Adaptive Activation Cancellation for Hallucination Mitigation in Large Language Models

Eric Yocam 1 , Varghese Vaidyan 1 , Gurcan Comert 2 , Paris Kalathas 3 , Yong Wang 4 , and Judith L. Mwakalonge 5 · 2026

Read Adaptive Activation Cancellation for Hallucination Mitigation in Large Language Models by Eric Yocam 1 , Varghese Vaidyan 1 , Gurcan Comert 2 , Paris Kalathas 3 , Yong Wang 4 , and Judith L. Mwakalonge 5 (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-2603-10195.txt.

Description

Yocam and coauthors cancel hallucination linked signals inside language models in real time, borrowing from adaptive noise cancellation in signal work. Their frame treats those signals as structured interference in the transformer residual stream. Layer wise probes first locate the nodes tied to false text, then a confidence weighted hook tamps them in generation, with no fine tuning or extra passes. Across three model scales the hook lifts truth scores while prose ease and reasoning stay intact.

Catalog record

CREATOREric Yocam 1 , Varghese Vaidyan 1 , Gurcan Comert 2 , Paris Kalathas 3 , Yong Wang 4 , and Judith L. Mwakalonge 5
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE57464
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.10195

Browse the full library · About the Tower of Babel