Local LLMs, or SLMs, are surprisingly efficient at document handling with vague inputs, which was a weakness of traditional search algorithms. You can use a mid tier GPU for that.

When the bubble pops, that’s where the research community will go.