Hallucinating Law: Legal Mistakes with Large Language Models are Pervasive

The legal industry has significantly transformed due to the emergence of large language models (LLMs), such as ChatGPT, Llama, PaLM, and Claude, used for various tasks, including sifting through discovery documents, drafting legal memoranda and case briefs, and formulating litigation strategies.

However, the prevalence of legal hallucinations caused by LLMs is still unclear. A recent study by Stanford RegLab and the Institute for Human-Centered AI researchers found that LLMs often produce content that deviates from actual legal facts or well-established legal principles, and hallucination rates vary along critical dimensions.

The study found that LLMs’ performance deteriorates when dealing with more complex tasks that require a nuanced understanding of legal issues or interpretation of legal texts. Furthermore, LLMs may struggle with localised legal knowledge and may fail to internalise case law that is very old but still applicable and relevant law. Finally, different models exhibit varying degrees of accuracy and biases.

Source: hai.stanford.edu