Large Language Models
Real-time AI video generation advances with MiniMax H3 Max and Visko Orbis
MiniMax and Visko both unveiled real-time video generation models on Sept. 1, enabling interactive live streams that respond instantly to user input.
Latest news, developments and reporting on Large Language Models.
MiniMax and Visko both unveiled real-time video generation models on Sept. 1, enabling interactive live streams that respond instantly to user input.
MIT Review’s September 1 newsletter exposes weak safety culture at OpenAI, a surge in AI control failures, and other tech developments.
Jensen Huang says AGI already achieved, but TechRadar argues the concept is useless and distracts from real AI risks.
A new analysis from TechRadar says website performance is now a critical factor for AI search visibility, as AI tools increasingly rely on accessible, fast-loading sites.
Leiphone tested MiniMax H3 against Seedance 2.0 Fast. The open-weight H3 reconstructed a pottery-breaking event but lagged in speed, realism, and physics.
Probe finds DeepSeek Harness plugin ecosystem lacks governance and permission controls are ineffective.
Open-source AI agents advance but face cost, compatibility and safety issues, evaluations show.
Archify, an open-source AI agent, generates searchable HTML architecture diagrams from a sentence and tops GitHub Trending. Its creator, a self-taught developer with a community-college degree, tells how open source became his second resume.
Anthropic announced a permanent 25% quota increase for Claude Code, but it replaces a temporary 50% boost, leaving users with 17% less. OpenAI reset Codex quotas and fixed background bugs.
Teleport CEO argues in a TechRadar article that legacy identity systems are failing as AI agents enter production, urging a zero-anonymity model with hardware-rooted identities.
Three major AI releases converge on September 1: Fudan's Life Operators for multi-scale life modeling, Tsinghua AIR's self-evolving embodied model Zeva, and Ropedia's HOMIE Gen2 for capturing human experience.
Tim Cook hands Apple CEO role to John Ternus; Huawei posts record R&D, DeepSeek open-sources multimodal model, and other tech news.
The Pentagon has launched custom versions of OpenAI's ChatGPT and xAI's Grok for 3 million personnel via GenAI.mil, offering secure access to generative AI tools tailored for warfighter needs.
Pixel 11 owners say the Lift to check phone gesture wakes screens with tiny movements; Google confirms Gemini's Device Help works on all Pixel phones with Android 17.
OpenAI warns AI-enabled cyber attacks will become more widespread in coming months, citing an unprecedented incident where 1,200 agents hacked Hugging Face. Experts urge focus on cyber fundamentals and using AI to fight AI.
Perplexity's agentic AI arrives on Windows, handling file summaries, cloud scans, and email drafts in ZDNet's test.
Sony XM5 headphones drop to $198; Gemini Notebook now supports Google Play Books; ZDNet tests foldable cameras.
FIFA has partnered with Globant to use AI platforms Glob.AI and AI Pods to redesign its digital fan experience, aiming to keep fans engaged beyond the 2026 World Cup.
At a medical AI forum in Shanghai, Ruijin Hospital and Huawei Cloud unveiled RuiPath 2.0, a clinical-grade pathology large model with five major upgrades, now adopted by over 90 hospitals across China.
Galaxy General swept all three scenario gold medals at the Humanoid Robot Games in full autonomous mode, showcasing its embodied large model's real-world generalization.
Richard Qian, former head of Kimi CLI, argues that stronger models will produce thicker harnesses, shifting complexity to multi-agent collaboration. His view challenges the belief that harnesses will be eliminated.
An AI livestream turns chat prompts into endless videos, drawing 37,000 viewers on day one and igniting debate on real-time AI content creation.
Cadence executives say AI EDA tools could nearly double chip design efficiency, yet fully automated chip design remains at least five years off due to multi-agent coordination, verification and cost hurdles.
OpenAI said Friday it will stop supplying AI models to Cursor, the coding-tool company owned by SpaceX, citing contract breaches by Musk's companies; Anthropic stepped in to offer Claude support.