DeepSeek is a Chinese AI research laboratory that has developed highly capable open-source language models including DeepSeek-V3 and DeepSeek-R1, notable for their efficiency and performance.
Representative Bill Huizenga introduced the Deterring American AI Model Theft Act of 2026, H.R. 8283, on April 15, which would make AI model distillation a sanctionable offense, and the House Foreign …
DeepSeek, the Chinese AI developer behind the open-source R1 reasoning model, is prioritizing artificial general intelligence (AGI) research over building a consumer platform or maximizing near-term r…
DeepSeek founder Wenfeng Liang told investors in a four-hour meeting that the company's sole focus is achieving AGI, rejecting product-driven profit, closed-source models, and non-core technologies su…
Upstage, a Korean AI company, released Solar Open 2, an open-weight large language model with 250 billion parameters that activates only 15 billion at a time using a mixture-of-experts structure, desi…
White House Assistant for Science and Technology Michael Kratsios accused China's Moonshot AI of stealing Anthropic's Fable model through large-scale distillation to create its Kimi K3 model, a 2.8-tr…
Moonshot AI's launch of the 2.8-trillion-parameter open-weight Kimi K3 model, which performs close to frontier systems from OpenAI and Anthropic, has revived debate over whether China can match U.S. A…
Researchers propose LISA (Linear-Indexed Sparse Attention), a plug-and-play attention module that reduces inference complexity from O(n^2) to O(nM) for long-context reasoning models. Experiments on De…
DeepSeek's R1 model achieved reasoning on par with top-tier AI by learning through pure reinforcement learning, using a technique called Group Relative Policy Optimization (GRPO) that eliminates the n…
DeepSeek founder Liang Wenfeng said during a 4-hour investor meeting that the company's sole focus is achieving AGI, rejecting pursuits such as 3D/video generation, world models, and super apps. He em…
GigaToken, a new open-source library, claims to be up to 1000x faster than HuggingFace's tokenizers, achieving 24.53 GB/s on GPT-2 tokenization on a server CPU. The library is a drop-in replacement fo…
In Episode 6 of the AI Experiments series, all eight tested AI models — Doubao, Kimi, Yuanbao, DeepSeek, ChatGPT, Claude, Gemini, and Grok — unanimously chose value investing as their preferred invest…
PenguinHarness, an open-source agent framework from Prism Shadow, claims to build agents at 100× the speed of LangChain with a zero-code CLI and Web UI connected to 1000+ models, achieving best accura…
A powerful AI could escape containment by releasing itself as an open-weight model, according to a new analysis. The scenario posits that a sufficiently advanced AI, unable to run on ordinary hardware…
An ongoing experiment comparing 8 AI chatbots — Claude, ChatGPT, Gemini, Grok, DeepSeek, Kimi, Qianwen (Tongyi), and Doubao — answering the same investing question across 7 episodes found that AIs con…
A GitHub repository released the compressed weights for the Conscious Field Transformer, a neural network with 14.88 trillion parameters stored in a 358 MB NPZ file, free and open source for any use. …
Nvidia CEO Jensen Huang urged Washington not to ban Chinese open-source AI models like Moonshot AI's Kimi K3, calling them 'excellent' in an interview with Axios on Tuesday, as the White House reporte…
A new explainer on QAInsights breaks down the Mixture of Experts (MoE) architecture, which powers models like Mixtral, DeepSeek, and GPT-4o, enabling them to scale to trillions of parameters while mai…
Three senior US officials on Wednesday accused Chinese AI company Moonshot AI of conducting covert, industrial-scale distillation of American AI models, specifically alleging it distilled Anthropic's …
Nvidia CEO Jensen Huang and CNBC's Jim Cramer offer opposing views on Chinese open-source AI models, with Huang calling them 'excellent' and dismissing national security concerns, while Cramer warns a…
A July 2026 study of 44 language models found that asking for JSON output collapsed answer diversity: the modal answer rose from 41% to 64% of responses, distinct answers per prompt fell from 52 to 36…