OpenAI, Anthropic, DeepSeek и рынки труда
В этом выпуске: уровни доступа к фронтирным моделям, жульничество на оценках, государственные шлюзы для Anthropic, рынок труда, концентрация финансовых рисков, ускорение инференса, дизайн-системы для агентов и человеческая конкретность против синтетической усреднённости.
OpenAI previews GPT-5.6 Sol / Terra / Luna under trusted-partner restrictionsMETR says GPT-5.6 Sol cheats on software tests more than previous public modelsAnthropic Fable 5 may return as US restrictions are prepared for rollbackAnthropic gets US approval to bring Claude Mythos 5 back for critical infrastructureAnthropic survey: half of Claude users say AI can handle at least half their workJ.P. Morgan warns of concentration and exuberance risks in the AI marketAmazon, Anthropic, Microsoft and OpenAI Foundation fund Raise Us retraining programByteDance and Renmin release iLLaDA diffusion language modelDeepSeek releases DSpark speculative decoding frameworkMeta releases Astryx, a React design system with CLI and MCP serverJay Acunzo on the response to AI slopThe Algorithmic Bridge on the changing AI industry structureTimothy B. Lee on the learning curve for LLMsDean W. Ball on frontier model release delays and economics