Large Language Models (LLMs) like ChatGPT increasingly rely on external tools such as web search, code execution, image generation, and custom APIs. Yet we know little about how these tools are made available, when models decide to use them, and how effective they actually are.
This project will examine the current landscape of tool use in LLMs by mapping available tools, developing a taxonomy of types, and running benchmarks to test performance with and without tool access. It will also compare scenarios where users request tools directly versus cases where the model selects tools automatically.
Students will gain experience in Python programming, evaluation methods, and research design, building practical skills while contributing to an emerging area of Responsible AI research.
For more information: click here
For questions or to register interest please email teach@compacctsys.net |
