Follow-up to our post on world models as code (https://news.ycombinator.com/item?id=49541458). We ran the same Union Square task with GPT-6 Astra: same brief, same test conditions, which one you like more?

The video: https://github.com/PhiloLabs/fable51-worlds/tree/main#%EF%B8...