Bridgewater fine-tunes a model on its investors' judgment
Working with Thinking Machines, Bridgewater fine-tuned an open-weight model on examples labeled by its investment experts. It averaged 84.7% on six internal research tasks versus the mid-to-high 70s for frontier models, at 13.8x lower cost per task.
Sources
- ai-street.co ↗ · Jul 2026
More use cases like this
Hedge fundsOpen-source earnings-season Claude/Codex skillHedge fundsCitadel's agent reproduces 8 weeks of PhD research in 2 hoursHedge fundsEx hedge-fund analyst turns 5-hour earnings preview into 20 minutesHedge fundsHow Citadel, Balyasny and Point72 use AI
Get The Brief: the best finds, weekly, every Friday