Curating Proprietary Data to Build THOMSON-1

How do you build a frontier legal AI model without a hyperscaler budget? In this episode, Ari Morcos (CEO, DatologyAI) sits down with Jonathan Richard Schwarz (Head of AI Research, Thomson Reuters) to break down how their teams trained THOMSON-1 — a frontier legal model — using under 400 GPUs and just $450K of compute. A conversation for anyone thinking about efficient model training, data quality, and what's possible outside the biggest labs' budgets.