Playing Improv with Python!
summary
Plays improv games by wiring local and hosted LLMs into Python: Llama 3.1 (local via Ollama), GitHub-hosted models and gbd-40-mini for vision, using OpenAI-compatible endpoints and the OpenAI SDK. Implements reverse-charades, Pixar story spine, sentence-at-a-time, word-at-a-time, multimodal prop-to-product and a “three things” game that uses function-calling with JSON Schema to force exactly three items. Finds small local models are slower and fail at counting or single-word turns, larger models follow rules better, and function-calling yields reliably structured outputs; code and slides are on GitHub.
AI-written summary · about
chapters (13)
- 0:00Introduction to improv with Python
- 3:00Personal improv journey
- 8:58Types of improv games
- 11:49Story spine with volunteers
- 12:11Playing improv with AI
- 14:00Running LLMs locally with Ollama
- 20:50Story spine with Python
- 24:18Sentence-at-a-time story
- 26:25Word-at-a-time story
- 31:29Improv with props
- 36:53Three things game
- 43:58Lessons learned from mixing improv and Python
- 45:09Q&A session
related talks
27mBuilding Useful Apps With Small, Local LLMs
27mAI Demystified Unveiling the Intelligence Behind the Machine
40mCreating Interactive GenAI applications powered by Streamlit, open source LLMs
01h 34mA Practical Starter's Guide to building LLM based projects
47mRamaLama: Making working with AI Models Boring
37mEmpowering Developers with local AI
46mBeyond Basic Prompting: Supercharging Open Source LLMs with LMQL's Structured Generation
43mLarge Language Models for Devs: from zero to your first LLM application