ИИ для самых маленьких: как работает LLM и ИИ-агент

0:00:00 Let's raise the level of understanding of the topic of AI! 0:01:10 LLM is just a very long mathematical function 0:03:23 Prompt, tokenization, model dictionary 0:05:52 Why a token ≠ a word 0:06:19 Embedding vectors 0:08:15 Model context window 0:10:01 The LLM function's job is to predict the next token 0:11:28 Why LLM always returns a probabilistic result 0:13:35 The function returns the probability distribution of the next token 0:14:37 English is more efficient than Russian in terms of tokens 0:15:22 500 tokens of an LLM response are 500 function calls 0:16:02 Reasoning, reasoning mode 0:16:33 LLM weights 0:17:04 Base and instruct models 0:17:54 LLM weights 0:18:55 What is inference? 0:19:07 AI agents 0:22:40 My AI agent for accounting documents 0:27:44 The agent can extract details even from photographs 0:29:55 "Yes, you can extract details using regular expressions without LLM" 0:32:03 Validating received details with deterministic code 0:32:31 Agent = deterministic code + LLM 0:33:03 What's more important—quality LLM or quality agent? 0:35:03 What is a high-quality AI agent? 0:36:53 Running LLM on a phone 0:39:47 AI agent for book reviews with LLM on a phone 0:43:23 Are local models just toys? 0:47:24 Local models don't store context? 0:48:49 Do local models have a too narrow context window? 0:51:35 Are local models useless because they write bad code? 0:54:55 Paid LLMs are inexpensive, why bother with local ones? 0:58:55 Local LLMs don't have Structured Output? 0:59:56 Are quantized models terrible? 1:00:24 Is 32GB of RAM on a computer prohibitively expensive? 1:01:24 Is an LLM that takes longer than a couple of seconds to generate a response garbage? 1:03:17 Should a local LLM fit into VRAM? 1:04:01 Will LLMs replace programmers? The video mentioned at the end is    • Тебе врут про вайбкодинг в 2026. Разбираю ...   My "Hardcore Web Development" course is https://course.to.digital Telegram channel is https://t.me/t0digital VK is https://vk.com/digitalize.team RuTube is https://rutube.ru/channel/24802975/ab... Zen is https://dzen.ru/id/6235d32cb64df01e6e...

ИИ-агенты — вот что действительно изменит разработку. Пишем ИИ-агент на Python, LangChain и GigaChat
▶︎

ИИ-агенты — вот что действительно изменит разработку. Пишем ИИ-агент на Python, LangChain и GigaChat

Тебе врут про вайбкодинг в 2026. Разбираю хайп на цифрах
▶︎

Тебе врут про вайбкодинг в 2026. Разбираю хайп на цифрах

Paid LLM vs. Local Model on a Home Laptop – Is There a Big Difference?
▶︎

Paid LLM vs. Local Model on a Home Laptop – Is There a Big Difference?

Local AI Agents - EVERYTHING You Need to Know: Parameters, Hardware, Speed, Privacy
▶︎

Local AI Agents - EVERYTHING You Need to Know: Parameters, Hardware, Speed, Privacy

Цепи Маркова — математика предсказаний [Veritasium]
▶︎

Цепи Маркова — математика предсказаний [Veritasium]

Writing browser e2e tests with local qwen3.6 and the pi development agent
▶︎

Writing browser e2e tests with local qwen3.6 and the pi development agent

Как AI скамит весь мир / Топ манипуляций твоим страхом
▶︎

Как AI скамит весь мир / Топ манипуляций твоим страхом

AI for IT Beginners — Lecture at MIREA
▶︎

AI for IT Beginners — Lecture at MIREA

THE COLLAPSE OF THE GERMAN AUTO INDUSTRY
▶︎

THE COLLAPSE OF THE GERMAN AUTO INDUSTRY

How AI is Changing Development in 2026: Key Insights from Major IT Conferences / Kirill Mokevnin
▶︎

How AI is Changing Development in 2026: Key Insights from Major IT Conferences / Kirill Mokevnin

Roman Yampolskiy: ARTIFICIAL INTELLIGENCE ALREADY UNDERSTANDS EVERYTHING
▶︎

Roman Yampolskiy: ARTIFICIAL INTELLIGENCE ALREADY UNDERSTANDS EVERYTHING

60 минут ПЕРЕВЕРНУТ ваше представление о СОЗНАНИИ – философ Антон Кузнецов
▶︎

60 минут ПЕРЕВЕРНУТ ваше представление о СОЗНАНИИ – философ Антон Кузнецов

I Wrote My Own Claude Code in 99 Lines. Meet chebupelka!
▶︎

I Wrote My Own Claude Code in 99 Lines. Meet chebupelka!

Critical LLM knowledge in AN HOUR! Everyone should know this.
▶︎

Critical LLM knowledge in AN HOUR! Everyone should know this.

Why Google Just Gave Away Gemma 4 for Free
▶︎

Why Google Just Gave Away Gemma 4 for Free

Why does AI even work?
▶︎

Why does AI even work?

Stanislav Krapivnik: Russlands Wut kocht über – Steht ein EU-Russland-Krieg bevor?
▶︎

Stanislav Krapivnik: Russlands Wut kocht über – Steht ein EU-Russland-Krieg bevor?

The tale of how Qwen and I matched messy Yandex Delivery data to FIAS
▶︎

The tale of how Qwen and I matched messy Yandex Delivery data to FIAS

Lecture by AI professor at Oxford
▶︎

Lecture by AI professor at Oxford

How to Run AI on a Computer Offline and for Free: A Complete Guide to Local Models
▶︎

How to Run AI on a Computer Offline and for Free: A Complete Guide to Local Models