LLMs reportedly often use text matching when search is needed

An author of “Querying Databases with Function Calling” reports that LLMs often use LIKE or exact-match operators when natural-language commands require SEARCH in Weaviate. The author also says their experiments found a strong bias toward exact text matching, which they think helps explain the popularity of agents paired with grep or BM25 rather than semantic search—search based on meaning. In their view, combining an LLM’s reasoning about a query with a semantic search engine is more effective than leaving the semantic work to the LLM alone.