GPT and Search

There’s a lot of talk about how GPT is going to take over search. Meaning, compete with or take down Google.

I get the excitement there, but there are some pretty serious barriers to having this happen immediately. First, GPT is non-deterministic, meaning you can ask it the same thing twice and get completely different results. In fact you’re likely to.

Next, I think the whole idea confuses what GPT and search are good at. Search looks up facts. It’s pulling from a database. While the “G” in GPT stands for generative. Meaning, it’s making things up. It’s often very correct when it does so, which is why we’re so impressed. But it’s also often wrong in truly comedic ways.

Imagine a young Jedi asking Yoda how many moons exist in the Dagobah system. He’d like answer something like:

Feeble my mind is for such things. Wisdom is that for which I focus. If need this data you do, go to a goddamn archive and look it up yourself.

So I think GPT could be amazing, with some likely soon-to-arrive tweaks, for answering basic and mostly static facts. Like, what’s the best advice for someone in an abusive relationship? Or what are some great ideas for traveling in the US? Those change, but not very often, and the insight you get from GPT is Yoda level rather than just a database lookup.

But if you’re asking what the best article is on learning Vim, that requires that you’ve looked at all of them and done some sort of ranking. That’s not GPT’s thing.

So I’d be a bit skeptical of the claims around GPT killing Google any time soon. Maybe at some point in the near future, if they incorporate a lot more database lookup functionality and combine that with results stability. But until then it’s going to be a lot more Yoda and and a lot less Einstien.