Each workshop is divided into chapters — watch at your own pace.
"Doing AI" can mean five completely different things — from basic research to end users. A Russian nesting doll analogy then clarifies how AI → Machine Learning → Deep Learning → LLM relate to each other, so you know what these terms actually mean.
What's the fundamental difference between Traditional AI, LLMs, and Agents? And how does an LLM generate text one token at a time through the prediction loop? This is the core mechanism behind all AI behavior.
LLMs don't see "words" — they see tokens. How are tokens billed, and why does Chinese cost more than English? How does temperature control randomness? And most importantly: why does AI fabricate papers? It picks "statistically likely" text, not "correct" text.
Five things AI can't do: real-time information, precise calculation, genuine novelty, long-term memory, objective judgment. More dangerous: it hides its own confidence scores, and the sycophancy problem — the more you push, the more it drifts from the truth.
AI has a limited context window — when it fills up, things get forgotten or compressed badly. The deeper truth: the moment you press send, AI is "born" and disappears after replying. The "conversation memory" is an illusion the app creates by re-packaging your history.
Assign a role, give examples, ask it to think, specify format, put the key point first — five techniques grounded in how LLMs work. A live car wash case study demonstrates the effect of each technique.
Verify outputs, protect data, disclose honestly — what these three principles actually mean. The Samsung leak shows the real data risk of online AI. Finally: US, China, Taiwan — what are the risk levels for AI tools from each camp?
A talk by Anthropic researcher Eric. Explores the model of letting AI handle most development work with developers shifting to a product manager role, and how to significantly boost productivity while maintaining safety and quality through verifiable test metrics and sufficient context.
For: Anyone new to Agents who wants to build the concept from scratch. Starting from two real cases — Klarna's AI customer service and the AI Scientist — this chapter explains the fundamental difference between Chat and Agent, and the three engineering stages behind the evolution.
For: Those with some AI background who want to find the right tool for their work. A full comparison of Agent tools from Google, Microsoft, OpenAI, open-source ecosystems, and Anthropic — with a three-step decision framework based on data sensitivity, task type, and existing tools.
For: Anyone who wants to see Agents in action right away. Covers the difference between Claude Cowork (desktop automation, for non-engineers) and Claude Code (code & research tool, for developers), with real demo scenarios and key things to watch out for when using either tool.
Go to each vendor's tool
A walkthrough of uploading documents, auto-summarization, Q&A with source citations, cross-document analysis, mind maps, and audio overviews. Completely free — all you need is a Google account.