About this document
Metacognitive Capabilities of LLMS: An Exploration in Mathematical Problem Solving by s9qi1zche is a document available to read on EtoBox.
The paper investigates the metacognitive capabilities of large language models (LLMs) in mathematical problem solving, demonstrating that they can identify and label skills relevant to tasks. By employing a prompt-guided interaction, the authors show that LLMs can cluster these skills into interpretable categories, enhancing their problem-solving accuracy on math datasets like GSM8K and MATH. The findings suggest that this methodology could be applied beyond mathematics to improve LLM performance in various
- Author
- s9qi1zche
- Language
- EN