The post discusses how researchers are examining the relationship between token embeddings in large language models (LLMs) and human cognitive processes of categorization. The discussion emphasizes the methods used, including static token-level embeddings, and argues about the relevance of these methods concerning how LLMs function. Critics point out that the reliance on static data without performing full model computations undermines the validity of the conclusions drawn. Some commenters express confusion about the methodology, especially in the scope of the analysis focusing on nouns. The discourse highlights the complexities of language and embedding in machine learning models. Overall, while exploring the intersection of LLMs and human thought processes is intriguing, skepticism arises regarding the analysis techniques employed.