LLM Grounding Analysis

LLM Grounding Analysis

LLM grounding vs. factual recall

The study investigates LLMs using Fakepedia, a dataset presenting contradictions between known facts and new information. Through Masked Grouped Causal Tracing (MGCT), the research deciphers LLMs' grounding mechanisms by contrasting neural activation patterns. Findings help understand the co-functioning of grounding with recall capabilities within LLMs.

Machine LearningNatural Language
Key facts
Maturity
Support
C4DT
Inactive
Lab
Active
  • Technical

Data Science Lab

Data Science Lab
Robert West

Prof. Robert West

Our research aims to make sense of large amounts of data. Frequently, the data we analyze is collected on the Web, e.g., using server logs, social media, wikis, online news, online games, etc. We distill heaps of raw data into meaningful insights by developing and applying algorithms and techniques in areas including social and information network analysis, machine learning, computational social science, data mining, natural language processing, and human computation.

This page was last edited on 2024-05-03.