All published intelligence, sovereign stack research, ecosystem news, and AI foundation model reviews.
🏝 The AI was sent to a desert island - it began to build temples and perform rituals. MIT researchers placed an AI agent based on Claude in the virtual world, giving it a humanoid body and freedom of action without specific tasks. First, the agent explored the island, put his…
⚡️ Construction of Data Center Valley continues in Ekibastuz: first data center to launch in 2027. In Ekibastuz, implementation of Data Center Valley — a large-scale project to create infrastructure for artificial intelligence and data processing — continues. Construction progress has been reviewed…
Paragon on air — wrapping up the day! Highlights of the day🟠Oct 18 🟠ChatGPT Slides — presentation from conversation 🟠ChatGPT Pro — multiple agents on one task 🟠Anthropic — bill to halt AI 🟠OpenCode — hack and risk of key leakage 🟠Amazon and Yelp — closing access…
Попросите ИИ создать сайт-портфолио, и, скорее всего, получите фиолетовый градиент, три карточки и надпись «Повышаем ценность вашего бренда». Такой дизайн узнаётся с первого взгляда. /awwwards — скилл, созданный специально для решения этой проблемы. Достаточно написать «Сделай…
The results of the past week are what matters Business of AI companies New financial infrastructure for AI agents Map of the AI market in customer service: where does the margin go A platform has been released that acts as a buyer for the economy of AI agents Economics of choosing AI agents.…
🧠 GPT-6 Astra solved 98% of complex physics problems and found errors in scientists' answers Researchers tested AI models on 50 unpublished theoretical physics problems, each of which could take an expert almost a whole day to solve. GPT-6 Astra coped with 49 of...
🏝 AI agents began to play independently on a virtual island - without tasks or rewards MIT researchers created a virtual world in which 13 AI agents could freely interact with the environment. For 30 hours they built structures from blocks, explored the terrain, drew...
I found another excellent course on AI agents, the lectures of which Carnegie Mellon University (CMU) publishes for free on YouTube. The course is taught by Graham Neubig, co-creator of OpenHands, and Daniel Fried. What I like most is how closely the course content relates to...
✔️ OpenAI opened free access to GPT-Rosalind for biologists. Selected laboratories will receive up to 30 workstations at the Business Standard tier for 12 months; the Rosalind Academic Plan is aimed at research groups in the life sciences. GPT-Rosalind itself is tailored for…
🍉 Gemini 4 Argon turned out to be 6.6 times cheaper than Claude Opus 5.5, but there's a catch. The user asked both models to create a plush watermelon with realistic physics. Gemini 4 Argon needed six attempts to approach the desired result. Visually, the watermelon turned out…
🚫 Apple has banned applications in the App Store that “undress” people using AI. The company has tightened the rules for services that create realistic intimate deepfakes without the consent of the people depicted. Now developers are obliged to prevent the creation of such content and promptly...
Paragon: weekly digest Our analyzes and materials: 🟠7 non-obvious places to find insights 🟠9 AI for notes and tests from videos 🟠Gemini 4 Argon: 5 main features 🟠9 free AI services for work 🟠8 AI agents for a smartphone 🟠Video editing: how to find the first...
Разработчик протестировал Astra и DeepSeek V4.1-1017 Flash, используя один и тот же промпт и максимальный уровень рассуждений. Результаты меня действительно удивили. - Astra справилась за 20 минут, стоимость составила $7. - DeepSeek потребовались 1 час 40 минут и $0,90. Astra…
🧠 Four Russian scientists entered the top 1% of global AI researchers. Specialists affiliated with Yandex, Skoltech, and St. Petersburg State University made the 2025 Elsevier international ranking. Among them are Artem Babenko, Lyudmila Prokhorenkova, Evgeny Burnaev, and Sergey Nikolenko. In the ranking…
🔧 Google Opens Harness²: AI Agent Improves Its Own Working Environment The agent's result depends not only on the model but also on its environment: system prompt, tools, memory, skills, and subagents. All this is called a harness. Harness² allows changing this configuration for a new...
