MegaTrain: Full Precision Training
MegaTrain enables full precision training of large language models on a single GPU. It stores parameters in host memory and uses GPUs as compute engines.
All 3,523 AI stories, newest first · page 147 of 147
MegaTrain enables full precision training of large language models on a single GPU. It stores parameters in host memory and uses GPUs as compute engines.
A new local data lake enables AI-powered data engineering and analytics without cloud setup overhead. It allows for quick and interactive data analysis with SQL, Py, and natural language querying.
A large language model was connected to an 8-bit shoot-'em-up game, receiving structured text summaries instead of traditional inputs. The model developed strategies and discovered an exploit in the game's AI.
A C++ LLM inference engine from scratch has output tokens costing 5x more. The article explores the reasons behind this increased cost.
Linggen is a local-first, open-source AI coding agent with P2P remote access. It proposes step-by-step plans before writing code and works with various models.
Iran's IRGC has threatened OpenAI's Abu Dhabi data center in response to potential US attacks on Iranian power plants. The IRGC vows to annihilate US-linked energy and tech companies in the region.
Intel is partnering with Elon Musk's Terafab to design and build an AI chip factory in Austin, Texas. The facility will supply AI chips to SpaceX and Tesla, supporting Musk's plans for self-driving cars and humanoid robots.
Holotron-12B is a new high throughput computer use agent. It is designed for efficient computing.
Holo3 breaks the computer use frontier with its open-source release. This innovation is set to revolutionize the way we interact with computers.
Google is investing in open source security for the AI era. The company is building new tools and developing code security to improve open source security.
Google is expanding Personal Intelligence across AI Mode in Search, Gemini app, and Gemini in Chrome. This move aims to bring personalized AI experiences to more users.
Google's Gemini now directs users to mental health resources faster during crises. The update aims to provide timely support for those in need.
OpenAI outlines the next phase of enterprise AI with accelerated adoption. This phase includes Frontier, ChatGPT Enterprise, Codex, and company-wide AI agents.
CongaLine is a self-hosted AI agent fleet that allows teams and individuals to run isolated AI assistants. It prioritizes security and offers flexible deployment options.
Anthropic is set to preview its powerful Mythos model to combat AI cyberthreats. This move aims to enhance security against emerging threats.
Anthropic's new AI model, part of Project Glasswing, identifies security vulnerabilities in major operating systems and web browsers. The model is offered to launch partners, including Nvidia and Google, to enhance cybersecurity with minimal human intervention.
AMD's AI director claims Claude Code has become dumber and lazier since its update. This statement raises concerns about the model's performance.
OpenAI's new Realtime Video API lets developers pipe live camera streams directly to GPT-5V, enabling sub-second visual understanding for robotics, accessibility tools, and consumer apps.
DeepSeek's latest model, R3, has set new records on AIME and HumanEval, outperforming GPT-5 and Gemini 2.5 Pro on multi-step reasoning tasks while remaining fully open-weights.