Mathematical Reasoning in LLMs
Investigating where and why LLMs fail at mathematical reasoning, and whether targeted interventions improve reliability.
Where and why do LLMs fail at mathematical reasoning? This project investigates systematic failure patterns across problem types — arithmetic, algebraic, geometric, and logical reasoning — and explores whether targeted training interventions, prompting strategies, or structured representations can make LLMs more reliable on formal problem-solving tasks.
Team: Ruchi Harge, Snehal Reddy