I'm skeptical frontier LLMs can actually do well (e.g. alpha 5%+) without fine-tuning, especially on historical market data. Presumably you support fine-tuned models?