News about foundation models, LLMs, multimodal models, benchmarks, and model releases from OpenAI, Anthropic, Google, Meta, Mistral, and others.
719 articles
An expert explains the key distinctions between AI model licensing types, noting that open-source models let developers view and modify the underlying code. The explainer clarifies why these differences matter for transparency, customization, and control over AI systems. Source: pbs.org
Open Dreamer is a JAX/Flax reimplementation of the Dreamer 4 world-model pipeline, released together with complete training code, configuration files, and notes on stability engineering. The project aims to make the advanced world-model approach more accessible and reproducible for researchers. Source: marktechpost.com
Anthropic has launched a new large language model, Claude Opus 5, on both its consumer chatbot app and developer platform. The company claims the model brings notable improvements in performance efficiency and safety compared to its predecessors. Source: siliconangle.com
LLMs generate text by predicting the next token based on patterns learned from massive datasets. Researchers use the concept of scaling laws to describe consistent relationships between model size, data, compute, and resulting performance. Source: aimultiple.com
A new study published in npj Digital Medicine examines whether large language models can help identify individuals at clinical high risk for psychosis. Early detection remains central to psychosis prevention, and researchers assessed how well LLMs perform this task. Source: nature.com
Hallucinations remain one of the most persistent issues with large language models, occurring when a model generates false or unsupported information. A new evaluation approach called GraphEval is designed to better detect and quantify these errors. Source: kdnuggets.com
US officials claim Chinese AI company Moonshot AI copied elements of Anthropic's advanced Fable LLM to build its new K3 model. The accusation adds to growing tensions over AI technology transfer between the US and China. Source: timesofindia.indiatimes.com
NASA's NAVI-Orbital project uses AI to analyze satellite imagery in space. Google's Gemma 3 compresses images into text, allowing faster data transfer and quicker disaster-response decisions. Source: spectrum.ieee.org
A White House science advisor accused Chinese company Moonshot AI of copying Anthropic's Fable LLM and deploying advanced Nvidia GPUs that are barred from export to China. The allegations point to both intellectual property theft and violation of chip export controls. Source: pluang.com
The White House accuses Moonshot AI, a Chinese AI startup, of illegally using American technology to develop its latest large language model. The case highlights ongoing US concerns about Chinese firms accessing restricted AI know-how. Source: news.az
The review examines how large language models, AI vision, and embodied intelligence turn a standard robotic arm into a smart AI-powered assistant. The DOFBOT Ultra is tested alongside other AI-related hardware as part of the hands-on evaluation. Source: hackster.io
New research argues that different questions require different answer formats — a single number with a source, a structured list, or a simple yes/no. The analysis identifies seven recurring error patterns and proposes a typed generation contract to reduce hallucinations in retrieval-augmented generation systems. Source: towardsdatascience.com
AI models are often labeled as open or closed, but the terms have little to do with personality and everything to do with access. A software engineering researcher breaks down what openness actually means for large language models. Source: theconversation.com
Researchers are testing whether large language models can reliably simulate human behavior in complex decision-making scenarios. Early findings raise questions about how faithfully LLMs replicate real human risk preferences. Source: miragenews.com
DGT has released an open AI model built to treat all 24 EU languages equally. Source: dig.watch
Upstage has unveiled its open-source large language model (LLM) 'Solar Open 2', optimized for AI agent tasks. Upstage announced on July 23 that it has... Source: chosun.com
China's Moonshot AI stole from Anthropic's most sophisticated AI large language model, Fable, to help create its latest K3 release and has acquired advanced... Source: nypost.com
Upstage has unveiled 'Solar Open2,' a large language model (LLM) aimed at its proprietary ... Source: m.ajupress.com
Today, Cisco is introducing Antares, a family of security small language models (SLMs) purpose-built for one of the hardest, most time-consuming and... Source: blogs.cisco.com
WASHINGTON, July 22 (Reuters) - China's Moonshot AI used Anthropic's most sophisticated AI large language model, Fable, to help create its latest K3 release... Source: ca.news.yahoo.com