LLMs Fall Short in High-Frequency Trading: Architect Founder Weighs In

Brett Harrison, founder and CEO of Architect Financial Technologies, is cautioning against relying on Large Language Models (LLMs) for building high-frequency trading systems. In a recent Medium post, Harrison argues that LLMs are fundamentally ill-suited for this task due to their inability to process stochastic market data.

Harrison's critique carries significant weight, given his 11-year tenure at Jane Street leading algorithmic trading system development. He acknowledges that LLMs can be useful for supporting tasks such as code generation and feature selection, but warns against relying on them for core quantitative trading models or real-time operations like continuous market monitoring.

Harrison's background in computer science and his experience at Jane Street make his views particularly relevant. He has also launched Architect Financial Technologies with backing from notable crypto-native investors, including Coinbase and Circle, to build trading infrastructure that blends human expertise with technological innovation.