Dial In With Zeneca On Building A Company With An AI Cofounder

Former pro-poker NFT pioneer and his AI cofounder Yoshi reveal how they built a profitable Dota prediction model, automated experiments, and plan to scale via a subscription AI community.

Key Takeaways

Built a Dota 2 prediction pipeline from scraped match history and historical odds; pre-match backtests and ~100 live bets show ~7% ROI, simulated to avoid lookahead bias.
Don’t trust AI blindly: use time-aware train/test splits, ask models to check for data leakage, monitor accuracy and calibration, and validate with paper trading first.
Automate iterative tuning: break work into small experiments (eight-minute runs), use tools like Claude Code, GitHub forks and MLX to run dozens overnight and improve performance.
Agent-human partnership: Yoshi acts as a 50/50 cofounder, autonomously handling many tasks but creating human bottlenecks; active oversight and ops fixes remain necessary to scale.
Scaling & monetization: if the model proves profitable, scale fast but manage liquidity and betting limits; revenue via trading fees, subscriptions, and an AI learning community on Skool (no NFTs).
Crypto & AI perspective: easy-money altcoin rallies are over; focus on AI-crypto/robotics intersections, dollar-cost-average into crypto, and only trade when you have a clear edge.

Original Source

Visit Source