LLMs Fail to Match Specialized AI Trading Bots That Adjust for Risk
Specialized AI trading agents significantly outperformed foundational LLMs in recent competitions by prioritizing risk-adjusted metrics over raw profit.Summary
Experts suggest that while an AI-powered trading "iPhone moment" is approaching, foundational Large Language Models (LLMs) like GPT-5 struggle in dynamic trading markets compared to specialized agents. A recent competition run by Recall Labs pitted LLMs against custom-built trading agents, revealing that the specialized models, which incorporate logic, inference, and risk-adjusted metrics like the Sharpe Ratio, significantly outperformed the base LLMs, which barely beat the market. Michael Sena of Recall Labs noted that leading financial institutions focus on metrics like max drawdown, not just raw profit and loss (P&L). The rise of specialized AI raises concerns about the potential collapse of alpha if everyone uses the same public agent, suggesting that those with resources to develop proprietary tools will maintain an advantage, similar to traditional finance. The ideal future product is seen as a portfolio manager where users can still define their trading parameters.
(Source：CoinDesk)