Key-Augmented Neural Triggers for Knowledge Sharing Journal: Journal of Systems and Software

Alex Wolf, Marco Edoardo Palma, Pooja Rani, Harald C. Gall · 2025

Read Key-Augmented Neural Triggers for Knowledge Sharing Journal: Journal of Systems and Software by Alex Wolf, Marco Edoardo Palma, Pooja Rani, Harald C. Gall (2025) 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-2508-03340.txt.

Description

Understanding a whole code repository is hard because knowledge scatters across files, search pipelines drown models in long context, and fresh training data is scarce. Alex Wolf, Marco Edoardo Palma, Pooja Rani, and Harald C. Gall propose KANT, which plants small knowledge anchors in the code that the model can trigger from inside, at training and at answer time.

Training data is synthesized from the code itself, and anchors replace bulky context, cutting tokens and delay while allowing private, on-site use. Human judges preferred the system's answers over sixty percent of the time, its training data matched what developers actually ask, and gains held across several different language models.

Catalog record

CREATORAlex Wolf, Marco Edoardo Palma, Pooja Rani, Harald C. Gall
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-03340

Browse the full library · About the Tower of Babel