GPT-4 vs Starcraft II – Strategic Decision Making using Large Language Models
summary
Uses GPT-4 via Azure OpenAI to read a numeric StarCraft II game state (time, supply, resources, units, structures) and return a one‑paragraph strategy plus a 20‑step build order, sent from a Python console that talks to the StarCraft II API + Python console. Chain‑of‑thought prompting improves play and “jumps a couple of levels” versus naive prompts but it still loses on hard and delegates micromanagement to hard rules. Telemetry uses TensorBoard. Also demos function calling and a pizza agent that, without guardrails, orders a million pizzas, and a working but flaky Vibe project‑manager / Vibe developer demo.
AI-written summary · about
chapters (13)
- 0:00Welcome
- 3:00Speaker background
- 8:58Starcraft II with LLMs
- 14:53Prompt engineering
- 25:06Strategic decisions in orgs
- 26:11AI agents and humans
- 31:24Agents and environment
- 34:13LLM limitations and risks
- 38:28Function calling in LLMs
- 41:20Pizza ordering demo
- 47:22Full Stack GPT demo
- 51:03Vibe project management
- 57:10Q&A
related talks
44mGPT-4 vs Starcraft II – Strategic Decision Making using Large Language Models
45mBringing the Power of AI to Your Application
01h 00mUnleashGPT: Mastering the Implementation of Large Language Models
45mHow to use GPT in your projects: A Case Study
43mBuild and Maintain your Company Copilot with Azure ML and GPT-4 Q&A
59mOrchestrate your AI with Semantic Kernel
25mFrom Prompt to Production: Building Autonomous AI Agents in Node.js
48mUsing AI in Software Design: How ChatGPT Can Help With Creating a Solution Architecture