Over the past few years, there has been a lot of talk about AI making human jobs obsolete, especially white-collar work. Tools like Grok and ChatGPT can be used to generate computer code, blog posts, and marketing material, putting many jobs at risk.
But could AI-powered robots eventually replace manual laborers too? In today's article, we're going to discuss REPPO, a new AI training data company that implements a prediction market to better train the AI models that power robotics.
AI Training Limits
In his article on X, the founder of Reppo explains how he previously worked with AI systems at the decentralized storage project Filecoin.
Despite having access to unlimited compute and smart people in the Filecoin ecosystem, the team lacked "high-quality human data about how the world behaves when the answers aren’t obvious".
At the end of the day, human judgement is still key to AI training, and necessary for resolving many difficult edge cases.
Prediction Markets
This is where prediction markets (something like Polymarket) come in.
The incentive structure of a prediction market motivates participants to voluntarily place bets on outcomes they can predict based on their knowledge and experience.
Instead of training models on static answers, you train them on how informed humans think about outcomes before they happen and how those beliefs evolve over time.
This kind of data, which "captures uncertainty rather than collapsing it", can be used to better train and evaluate AI models.
Sourcing Expertise
Sourcing genuine expert input in the field, whether from engineers, clinicians, or other professionals, has always been the weak point of traditional AI data pipelines.
A prediction market solves the problem by rewarding the most qualified people for providing accurate answers.
Instead of trying to hire every expert or guess who to trust, Reppo lets expertise reveal itself through performance.
This is especially important in the area of AI-powered robotics, which operate in a world full of edge cases, where systems require constant feedback about expectations, failures, and what should happen next.
This requires a continuous “human-in-the-loop” that makes judgements in real-time. Decentralized prediction markets like Reppo can be used to capture that judgement, and turn it into usable training data for AI.
REPPO Token
The REPPO token was launched on Base (Ethereum Layer 2). In addition to governance, its main purpose is to incentivize participants to make decisions on how data should be judged for training advanced AI systems.
Domain experts stake REPPO on opinion contracts, essentially betting on which data is highest quality. They are putting real capital at risk on their judgment, and rewarded for being consistently accurate.
New Investment
On April 23rd, the Reppo Foundation announced a $20 million strategic capital commitment from Bolts Capital to scale the protocol, and solve the AI training-data bottleneck.
That probably explains why the token surged in price over the past few days.
Along with many other AI tokens, REPPO can be traded on Base via the DEXs Uniswap and Aerodrome. Of course, none of this is financial advice, just speculation on what else could potentially be the future of money.
Until next time...
If you learned something new from this article, be sure to check out my other posts on crypto and finance here on the Hive blockchain. You can also follow me on X or InLeo for more frequent updates.
Sources
- Reppo Founder's Article On X