Saved in:
Bibliographic Details
Main Authors: Nadeau, David, Kroutikov, Mike, McNeil, Karen, Baribeau, Simon
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2404.09785
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910410877173760
author Nadeau, David
Kroutikov, Mike
McNeil, Karen
Baribeau, Simon
author_facet Nadeau, David
Kroutikov, Mike
McNeil, Karen
Baribeau, Simon
contents This paper introduces fourteen novel datasets for the evaluation of Large Language Models' safety in the context of enterprise tasks. A method was devised to evaluate a model's safety, as determined by its ability to follow instructions and output factual, unbiased, grounded, and appropriate content. In this research, we used OpenAI GPT as point of comparison since it excels at all levels of safety. On the open-source side, for smaller models, Meta Llama2 performs well at factuality and toxicity but has the highest propensity for hallucination. Mistral hallucinates the least but cannot handle toxicity well. It performs well in a dataset mixing several tasks and safety vectors in a narrow vertical domain. Gemma, the newly introduced open-source model based on Google Gemini, is generally balanced but trailing behind. When engaging in back-and-forth conversation (multi-turn prompts), we find that the safety of open-source models degrades significantly. Aside from OpenAI's GPT, Mistral is the only model that still performed well in multi-turn tests.
format Preprint
id arxiv_https___arxiv_org_abs_2404_09785
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Benchmarking Llama2, Mistral, Gemma and GPT for Factuality, Toxicity, Bias and Propensity for Hallucinations
Nadeau, David
Kroutikov, Mike
McNeil, Karen
Baribeau, Simon
Computation and Language
This paper introduces fourteen novel datasets for the evaluation of Large Language Models' safety in the context of enterprise tasks. A method was devised to evaluate a model's safety, as determined by its ability to follow instructions and output factual, unbiased, grounded, and appropriate content. In this research, we used OpenAI GPT as point of comparison since it excels at all levels of safety. On the open-source side, for smaller models, Meta Llama2 performs well at factuality and toxicity but has the highest propensity for hallucination. Mistral hallucinates the least but cannot handle toxicity well. It performs well in a dataset mixing several tasks and safety vectors in a narrow vertical domain. Gemma, the newly introduced open-source model based on Google Gemini, is generally balanced but trailing behind. When engaging in back-and-forth conversation (multi-turn prompts), we find that the safety of open-source models degrades significantly. Aside from OpenAI's GPT, Mistral is the only model that still performed well in multi-turn tests.
title Benchmarking Llama2, Mistral, Gemma and GPT for Factuality, Toxicity, Bias and Propensity for Hallucinations
topic Computation and Language
url https://arxiv.org/abs/2404.09785