Tula Masterman
I'm a researcher in Artificial Intelligence focusing on language models and AI safety. My work concentrates on developing lightweight and efficient methods for content safety classification and understanding AI agent architectures.
My recent research has focused on developing novel techniques for content safety and prompt injection classification using Layer Enhanced Classification (LEC). This work demonstrates how we can leverage intermediate transformer layers of language models to create more efficient and effective classification systems. I'm particularly interested in finding ways to make AI systems both more efficient and safer through innovative architectural approaches.
I've also been conducting extensive research on AI agent architectures, examining how different systems approach reasoning, planning, and tool execution. This work aims to better understand the landscape of emerging AI agent implementations and their capabilities. Through this research, I'm working to identify key patterns in agent design and communication that can inform the development of more robust AI systems.
Publications
Lightweight Safety Classification Using Pruned Language Models
Mason Sawtell, Tula Masterman, Sandi Besen, Jim Brown
arXiv.org 2024
The Landscape of Emerging AI Agent Architectures for Reasoning, Planning, and Tool Calling: A Survey
Tula Masterman, Sandi Besen, Mason Sawtell, Alex Chao
arXiv.org 2024