Large language models converge toward human-like concept organizationLarge language models show human-like performance in knowledge extraction,reasoning and dialogue, but it remains controversial whether this performanceis best explained by memorization and pattern matching, or whether it reflectshuman-like inferential semantics and world knowledge. Knowledge bases such asWikiData provide large-scale, high-quality representations of inferentialsemantics and world knowledge. We show that large language models learn toorganize concepts in ways that are strikingly similar to how concepts areorganized in such knowledge bases. Knowledge bases model collective,institutional knowledge, and large language models seem to induce suchknowledge from raw text. We show that bigger and better models exhibit morehuman-like concept organization, across four families of language models andthree knowledge graph embeddings.arxiv.org