AI is trained on the way the words are used, context, when they appear in which order, appropriate priority, among other stats of that nature, in order for its LLM to output those desired patterns. Dictionaries do the same thing, in order to craft dictionary entries, just with fewer or different stats. Lexicographers have been studying the way people use words, in order to discover the most common ways they are used, by observation, in order to fill their dbase/corpus with citations of said methods, and craft an entry for each word based on stats. The reason a single entry for one word has multiple uses, is that the 1-2-3 order of uses is a statistical ranking of how frequently those uses have been observed.