Expert Ratings of LLM-Generated Basketball Layup Lesson Plans

A 2026 expert survey comparing lesson plans generated by three large language models for teaching a basketball layup. Benedikt Meixner submitted a prompt to GPT-4o, Claude Sonnet, and Google Gemini, and teaching experts rated the plans on 28 predefined quality criteria using 5-point Likert scales. The dataset likely contains the expert ratings and statistical analysis results.

Use Cases

• Compare the quality of lesson plans generated by GPT-4o, Claude Sonnet, and Google Gemini based on expert ratings.
• Analyze the distribution of Likert scale ratings across 28 predefined quality criteria.
• Identify criteria where LLM-generated lesson plans show significant differences based on Friedman test results.
• Benchmark AI-generated lesson content for physical education against expert-defined standards.

Strengths

• Expert evaluation using 28 predefined quality criteria provides a structured assessment framework.
• Comparison includes three prominent LLMs: GPT-4o, Claude Sonnet, and Google Gemini.
• Results include statistical analysis (Friedman test) to identify differences among models.

Limitations

• Dataset scope is limited to a single physical education topic (basketball layup) and is very small (22.6 KB).
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

figshare

Collection Method

Expert survey rating LLM-generated lesson plans.

Freshness

Last updated 2026-03-19 06:48:14