Search algorithms are, depending on the specifics, potentially “ai” now. If we’re tokenizing out vectors and running a straight match query (i.e. postgres full text search) that’s not AI - that’s just string matching. Some of the offerings get into NN guided or LLM powered… these tend to suck though because they’re unpredictable and inconsistent. That may just be the novelty of the technology though, we’ve had decades to work on small word exclusion and language specific dictionary mapping - it’s possible the consistency will get up and, at least when it comes to searching, everything really good already uses weird heuristics so it’s not like we can reason on why specific results are preferred, we just know they’re consistent.
Search algorithms are, depending on the specifics, potentially “ai” now. If we’re tokenizing out vectors and running a straight match query (i.e. postgres full text search) that’s not AI - that’s just string matching. Some of the offerings get into NN guided or LLM powered… these tend to suck though because they’re unpredictable and inconsistent. That may just be the novelty of the technology though, we’ve had decades to work on small word exclusion and language specific dictionary mapping - it’s possible the consistency will get up and, at least when it comes to searching, everything really good already uses weird heuristics so it’s not like we can reason on why specific results are preferred, we just know they’re consistent.