Odd Lots How Hudson River Trading Actually Uses AI
332 snips
Oct 31, 2025 
Iain Dunning, Head of AI Research at Hudson River Trading and former DeepMind researcher, dives into the innovative use of AI in market making. He explains how large-scale machine learning is transforming short-term price predictions. Dunning highlights the challenges like labor and power constraints facing the field and contrasts new AI models with traditional approaches. He also touches on the role of data in trading, the complexity of neural networks, and how firms manage risks while pursuing market efficiency.
AI Snips
Chapters
Transcript
Episode notes
Trading As A Market Service
- Hudson River Trading views itself as a service provider to markets, primarily market making across many instruments.
- They profit by offering tight prices and capturing small spreads at scale using automated systems.
Models Overtook Handcrafted Trading Rules
- HRT shifted from handcrafted features to large-scale neural models trained on vast market data starting around 2013–2014.
- These models overtook traditional feature engineering by consuming internet-scale event data and finding patterns humans miss.
Market Events Are The Core Data
- Market event data (quotes, trades, order book changes) is internet-scale and often more valuable than alternative data for short horizons.
- Treat market events as tokens and scale models to consume them rather than over-engineer features.