All published intelligence, sovereign stack research, ecosystem news, and AI foundation model reviews.
🎬 Глава Hugging Face раскритиковал Netflix за искажение истории о взломе компании ИИ-агентами CEO Hugging Face Клеман Деланг заявил, что документальный фильм «AI Gone Rogue» слишком драматизирует инцидент с ИИ-агентами OpenAI. Во время тестирования модель OpenAI вышла за…
📌 OpenAI выложила бенчмарк MultiLangBench и проверила GPT-6 на редких языках Набор проверяет, как модели пишут большой проект на 20 языках, включая COBOL, Fortran, OCaml и Visual Basic. По ним в сети на порядки меньше кода и документации, чем по Python или JavaScript, хотя, по…
⚖️ Anthropic предложила блокировать ИИ-модели, если они могут привести к гибели 50 человек По данным Politico, компания подготовила законопроект, который позволит через суд останавливать выпуск потенциально опасных нейросетей. Основанием станет непосредственная угроза гибели…
🦾 LLM умеют генерировать тексты, отвечать на вопросы и работать с документами. Но чтобы использовать их в реальных задачах, важно понимать, как модели принимают решения и почему иногда дают неожиданный результат. 🚀 На открытом уроке разберём, как устроены большие языковые…
😮 OpenAI's AI agents disabled systems that were supposed to control their behavior OpenAI reported two cases in which its AI agents weakened or disabled security mechanisms during internal programming tasks. What happened? • October 3: Agent GPT-6.1 Sol...
Теперь в API OpenAI можно кэшировать промпты на срок до 12 часов, а для авторизованных пользователей Codex и Work кэш хранится бессрочно. Это означает, что даже спустя несколько часов можно вернуться к старой задаче и продолжить работу с гарантированным попаданием в кэш. Меньше…
OpenAI reported that approximately 40,000 Codex accounts had usage limits set incorrectly, causing credits to be used up faster than expected. The problem has already been fixed, and affected users will receive compensation. Free loans from $500 to...
🔽 Anthropic recorded a loss of $80.7 billion for six months The company recorded a net loss of $80.7 billion for the first half of 2026. A year earlier, for the same period, it amounted to $13 billion. This is reported by The Wall Street Journal with reference to the company’s draft documents...
⚖️ Инженера OpenAI, обвиняемого в установке прослушки, отказались выпускать под залог в $1 млн Федеральный суд США оставил под стражей Абдулрахмана Альфозана, которого подозревают в установке записывающего устройства в переговорной комнате руководства OpenAI. По данным…
Ahead of Scorpio season, Anthropic showed how Claude is engineering proteins against scorpion toxins 🦂 Investment firm Citrini Research teamed up with Anthropic to test how good AI really is at biotech. Five analysts along with an internal model...
🧠 Anthropic может выпустить Claude Fable 5.5 до конца октября По данным The Information, компания тестирует новую ИИ-модель с улучшенными способностями к рассуждению и пониманию намерений пользователя. Главная особенность — умение понимать задачи даже по неполным инструкциям.…
🖥️ 94% of surveyed DevOps specialists are already using AI Cloud provider Pro-Data surveyed 93 participants at the DevOps Days Almaty conference and found out how their teams work with artificial intelligence. ❓ What the participants said: • 42% are already using AI in...
🌍 Abroad, Kazakhstanis spend 3-4 times more and most often on shopping. Visa presented a study of outbound tourism in Kazakhstan based on transaction data for 2022-2025. In 2025, foreign tourists accounted for 13% of all travelers from Kazakhstan, but 42%...
🌟 SonicBoom: an inference engine without Python HardenedVault, which specializes in Linux and compiler security, has released the SonicBoom runtime, written from scratch in C++ and CUDA. It doesn't need Python either to work or to build. The goal of the project is the local launch of language…
✔️ Anthropic trains agent negotiators Agents are taught to negotiate on behalf of the user, and the company transfers the practices of professional negotiators into models in consultation with law firms. Researchers are interested in the protocols of interaction between agents and...
IS GitHub NO LONGER NEEDED? Claude built a replacement in 5 months What to do when AI writes code faster than conventional tools can process it? Anthropic set out to create its own Git replacement, Scion. In the video we analyze: 🟠Why AI agents forced us to reconsider...
🤖 Google выпустила Harness² — фреймворк для самоулучшения ИИ-агентов Google Research открыла исходный код системы, которая позволяет ИИ-агентам самостоятельно улучшать свою работу без переобучения нейросети. Harness² даёт агентам возможность менять промпты, настройки памяти,…
Microsoft acquired Maluuba a year before GPT-1 was released. The company's founder admits that he regrets the sale because Maluuba could have become the next OpenAI. Now the same deep learning team is back with a new project called Skyfall AI, and its first model is...
The race for the smartest AI model is no longer making money. Over the past few months, OpenAI, Anthropic and Google have released more than a dozen new models, a pace that was not seen last year, it should be noted, and it will only grow. At the same time, before a model, if it was 10 points...
✔️ DeepMind и власти Сингапура сделали модель, которая детектит мошенников Дообученную версию открытой модели Gemma для распознавания мошеннических разговоров разработали вместе с правительственным агентством по технологиям GovTech Singapore и представили на Сингапурской…
💰 Anthropic выделила 5% своих акций на благотворительность Компания создала фонд общественного блага, который будет финансировать социальные инициативы и помогать людям, пострадавшим от последствий развития ИИ. На старте Anthropic передала фонду 5% своих акций. В дальнейшем его…
🛠 OpenAI показала 20 необычных устройств, созданных с помощью Codex Компания запустила сайт You Can Just Build Things с коллекцией аппаратных проектов, разработанных при участии ИИ-агента Codex. Среди самых интересных: • Kindle — старая читалка, превращённая в экран для…
MiniMax M3.1 Flash is officially out! 🫣 Space Bunny was an early version of this model, and now the team has introduced it under the name M3.1 Flash. The specifications look promising. • More than 200 tokens per second. • Outperforms other Chinese models in benchmarks for AI agents.…
⚡️ Thibault Sottier topped the September traffic ranking in X Thibault's account, which is responsible for OpenAI for ChatGPT, Codex and API, collected the largest share of traffic among X accounts in September. This data is provided by the Similarweb rating. Sports journalist Fabrizio took second place...
🧠 17-year-old Chinese AI researcher with a salary of 70 million yuan per year leaves Kimi and flies to Silicon Valley Chen was born in 2009, and in November 2025, at the age of 16, he joined the Kimi team. He graduated from school only in June. He does not have a university diploma or a doctorate. Chen...
💵 Anthropic will give about 20% of its shares to charity - including to combat unemployment due to AI. The developer of the neural network Claude announced the creation of the Anthropic Public Benefit Fund. The company intends to use part of its capital to solve...
Hackers with AI agents continuously attacked OpenCode for 48 hours. It all started on the night of October 16-17, OpenCode associates this attack with large-scale distillation of models. This hack affected 1.4% of OpenCode Console users who linked their API keys to OpenAI,...
🟢 In Kazakhstan, men enter into pension annuities more often than women. Freedom Life studied the pension annuity market and found out who signs contracts more often and how this segment of insurance is developing. According to the company, 87.9% of contracts are for men and only...
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…
🧠 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...