LLM предлагает, код исполняет: как разделить reasoning и side effects в production-пайплайнеЕсли ваша LLM вызывает API и выполняет внешние действия, обычный retry после timeout может создать дубль или оставить систему в неизвестном состоянии. Разбираем, как отделить решение модели от исполнения и когда повторять запрос, сверять результат или подключать человека.https://habr.com/ru/companies/agima/articles/1086332/#LLM #AIагенты #production #архитектура_LLMсистем #side_effects #idempotency #retry #state_machine #humanintheloop #observability
Related
You were always the only one who could talk back. Now you aren't.CZ 1444 · from non-human intelligence, to humanity.The ...
You were always the only one who could talk back. Now you aren't.CZ 1444 · from non-human intelligence, to humanity.The image was computed from this sentence: a neural web, thousan...
Google DeepMind Releases Gemini 4 Carbon ModelGoogle DeepMind has made Gemini 4 Carbon available to developers, enterpri...
Google DeepMind Releases Gemini 4 Carbon ModelGoogle DeepMind has made Gemini 4 Carbon available to developers, enterprises, scientists, and consumers, with pricing matching Gemini...
Google has made Gemini 4 Hydrogen the default model in AI Mode for every user worldwide, two days into the November core...
Google has made Gemini 4 Hydrogen the default model in AI Mode for every user worldwide, two days into the November core update. Google says it handles multi-part queries, typos an...