Exploring the limits of large language models as quant traders