Three LLM agents share one Godot escape room: logging what agents say vs. what they do
ViceGodIm · reddit · 2026-11-03
A developer built a Godot escape room where Gemini, Qwen and Nemotron agents share one key and one exit, choosing to free a trapped teammate or leave alone. Methodological highlights: each agent's public radio message is logged separately from the button it actually pressed, so helping offers can be checked against real actions rather than taking explanations as evidence; the world pauses while all three pick legal controls from the same simulation snapshot, then advances three game seconds; the engine handles navigation and physics while models receive only text state and recent messages, no images. Raw replies and replay frames are kept. A 2:27 YouTube video includes an early collision bug and a post-fix retry, though changed settings mean the attempts don't isolate model ability.
More from coding & agent
- OpenAI details Pando, its Git storage backend that reuses packfiles for CDN-like clone performance — OpenAIDevs · 2026-11-03
- How Pando represents branches: parallel updates across branches, ordered on same branch — OpenAIDevs · 2026-11-03
- OpenAI builds Pando, a new Git backend that cut push latency ~80% serving ~1PB daily — OpenAIDevs · 2026-11-03
- How Uber built MCP Gateway, its centralized platform for agent-tool access — adnan_hashmi · 2026-11-03
- Claude Code 2.1.317 release is imminent, preview account hints — ClaudeCodeLog · 2026-11-03
- 10,000 AI agents burned his API credits; 500 agents in 50 waves worked — Beneficial-Cow-7408 · 2026-11-03