, and reliability. Any open-weight model from the Hub can be called through the same API using a Hu...
Fine-tuning is a process for adapting pretrained large language models to specific domains, tasks, or styles for businesses, delivering better accuracy and reducing costs. Parameter-efficient methods like LoRA have emerged, but their low-rank nature limits their representation power, compromising performance on complex tasks. Amazon SageMaker AI Serverless Model Customization now supports Full ...
This article discusses the challenges and inaccuracies of estimating the fine-tuning time for generative language models using GPU clusters. It compares the traditional Floating-Point Operations (FLOPs) calculation method against a more advanced CatBoost regressor that considers training-specific signals and learns nonlinear interactions. The study, which analyzed 272 real supervised fine-tunin...
RadixArk's first model, DaisyRose-27B, built on Qwen3.8-27B and trained with reinforcement learning using Miles, achieves significant improvements on long-horizon agent tasks, surpassing the performance on Terminal-Bench 4.1 by 6.4 percentage points. The model also delivers gains on a variety of reasoning, instruction following, and coding benchmarks. The trained model, available under Apache 2...
New compact desktops from ASUS, Dell, and MSI, powered by NVIDIA RTX technology, are now available. These desktops offer local implementation of AI agents, image and video generation, and model fine-tuning on a dedicated Windows PC. With up to 128GB of unified memory and NVIDIA's AI software stack, these systems cater to larger local models while providing access to creative and development too...
The Nutrient Data Extraction API is a document parsing tool that extracts values from PDFs, images, and Office files, converting them into JSON or Markdown. It offers various parse modes, including text, structure, and agentic, with different pricing structures. The API provides performance enhancements, such as confidence scores, bounding boxes, and match labels. Users can test the API through...
Researchers at MIT have developed a flapping-wing aerial microrobot with onboard sensors and computation, allowing it to follow complex trajectories and land precisely outside of a laboratory motion-capture environment. The robot, which is roughly the size and shape of a microcassette, is powered by soft actuators and can carry necessary electronics for sensing and computation. This is the firs...
LLama.cpp is a versatile and lightweight platform for developing cross-platform AI products, supporting multi-modal large language models across various hardware from local machines to enterprise servers. Through techniques like quantization, optimized backends, and C/C++ implementation, it ensures AI inference is accessible at every level. With a growing community of skilled contributors, the ...
A new benchmark called XYBench has been developed to test AI assistants' ability to help users with misguided questions, which often stem from misconceptions about the problem at hand. Developed in collaboration with researchers from Carnegie Mellon, NVIDIA, and Johns Hopkins, XYBench contains over 8,000 real-world questions spanning various domains. The main objective of XYBench is to evaluate...
A teacher who believes a student’s term paper may have been written by a large language model (LLM) can use a tool for detecting artificial intelligence-generated text to help confirm their suspicions. But the teacher may want to think twice before putting all their trust in the detector. Researchers from MIT and Cornell University showed that such detection tools, which are AI models ...
Introducing GPT‑6.2 Astra and Sol, with better professional outputs, coding, computer use, and efficiency at lower cost across ChatGPT Work, Codex, and the API.
How OpenAI is working with NATO and allied cyber agencies to help trusted defenders find vulnerabilities, test fixes and protect essential systems.
Astra for Law ranked first in a launch evaluation spanning 24 areas of law. Every ChatGPT Enterprise customer can now request access.
A simple decision tree for regression will almost always overfit the training data where the model accuracy on the training data is near 100% but accuracy on new, previously unseen data is very poor. The simple idea of random forest regression is to create a collection/ensemble of many simple decision trees, where each tree is trained on a different subset of the rows of the training data, a...
OpenAI and the Government of Chile are partnering to support responsible AI adoption, build public-sector skills and help local startups develop AI applications.