Complex search: the hard part is asking the right question first

Let me say something. I've been using Pro Search pretty heavily lately, and my takeaway is that its value isn't in the answer itself, it's in the step where it breaks down the question.

Last week I threw "how to prepare for a machine learning interview" at it. Regular search dumped a pile of interview-experience links on me, each title more identical than the last. Pro Search was different—it first broke that sentence apart, split it into several directions, then went looking separately. Some of the material I found also described it this way, saying it thinks like a human about the real need behind the question. My own testing basically lines up with that.

The difference is exactly at that layer. Search has been around for twenty years, and it's all been about indexing and ranking—the logic being, you ask precisely, I find fast.

Pro Search flips the approach: it assumes you asked imprecisely, and fills in that middle layer for you. So it's better suited to questions you yourself haven't figured out.

But breaking down the question has a cost. When it splits in the wrong direction, what you get is structurally complete, clearly organized, and reads perfectly smoothly—it's just not answering what you wanted. That's harder to notice than traditional search dumping ten irrelevant links on you, because with the latter you know at a glance you need to search again.

My judgment is that the bottleneck for complex queries has always been problem definition; the retrieval end has never lacked tools. Pro Search is grabbing that segment. Grab it right, and the experience is smooth; grab it wrong, and you have to come back and retype the question yourself. And the more decently it breaks things down, the more you can't be bothered to go back and fix it.