Wall Street Keeps Testing AI Traders, But Most Are Still Underperforming

Wall Street Keeps Testing AI Traders, But Most Are Still Underperforming Recent trading competitions suggest large language models are still unreliable portfolio managers, according to Bloomberg. Tests involving models from OpenAI, Anthropic, Google, and xAI have often delivered underwhelming results: many lost money, traded excessively, and made erratic decisions despite receiving identical prompts. In several cases, models appeared unable to stick to coherent strategies for more than a few trading sessions. Bloomberg writes that one of the clearest examples came from Alpha Arena, a competition created by startup Nof1. Eight models were...