Exploring
Finance training data
Expert-built datasets that teach AI models real banking work.
- Problem
- AI models learn finance mostly from public text. The work bankers actually do, such as building models, reading merger documents and running comparables, is rarely written down in a form a model can learn from.
- Approach
- Create and sell datasets and evaluations made by people who have done the work, so AI labs and finance firms can train and test models on real tasks.
- Why me
- I've done M&A at Goldman Sachs and Standard Chartered, and evaluated training data at Isidor. I know both what good banking work looks like and what makes data useful for training.
- Next steps
- Talk to AI labs and finance firms about which tasks they most need data for
- Build a first sample dataset and evaluation