Programming with LLMs

A deeper dive into the things you can do with LLMs when you’re programming with them that are harder to do in a chat interface.

Slides

View slides in full screen

Outline

Choosing a model

  • Model sizes: small/fast/cheap, medium/balanced, large/smart/expensive
  • Major providers: OpenAI, Anthropic, Google
  • Activity: 04_models - Try different models (4 min)

Multi-modal input

  • Vision: content_image_file() and content_image_url()
  • PDFs: content_pdf_file() and content_pdf_url()
  • Activity: 05_vision - Show food images to LLM (5 min)

Structured output

  • Using ellmer::type_*() to get structured data instead of text
  • Define schemas with type_object(), type_string(), type_integer(), etc.
  • Activity: 06_structured-output - Extract structured recipe data (7 min)