Auditing LLMs and Twitter
Data Skeptic
00:00
Graph-Based Insights in LLM Auditing
This chapter examines the transition from graph work in distributed systems to auditing large language models (LLMs). The speakers highlight the challenges of bias detection and transparency in AI while exploring LLMs' ability to respond to graph-related prompts, particularly the Karate Club. They introduce the Graph Atlas Distance as a new metric for evaluating LLMs' performance in generating graph representations and discuss the complexities surrounding prompt engineering and error diversity among models.
Transcript
Play full episode
