No — not directly, and not well as a predictor even indirectly. A language model generates text. It doesn’t observe live prices, doesn’t hold a position, and doesn’t place orders. Anything that lets ChatGPT “trade” is a wrapper that turns its text into API calls, and the quality of the trades is the quality of the text.
Why an LLM is a poor price predictor
It has no live data unless something feeds it, and by the time text is generated the price has moved. It reasons in language, not in the statistical relationships that actually predict short-horizon returns. It’s confident regardless of whether it’s right. And it was trained on the internet, which is a poor source of trading edge.
Asking ChatGPT whether to buy is asking a very well-read person with no market data and no accountability.
What LLMs are useful for
Generating hypotheses. A model that has read every finance paper can propose candidate signals — “does funding rate divergence between venues predict short-term reversal?” — faster than a human researcher can type them. Those hypotheses then get tested statistically, and most fail, but the throughput is higher.
Processing text. News, filings, governance proposals, social sentiment. Turning unstructured text into features a trading model can use.
Writing and reviewing code. Backtesting infrastructure, data pipelines, monitoring.
The division of labour
LLMs propose; quantitative models predict; execution systems act; people set the boundaries. Products that skip the middle two and connect an LLM straight to a wallet are the ones to avoid.