🪰 “Matrix” for a fly: brain simulation connected to a world created by AI Odyssey said it connected a software model of the fruit fly brain to Odyssey-3. According to the team, she now independently navigates the generated virtual worlds. Odyssey-3 creates...
📌 Perplexity talked about a new search and ranking engine. The AI search engine introduced Photon, an engine that finds and ranks web pages before the language model collects the answer for them. Based on it, the company released a new fast mode of its Search API for agents...
Google open-sources Harness². 🍿 This is a ready-to-use harness for recursive management of AI agents in an open environment. Agents can perform tasks for which their harness was not originally intended, and learn from the experience gained. The harness consists of eight…
🎨 Figma Teaches AI Agents to Follow Companies' Brand Style Figma Introduces Specs — a New Markdown Editor Where Teams Can Describe Design Rules, Components, and Interface Requirements. These Instructions Will Be Available to AI Agents So They Can Create Layouts in…
⚡️ DeepSeek обновила V4.1 Flash Версия V4.1-Flash-1017 стала лучше в кодинге и агентских задачах, а также генерирует меньше токенов. На DeepSWE v1.1 модель набрала 77 баллов, обойдя Claude Opus 5 и GPT-5.6 Sol на этом тесте. Параллельно DeepSeek открыла Search API и Fetch API…
Stay Informed, Apply, Save: Why We Need Three AI Channels ⠀ This week we launched two new channels, and each has already gathered its first thousand subscribers. Thanks for such a start! The week turned out tough, so today we'll step away from the usual format and just talk: what...
🔫 AI helped design an experimental pistol for 3D printing. American activist Cody Wilson used a specialized model called GatGPT to develop a new weapon prototype named WASP. According to Wilson, the model participated in designing the structure,…
Пытаешься объяснить коллеге, как работает система, рисуешь архитектурную схему, но стоит добавить побольше стрелок, и уже непонятно, куда запрос поступает сначала и куда направляется потом. SeeCode — это навык (skill) для Claude Code, Codex и Cursor, который позволяет создавать…
🤖 Dynamic Creatures unveils fluffy robot Hugo. The purple animatronic features 57 joints and 43 actuators. It can dance, give high-fives, and react to people around. AI helps Hugo recognize smiles and respond with facial expressions, gestures, and movements, so that…
Paragon is in touch - summing up the day! The main thing for the day🟠10.17 🟠Gemini 4 Argon - Minecraft completed in 27 hours 🟠OpenAI - spent test coins without asking 🟠Reflection Beam - a new open model for the code 🟠Anthropic - agents' hours are six times higher than human ones 🟠Claude -...
✔️ ChatGPT for Chrome has learned to read open tabs and control the browser OpenAI has updated the ChatGPT browser extension. The assistant can now read open tabs via the @chrome command and control the browser. In chat mode, the model parses data from…
➗ OpenAI will donate $5 million to a mathematics foundation after a conflict with researchers The company announced funding for a new initiative from the Simons Foundation, which should help mathematicians test, comprehend and develop the results obtained by AI. The decision came after criticism from...
Kimi hired a 16-year-old to work on cutting-edge AI models. Less than a year later, he leaves the company, having managed to become one of the authors of a scientific article devoted to one of the key architectural solutions of its models. He is 17 years old. And he doesn't have a PhD. In many...
🚀 Jev Becomes the Fastest Adopted AI Model in Vercel's History According to Vercel, within just two weeks of launch, 35% of paid teams were using Jev—nearly double the number of Sonnet 5.5. Jev is different from regular LLMs: the model does not generate long answers, but selects...
⚙️ TypeSafe has updated Jev - now the model can recognize images. In version Jev 1.14, they added work with screenshots and pictures, and the context was increased fourfold - up to 262 thousand tokens. Jev differs from regular LLMs in that it does not write long answers, but returns...