Benchmarking Llama2, Mistral, Gemma and GPT for Factuality, Toxicity, Bias and Propensity for Hallucinations
This work addresses the need for robust safety evaluation in enterprise applications of LLMs, though it is incremental as it builds on existing benchmarking efforts with new datasets.
The paper introduced fourteen new datasets to evaluate the safety of large language models in enterprise tasks, finding that while OpenAI GPT excelled across all safety metrics, open-source models like Llama2, Mistral, and Gemma showed varied performance with specific weaknesses such as high hallucination rates or poor toxicity handling, and safety degraded significantly in multi-turn conversations except for Mistral.
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.