Decentralizing rewards in Reinforcement Learning from Human Feedback (RLHF) is the missing piece to making AI—and AGI—scalable and affordable. Today’s AI training pipelines rely heavily on centralized, high-cost methods to collect and process human feedback. This approach not only limits scalability but also creates economic inefficiencies that could stifle progress toward AGI.

Consider Deepseek: it built a competitive alternative to ChatGPT for just $5.6 million— a fraction of OpenAI’s reported development costs. This achievement stands in stark contrast to the hundreds of millions of dollars reportedly required to develop OpenAI’s state-of-the-art language models. Deepseek’s success underscores a pressing need: How can we make cutting-edge AI—and ultimately AGI—both affordable and accessible?

Why Decentralized Rewards Matter for RLHF

The answer lies in decentralization. By distributing the tasks of providing, evaluating, and rewarding human feedback across a global network of contributors, we can dramatically reduce costs while increasing efficiency. By distributing the workload globally, decentralized RLHF turns AI training into a collaborative, incentive-driven system where anyone—from casual users to specialists—can contribute and be fairly rewarded.

A decentralized rewards mechanism aligns incentives more effectively than centralized systems. Instead of rigid quotas or predefined benchmarks, contributors are rewarded based on measurable improvements to the AI model. Blockchain and smart contracts can automate fair compensation, reducing bureaucracy and ensuring accountability.

Scaling AGI Development Affordably

The need for scalable, cost-effective solutions has never been greater. The U.S. government’s $500 billion Stargate Project aims to accelerate AGI with massive computing power—but it still relies on traditional, centralized data collection, limiting scalability. Even with vast computational resources, the reliance on expensive and top-down methods to gather diverse, high-quality training data will struggle to keep pace with the growing complexity of AGI development.

Decentralized RLHF offers a way to overcome this bottleneck. By opening participation to a global audience, it ensures a more diverse pool of feedback, reducing cultural and systemic biases. It also accelerates innovation, as contributors from different backgrounds can bring fresh perspectives that enhance the model’s general intelligence. This democratization of the RLHF process is critical for creating AGI systems that truly reflect the diversity and complexity of human experience.

The Path Forward

Decentralized rewards for RLHF could be the catalyst that enables scalable and affordable AGI. By lowering the cost barriers, this approach democratizes access to AGI development and fosters competition, driving innovation at an unprecedented scale. In a world where AI systems are poised to influence economies, governance, and personal lives, decentralization is no longer just a technical option—it is a societal imperative.

Deepseek’s efficiency and the Stargate Project’s ambition highlight two paths toward AGI: one lean and decentralized, the other massive and centralized. Only decentralized RLHF can make AGI truly accessible and scalable. While centralized systems have driven significant progress, they cannot scale affordably to meet the demands of AGI. Decentralized RLHF, powered by a robust rewards mechanism, is the key to unlocking the next phase of AI innovation.