Logical inference is a fundamental aspect of intelligence, encompassing induction (e.g., Raven’s IQ test), deduction (e.g., Sherlock Holmes’ reasoning), and abduction (e.g., scientific discovery). This project seeks to explore and enhance the logical inference capabilities of large language models (LLMs) and multi-modal large language models (MLLMs) across diverse datasets, environments, and task scenarios. Additionally, we aim to diagnose the rationales behind the models’ decision-making processes in logical inference. This project is highly research-oriented, and it requires applicants to possess strong self-motivation and determination.