Karpathy's Autoresearch On My AI Polymarket Trading Bot

2026-03-11 · Source: All About AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Data Science & Analytics · Depth: Advanced, long

Summary

This content details the development and live testing of an automated arbitrage trading bot on PolyMarket, inspired by Andrej Karpathy's auto-research project. The bot operates within a "trading auto-research loop," using a GitHub repository to evolve its strategy. A markdown file, "training program," defines experiment logic, including how strategies are chosen, run, and evaluated. The system updates strategy code, creates new experiments, and tests them on PolyMarket's five-minute Bitcoin up/down market in dry mode, with a brief live test demonstrating a profit of $2 over approximately 20 minutes. Weak experiments are discarded, while improvements are kept, with a confirmation step for strong results due to data noise. The bot successfully executed five arbitrage trades, maintaining a 100% win rate during the live demonstration.

Key takeaway

For AI Engineers developing automated trading systems, consider adapting an auto-research loop to autonomously evolve and optimize your strategies. By defining experiment logic in a "training program" and using GitHub for code evolution and memory, you can continuously improve bot performance and identify profitable arbitrage opportunities, as demonstrated by the $2 profit in 20 minutes.

Key insights

An auto-research loop can autonomously evolve trading strategies for arbitrage bots.

Principles

• GitHub can serve as a memory and evolution platform for agent-driven code.
• Experiment logic should be codified in a "research playbook" for autonomous agents.

Method

An agent updates strategy code based on a markdown-defined "training program," tests it in a live environment, evaluates results, and iteratively refines the strategy, committing changes to a GitHub repository.

In practice

• Implement a confirmation step for strong results in noisy data environments.
• Use dry runs for strategy testing before live deployment to avoid financial loss.

Topics

• Auto-research
• Algorithmic Trading
• Arbitrage Bots
• Poly Market
• Strategy Evolution

Best for: AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• Building a Polymarket AI Trading Bot From Scratch — AI coding agents can rapidly develop and deploy high-frequency trading bots on prediction markets like Polymarket.
• 🔮 Exponential View #565: Autoresearch; the solar supercycle; an agentic nation; ChatGPT Olympian, seeing fraud & moving asteroids++ — Declining solar costs, rapid AI agent adoption, and AI-driven research automation are reshaping global energy and technology landscapes.
• The Future of Automated Trading With AI Trading Bot Forex — Modern AI Forex trading bots utilize reinforcement learning, LSTM, and NLP to adapt strategies dynamically, learning from data and market sentiment.
• Improve Your Agentic AI Trading With a Great Data Pipeline — High-quality, diverse data pipelines are crucial for successful agentic AI trading, outweighing model sophistication.
• My Adventures Building an AI Money Machine — AI can enable arbitrage in "juicy" prediction markets by exploiting inefficiencies.
• EquiLibre Technologies Raises Series A at €438M Valuation to Scale AI Trading Agents - The Recursive — Reinforcement learning-powered AI agents can autonomously trade billions in highly liquid financial markets.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by All About AI.