AI is getting better at writing code, generating images, even holding a conversation. But there’s still one thing it can’t produce on its own: human judgement. In this episode, Tim sits down with RG, founder of Reppo, a protocol using prediction markets to pay people to generate and curate AI training data. Instead of scraping the internet or outsourcing “labelling” for pennies, Reppo turns the whole process into a market. RG breaks down why “belief change” and disagreement are the most valuable signals for post-training and what it means when people can earn meaningful income just by providing feedback. Also in this episode: Why evaluation and context are the human edge in AIHow Reppo’s synthetic prediction market works (commit/reveal, on-chain settlement)Why post-training data is where demand is headed nextWhat “permissionless privacy” looks like for proprietary datasetsHow founders sharpen narratives by getting rejected 100+ timesSubscribe for new episodes every week: ‪@TheLocalWMC‬  Also on Spotify, Apple Podcasts, and wherever else you get your fix.  Follow us on X: https://x.com/TheLocalWMC