A study compared semantic search dynamics between 82 human participants and three large language models (GPT-4o, Gemini-2.5-Pro, Claude-Sonnet-4.5) using verbal fluency data. The analysis quantified entropy, distance to next, and distance to centroid across eight temperature settings for each model.
- Humans exhibited higher entropy, larger semantic steps, and broader dispersion than all tested LLMs.
- Temperature tuning produced only partial alignments between human and model metrics.
- No configuration reproduced the complete human profile across all dimensions.
The findings suggest that human semantic search implements a distinctive balance between local exploitation and global exploration that current model architectures fail to reproduce.