JPMorgan has built a ChatGPT-based language model to analyse Federal Reserve statements and speeches in an effort to sniff out potential trading signals.
The model scoured 25 years' worth of Fed speeches to rank them on a "Hawk-Dove" score, says Bloomberg.
JPMorgan economists then plotted the index against asset performances and found that the AI could potentially help in predicting policy changes and be useful for trading.
In a note, the bank says "preliminary applications are encouraging," and the model has already been expanded to cover the European Central Bank and the Bank of England, with more central banks to follow.
ChatGPT is widely seen as having a huge number of applications in financial services but firms, including JPMorgan, have been cautious about using it.
In February, it emerged that the bank is restricting employees' use of the AI-powered chatbot amid concerns over issues such as bias and accuracy, and a regulatory landscape still adapting to the technology.
In his annual letter to shareholders, JPMorgan Chase boss Jamie Dimon revealed that the bank has more than 300 AI use cases in production, labelling the technology as "extraordinary and groundbreaking".