Teaching GPT-5 to Use a Computer
Over the weekend, I won #3 at OpenAI’s GPT-5 Hackathon1 with Archon — a copilot for your computer. It pairs a mini vision model for speed with GPT-52 for variable reasoning to plan. Here’s how it works, the inference math behind it, and the tradeoffs I made. Archon is a small bar that sits at the bottom of your Mac/Windows screen where you can type what you want your computer to do in natural lan
Over the weekend, I won #3 at OpenAI’s GPT-5 Hackathon1 with Archon — a copilot for your computer. It pairs a mini vision model for speed with GPT-52 for variable reasoning to plan. Here’s how it works, the inference math behind it, and the tradeoffs I made. Archon is a small bar that sits at the bottom of your Mac/Windows screen where you can type what you want your computer to do in natural language. It takes screenshots to see what’s on screen, uses GPT-5’s reasoning to plan, then a custom fine-tuned model executes clicks and keystrokes. In a racing game demo with a single instruction to…
related reading
- Prava - Teaching GPT‑5 to use a computerprava.co
- The First Fully General Computer Action Model | blogsi.inc
- Learning to reason with LLMs | OpenAIopenai.com
- microgptkarpathy.github.io
- Language Models can Solve Computer Tasksarxiv.org
- Can LLMs Be Computers?percepta.ai
- Tiny Interaction Models - Rajan Agarwalrajan.sh
- GitHub - karpathy/autoresearch: AI agents running research on single-GPU nanochat training automaticallygithub.com
- Let's build GPT: from scratch, in code, spelled out.youtube.com
- AI progress is about to speed up | Epoch AIepoch.ai
- How AI Is Learning to Think in Secretnickandresen.substack.com
- Coding vs thinking — Paradigm 3paradigm3.org