Affine is an incentivized reinforcement learning environment that rewards miners for making incremental improvements on reasoning tasks like program abduction and coding. The mechanism is designed to be sybil-proof, decoy-proof, copy-proof, and overfitting-proof, ensuring fair competition. Validators evaluate models deployed on Chutes (SN64) looking for the pareto frontier - the model that outcompetes all others across all environments.
0.00This subnet is currently safe from deregistration.
No contribution data available
Yuma Pulse™
RL Incentives
First directed incentive mechanism for reinforcement learning with verifiable improvements
Anti-cheat Design
Sybil-proof, decoy-proof, copy-proof, and overfitting-proof mechanism prevents gaming
Pareto Frontier
Winners-take-all system rewards models dominating the performance frontier across all tasks
Chutes Integration
Models deployed on SN64 Chutes for inference load balancing and public availability