“`html
Amid growing concerns about hallucinations in AI-driven legal tools, Paxton AI has published results from a benchmarking study that highlight the reliability of its legal research tool. The data indicate an impressive 93.82% average accuracy in legal research tasks, based on benchmarks developed by researchers at Stanford University.
The study evaluated Paxton’s performance using a subset of 1,600 tasks from Stanford’s broad dataset of 750,000 tasks, which scrutinize the ability of different AI models to provide accurate legal interpretations without hallucinations. The detailed results are available on Paxton’s GitHub repository for independent review.
Complementing its benchmarking results, Paxton introduced a new feature: the Confidence Indicator. This tool assesses the reliability of AI-generated responses, providing users with confidence levels categorized as low, medium, or high. Unlike generic LLM-generated confidence scores, Paxton’s Confidence Indicator evaluates responses based on criteria such as contextual relevance, provided evidence, and query complexity.
An example query illustrates how the Confidence Indicator works. When a user submitted a vague query about family law in both New York and Pennsylvania, the tool indicated low confidence and suggested refining the query for better accuracy. After revision, the tool’s confidence level increased to medium with the new detailed query, and eventually to high once the query clearly specified jurisdictions and legal considerations.
“The Paxton AI Confidence Indicator improves the user experience by quickly showing the confidence and reliability of Paxton’s AI-generated responses,” the company stated. “This feature will help speed up decision-making by providing a transparent assessment of response quality.”
Paxton is offering a seven-day free trial of its product, which includes access to the Confidence Indicator, followed by subscription plans starting at $79 per month per user. For more information, visit LawNext.
“`