Large Language Models (LLMs) for high-stakes domains like finance and legal QA often provide brief answers without explanations, reducing user confidence.
Domaino1s introduces a method to enhance LLMs' reasoning abilities in high-stakes domains through supervised fine-tuning and tree search.
Domaino1s utilizes CoT-stock-2k and CoT-legal-2k datasets for fine-tuning, activating domain-specific reasoning steps.
Selective Tree Exploration and PROOF-Score metrics are proposed to improve model performance and explainability in stock investment and legal reasoning tasks.