Can Skills Learned in Games Transfer to Real-World Work?
Good Start Labs trains models in strategy games like 1830 and Diplomacy, showing terminal-agent training transfers to financial research benchmarks.
Good Start Labs, spun out of Every with $3.6M from General Catalyst and Inovia, trains AI models in verifiable strategy games. A 30B model trained as a multi-turn terminal agent in 1830: The Game of Railroads and Robber Barons improved Finance-Agent benchmark performance, while single-turn QA training did not transfer. The founders also co-authored COS-PLAY, a paper on co-evolving LLM decision and skill-bank agents for long-horizon tasks.
- $3.6M-funded startup trains models in verifiable strategy games
- Multi-turn terminal-agent training in 1830 improved Finance-Agent benchmark
- Only agentic training design transferred skills outside the game
- COS-PLAY paper introduces learnable skill bank for long-horizon tasks
From the original article
6 min read at latent.space
Good Start Labs trained an AI on a railroad game — and one version improved at financial research. The difference was the training design.
By Richard MacManus · latent.space
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.