You can train a Bittensor subnet model to learn and improve over any useful computation that can be precisely measured ... then drive its production through a pemissionless free-market mechanism -- incentivizing continuously improvement.

Bittensor upgrades open source itself.