Context and Layers in Harmony: A Unified Strategy for Mitigating LLM Hallucinations

Large language models, despite their strong performance, frequently produce hallucinated content due to excessive reliance on pre-trained knowledge while insufficiently integrating newly provided context. We introduce LACD, a technique that dynamically rebalances probability distributions across lay...

Full description

Saved in:
Bibliographic Details
Main Authors: Sangyeon Yu, Gyunyeop Kim, Sangwoo Kang
Format: Article
Language:English
Published: MDPI AG 2025-05-01
Series:Mathematics
Subjects:
Online Access:https://www.mdpi.com/2227-7390/13/11/1831
Tags: Add Tag
No Tags, Be the first to tag this record!