Spaces:

fava-uw

/

fava

like 46

Runtime error

This is an academic research project to understand and fix hallucinations in LLM outputs.

Title: Fine-grained Hallucination Detection and Editing For Language
Authors: Abhika Mishra, Akari Asai, Vidhisha Balachandran, Yizhong Wang, Yulia Tsvetkov, Graham Neubig, Hannaneh Hajishirzi
Author affiliations: UW, CMU, AI2
Abstract:
Large language models (LLMs) are prone to generate diverse factually incorrect statements, which are widely called hallucinations. Current approaches predominantly focus on coarse-grained automatic hallucination detection or editing, overlooking nuanced error levels.
In this paper, we propose a novel task---automatic fine-grained hallucination detection---and present a comprehensive taxonomy encompassing six hierarchically defined types of hallucination. To facilitate evaluation, we introduce a new benchmark that includes fine-grained human judgments on two LLM outputs across various domains, leveraging external web documents. Our analysis reveals that ChatGPT and Llama 2-Chat exhibit hallucinations in 60% and 75% of their outputs, respectively, and a majority of these hallucinations fall into categories that have been underexplored.
As an initial step to address this, we train FAVA, a retrieval-augmented LM by carefully designing synthetic data generations and fine-tuning an expert LM to detect and correct fine-grained hallucinations.
On our benchmark, our automatic and human evaluations show that FAVA significantly outperforms ChatGPT
on fine-grained hallucination detection by up to 38%, though a large room for future improvement still exists.
FAVA's suggested edits can also improve the factuality of LLM-generated text, resulting in 5-10% FactScore improvements.