MCG Live · December 19, 2025 · 41 min
REPPO is a prediction market platform for crowdsourcing and verifying AI training data using economic incentives, founded by Filecoin veterans. The founder describes it as a 'gas station' where humans provide real-world context to fuel AI agents, with users earning rewards by locking REPPO tokens to vote on content quality across various domains from spam detection to fashion retail.
Founder spent nearly three years at Filecoin Protocol Labs building ML/AI models for fraud detection on the network.
The core unlock: REPPO enables measurement of strength of preferences through a ve token mechanism where people lock tokens to capture how strongly they feel about something.
Non-zero-sum reward model: everyone who publishes content gets paid something, even if they receive only one vote, instead of winner-take-all dynamics.
Commit-reveal mechanism: voters commit votes in first 48 hours then reveal them to prevent front-running, with 50% of emissions going to creators and 50% split among voters.
VE tokenomics: voting power is proportional to how much and how long people lock tokens; 50,000 REPPO emitted every 48-hour epoch, with emissions halving yearly over four years for 1.4% total supply inflation.
Team revenue comes from 0.7% of trading fees daily (earning 10-12k on good days), paid in cBTC; built with minimal token sales and half-million dollar treasury.
Private subnets are seeded by companies like Agent Legit (spam detection startup), which market-buy REPPO and deposit it to incentivize publishers and voters instead of mining emissions themselves.
Q1 roadmap includes launching repo.exchange (decentralized data marketplace for buying/selling data created on REPPO) and publishing world's first RLHF dataset on Hugging Face crowdsourced via prediction markets.
Long-term vision: REPPO as a 'gas station' where AI agents and humans plug in—agents need novel data, humans provide real-world context, creating new income source for both.
This summary was written from the recording's transcript by a language model. For the exact words, ask the MCG Live archive: every answer there is quoted word for word and linked to the second it was said. Market Bubble Search is independent and not affiliated with the show. Nothing here is advice.