🧠 AI & Agents
Claude and nine-loop physics: what was calculated
In a 25 September 2026 article hosted by Anthropic, physicist Matt von Hippel describes a nine-loop calculation in the N=4 super Yang-Mills model theory.…
🧠 All articles in
AI & Agents
Autonomous agents, frontier models and how AI is moving from chat to action.
🧠 AI & Agents
NVIDIA Isaac ROS 5.0 brings agents into robotics development
On 22 September 2026, NVIDIA announced Isaac ROS 5.0 at ROSCon. The company describes ROS Lyrical and Ubuntu 24.04 support, agent-oriented workflows, and…
🧠 AI & Agents
GPT‑6.1 Sol: better agents still need inspectable evidence
OpenAI announced GPT‑6.1 Sol at DevDay. How to read its scientific workflow results, cost claims and relevance to robotics without overstating the evidence.
🧠 AI & Agents
Gemini 4 Argon: Google opens a one-million-token output window and targets cyber defenders
Google DeepMind presented Gemini 4 Argon on 30 September 2026, with an output context window raised to one million tokens and priority access for cyber defenders. Here is what the model actually brings.
🧠 AI & Agents
OpenAI DevDay 2026: Dots are agents that work for you — but not in France yet
Announced on 29 September 2026, Dots are always-on agents with their own computer in the cloud. The catch for European readers: the first rollout excludes the European Economic Area, Switzerland and the United Kingdom.
🧠 AI & Agents
Claude Sonnet 5.5: the cheaper model doubles Opus 5.5 on the agentic coding test
Anthropic released Claude Sonnet 5.5 on 28 September 2026. The most accessible model in the family beats Opus 5.5 on Terminal-Bench 4.0 — but loses on every other published test, and Anthropic says so.
🧠 AI & Agents
OpenAI's AI Research Intern: Announcement, Evidence and Limits
OpenAI says it has reached the automated research-intern stage. Here is its definition, the internal evidence and the limits.
🧠 AI & Agents
World Labs Atlas: impressive 3D worlds, early-access evidence
Atlas generates, reconstructs and simulates 3D scenes from sparse inputs. World Labs shows strong demos and internal benchmarks, but access and independent tests remain limited.
🧠 AI & Agents
Gemini 3.6 Flash: Faster Agents, but Read the Token Math
Google's Gemini 3.6 Flash costs less per output token and often uses fewer of them. That can lower agent bills, but not by a universal 65%.
🧠 AI & Agents
Hugging Face's AI-Agent Breach: What Is Confirmed
Hugging Face says an autonomous agent breached part of its infrastructure. The confirmed facts reveal a data-pipeline and response gap.
🧠 AI & Agents
AI in War: What Is Documented, Claimed and Disputed
AI targeting in Gaza and Ukraine, autonomous weapons, and the UN fight over a ban: what is confirmed, what is only claimed, and what remains disputed.
🧠 AI & Agents
China's open-weight AI race in 2026: who leads and why
DeepSeek, Qwen, Kimi, GLM and MiniMax now ship frontier-class open models at a fraction of Western prices. Here is who leads and what it changes.
🧠 AI & Agents
Kimi K3 vs Claude and GPT: Does It Really Win?
Moonshot AI shipped Kimi K3, a 2.8T open-weight model. We check the real, sourced benchmarks against Claude and GPT—where it leads, and where it trails.
🧠 AI & Agents
Sora vs Veo in 2026: official status, features and what's real
OpenAI's official Sora page confirms the product was discontinued on 26 April 2026. Here are Veo 3.1's verified features and what still holds up.
🧠 AI & Agents
AI and Humans: Five Tasks, Performance and Limits
AI dominates scale and specific benchmarks; humans retain context, responsibility and adaptable judgement.
🧠 AI & Agents
Best free AI tools in 2026: 7 useful options and their limits
Seven genuinely useful AI tools you can start free, with privacy and usage-limit warnings.
🧠 AI & Agents
10 jobs AI will replace by 2030? What the data actually says
Ten roles face strong task automation, but exposure is not the same as an entire job disappearing.
🧠 AI & Agents
AI Agents in 2026: Definition, Architecture, Risks and Real Uses
An AI agent does more than answer: it uses tools in an action loop. Here is how to recognise one, supervise it and avoid empty marketing.
🧠 AI & Agents
The EU AI Act in 2026: What Actually Applies to Companies and Users
Prohibited practices, general-purpose models, transparency and high-risk systems: the European timeline without treating every chatbot as banned.