Kruskal–Wallis Tests for NLAT and DISCERN AI Chatbot Evaluations

Meisam Dastani published a dataset on figshare containing statistical test results for evaluating four large language models on medical questions. The dataset includes Kruskal–Wallis test results from an evaluation of ChatGPT, Gemini, Copilot, and Grok using the DISCERN-AI and NLAT-AI assessment tools. The data was last updated on 2026-05-11.

Use Cases

• Compare the statistical performance of AI chatbots based on DISCERN-AI and NLAT-AI scores.
• Benchmark LLM responses for accuracy and transparency in the context of multiple sclerosis.
• Assess model strengths and weaknesses across medical domains like diagnosis, treatment, and disease management.
• Inform the selection of AI models for generating patient-facing medical content.
• Validate assessment tools like DISCERN-AI and NLAT-AI for evaluating AI-generated medical text.

Strengths

• Evaluation is based on 25 specific medical questions across five key domains.
• Compares four prominent, publicly accessible LLMs: ChatGPT, Gemini, Copilot, and Grok.
• Uses established assessment tools, DISCERN-AI and NLAT-AI, for structured evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The dataset is very small at 5.5 KB, indicating limited scope.

Provenance

Source

Meisam Dastani via figshare.

Collection Method

Responses from four LLMs to 25 medical questions were evaluated using DISCERN-AI and NLAT-AI tools.

Freshness

Last updated 2026-05-11 17:38:25; freshness should be verified.