I semi-regularly remember reading something without knowing the exact wording used for the title, particularly for terrible titles like "A new approach" (becaus

I semi-regularly remember reading something without knowing the exact wording used for the title, particularly for terrible titles like "A new approach" (because that's the original title even though it helps nobody), so the Algolia HN search is hard to use because I need to try a lot of variants ("was that article with that shell script trick about Bash? Linux? Shell scripting in general?") and (near-)synonyms ("emission", "greenhouse", "climate", or "warming" could all be used in an article about environmental policies). Being able to ask "give me the story from last week about global warming", and it being able to recognize "Climate change goals adjusted at government meeting" as being a match, without it matching any of the actual words, can be very helpful. I believe that is one of the things Google silently innovated/implemented a ~decade ago, or at least I never heard anyone talk about it the way that people talk about PageRank as magic sauce, but finding such topical synonyms is key to finding more relevant results. (Or super detrimental in other cases, but the quotes operator can help turn that down.) I would be curious how they did it, given that LLM wasn't around then