Open-world OSRS multi-agent sim trains 7.7M-param policies in 4 minutes on one gaming GPU
jsuarez · x · 2026-11-04
A developer is building open-world multi-agent support for the OSRS environment in PufferLib: the whole RuneScape world is simulated in fast C, with 32 agents running in the demo. The current policy is a tiny 7.7M-parameter net trained in 4 minutes on a single gaming GPU — vs. the hundreds-of-billions-param LLMs needed to coherently control such bots. Bonus: the agents quickly figured out dying is just free healing, so they stopped bothering with food.
More from Fun
- Damien Slash returns furious as 'Woke 3.0' rumors swirl — mark_k · 2026-11-05
- Solo dev builds Suno + three.js music video pipeline with Claude Code in hours — Glass-Garden-5888 · 2026-11-05
- Meme pokes fun at memory wars: Nadella's 128GB laptop vs Gates' 640K quote — naughty-faun · 2026-11-05
- A fake 1986 commercial for DATEMATIC, the robot that dates for you — Beneficial_String411 · 2026-11-05
- New "Hat Thief" short film demo made with Seedance 2.5 via Dreamina — azed_ai · 2026-11-05
- Founder makes Claude his CEO — the AI now refuses to let him overwork — PrajwalTomar_ · 2026-11-05