AFFINE
Affine
Live
Back to Blog
Dec 12, 2024·8 min read

Introducing Affine: Mining Open Reasoning

Affine is an incentivized RL environment that pays miners who make incremental improvements on a set of tasks. Learn how we are commoditizing reasoning and breaking the intelligence sound barrier.

The Dawn of Incentivized Reasoning

We are excited to announce Affine, a groundbreaking incentivized reinforcement learning environment that pays miners who make incremental improvements on a set of tasks such as program abduction and coding.

For the first time, we have created a mechanism that is:

  • Sybil-proof — resistant to fake identities gaming the system
  • Decoy-proof — resistant to misleading or low-quality submissions
  • Copy-proof — resistant to simple copying of existing solutions
  • Overfitting-proof — resistant to gaming the evaluation metrics

Why Affine Matters

Directed incentives for reinforcement learning have never been achieved at scale. The ability to direct intelligence and aggregate the work effort of a large, non-permissioned group of individuals on RL tasks will unlock rapid advancement in intelligence.

Traditional approaches to AI development rely on centralized teams and massive compute budgets. Affine flips this model on its head by creating an open marketplace where anyone can contribute to improving models and be rewarded for their efforts.

How It Works

Affine validators incentivize miners to submit models to Subnet 64 on Bittensor (also known as Chutes), where they are inference load balanced and publicly available. These models are evaluated on a set of RL environments, with validators looking for models that dominate the Pareto frontier.

The Pareto frontier represents models that outcompete all other models across all environments. The network uses a winners-take-all mechanism where miners are incentivized to:

  1. Download and study the current best model
  2. Make meaningful improvements
  3. Submit the improved version
  4. Receive rewards proportional to their contribution

Our Vision

We intend to commoditize reasoning — intelligence's highest form — and break the intelligence sound barrier. By creating proper economic incentives for open-source AI improvement, we believe we can accelerate the development of artificial general intelligence in a way that benefits everyone.

Getting Started

Ready to participate? Check out our GitHub repository and join our Discord community to connect with other miners and validators.

The future of AI is open, incentivized, and decentralized. Welcome to Affine.