Anatomy of a conversation

Programmatically conversing with LLMs

Slides

View slides in full screen

Outline

• (10m) Think empirically, be pragmatic
  • Getting into the right mindset for working with LLMs
  • How to approach this workshop
  • Extra time to work with participants on setup if needed
• (20m) Anatomy of a conversation
  • The same conversation from R and Python with ellmer and chatlas
  • Message roles: system, user, assistant
  • Your Turn _exercises/02_conversation: same question with and without a system prompt
  • How does the LLM remember the conversation? _demos/03_clearbot
• (20m) How do LLMs work?
  • How to Talk to Robots slides
  • Tokens as the fundamental unit
  • Example: _demos/04_token-possibilities
• (20m) Shinychat basics
  • live_console() and live_browser() (or chat.console() and chat.app())
  • Your Turn _exercises/05_live: write a roast of Hadley Wickham with live_console()
  • Making your own shinychat app: chat_app() for a quick app, page_chat() + chat_server() in R or Chat(client=) in Python for an app others can use
  • Your Turn _exercises/06_word-games: 20-questions-style game — the model picks a word, you guess