What are They Thinking? Delineation, Probing and Tracking of Concepts in LLMs

Mohamed Abdelwahab, Michelle Yu Collins, Sihan Chen, Yi Cheng Zhao, Jiading Zhu, Soliman Ali, Jonathan Rose · 2026

Read What are They Thinking? Delineation, Probing and Tracking of Concepts in LLMs by Mohamed Abdelwahab, Michelle Yu Collins, Sihan Chen, Yi Cheng Zhao, Jiading Zhu, Soliman Ali, Jonathan Rose (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-28823.txt.

Description

Mohamed Abdelwahab and seven colleagues take first steps toward probes detecting abstract concepts inside LLM embeddings, revealing model thinking. Such probes should be easy to apply to any model, so many concepts can be monitored during operation. The work defines three key tasks: delineating a concept, training and testing linear probes, and tracking concepts. Delineation builds a set with the concept present then absent. Probing detects the concept on any layer, exploring probe complexity. Finally they show such probes can track concepts across contexts, on four concepts and three models.

Catalog record

CREATORMohamed Abdelwahab, Michelle Yu Collins, Sihan Chen, Yi Cheng Zhao, Jiading Zhu, Soliman Ali, Jonathan Rose
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE85469
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.28823

Browse the full library · About the Tower of Babel