Table 1 Dataset statistics.

From: The “LLM World of Words” English free association norms generated by large language models

Network

Unique cues

Total responses

Unique responses

Missing responses

Humans

11,545

3,148,578

116,640

9.1%

Mistral

11,545

3,268,206

41,369

5.6%

Llama3

11,545

3,348,049

105,367

3.3%

Haiku

11,545

3,403,644

15,275

1.7%

  1. Cue and response statistics for all datasets after preprocessing. All networks have the same unique cues, but different numbers of total responses and unique responses. The Human network has the largest percentage of missing responses, but also the largest number of unique responses compared to all LLMs.
Back to article page