Do you use AI, agents, etc?
We keep our approach deliberately simple and are selective about adding complexity. Our strategies rely on a handful of high-quality signals. Every attempt so far to incorporate AI or agents in search of additional alpha has only degraded performance. In several instances, statistical analyses produced by leading LLMs have been outright wrong — missing nuances and misreporting failures as successes.
Where AI has proven genuinely useful is in the rapid generation of code to test new hypotheses. Over the past year, we have aggressively optimized our backtesting pipeline, cutting its runtime from 48 hours to 24, then to 6, then to 1, and finally to 2 minutes — on the exact same 13th-generation Intel Core i7 processor.