Winning the Virtual Worlds Hackathon

Generative AI, Multiplayer Prompting & Persistent Worlds

My team recently won the virtual worlds hackathon that took place during the a16z Tech Week. I’ve written a couple previous articles about some of the techniques we used (linked in the show notes below). This is a video discussion with Ali El Rhermoul, CTO of Beamable, about what we built—and what we discovered along the way.

Show Notes

To learn more about the semantic programming techniques used in the project, read Semantic Programming and Software 2.0.

For a little more on the multiplayer prompting techniques we developed, check out Multiuser Prompting Architecture with Generative AI.

Two of the technologies we used in the hackathon included Scenario (for 2D art) and Blockade Labs (their generative skyboxes, for immersive environments). I previously had conversations with folks from each company:

• Blockade Labs: Skybox AI with Adam B. Levine

• Scenario: Generative Art Assets for Games with Emmanuel de Maistre

I mentioned Greg Kamradt’s vector database “chat with a book” tutorial (link).

I also mentioned Hilary Mason’s observations of how hallucination is a feature of language models when it’s used for creating fictional narratives. You can find that conversation here: Generative Games with Language Models | Hilary Mason.

Lastly, I mentioned the work we had done back at Disriptor Beam with Jason Booth, building procedural content systems for Star Trek Timelines.