ChatGPT won’t replace search engines any time soon

I know this thread is about ChatGPT, but if you're still using Google then I suggest giving Kagi a try. It's a paid search engine, but I've been using it for quite a while now and really enjoy it. With Kagi you can rank and downrank domains as well as block them. There's also this think called "lenses" which are basically filters for specific kinds of searches. For instance, I have a "Programming" lens and a "Academic Research" lens.

> Personally, I wouldn't go long on search engines that don't have a strong ML component to them in the future.

What's kind of ironic about this is I think search engines may have mistakenly moved away from strong ML in the sense that you're thinking of.

Yes, ML is being used for recommendations much more than ever, but in terms of heuristically finding pages with the keywords you entered, mainstream search engines have become significantly worse at it. I remember a time when The Google would find any pages with the keywords you entered. In recent years (before I stopped using it), I noticed an increasing number of times where I knew it had a page indexed but it would refuse to include it in the results for whatever reason. Either its ability to fuzzy search pages seemed diminished or it would just not match something word-for-word. I could sometimes figure this out when the page I was looking for previously accidentally came up in the results for another barely-related search, so I knew it wasn't that the search engine was culling old pages. Though I'm sure they're doing that as well where they think they can get away with it.

Recommendations and curation are largely overrated, and that's where a lot of ML has been mistakenly applied. Well, I say mistakenly in the sense that it benefits the individual and society. Recommendation engines do serve the purpose of the company selling those recommendations.

A true application of machine learning to answer engines is the future and will be a big problem for companies that fought the advertising wars by banking on recommendation engines. That is unless they turn their ship soon enough.

And comparing these two results for the same question serves to illustrate how ChatGPT does not in any meaningful sense _understand_ what it is "talking" about. It basically just strings together relevant-seeming words in a plausible way. The result may be useful, or it may be nonsense.

You'd better know enough (and be alert enough) to tell the difference...

It will for content that doesn't exist yet.

One thing the article misses is that users don't just use Search engines/Google/etc to just search. We use it to solve a problem we have. Find the nearest X, Cheapest Y, Answer to Z. A few of these can be done with the localized data that ChatGPT has. Ofcourse, scaling it to search engine levels is an insanely hard challenge. But it doesn't need to go all the way in to do that. Selectively targeting certain portions of what we use Search for can easily get you to a lot of daily queries that you google. Imagine if ChatGPT didn't need to crawl but you as a Service/Business had to integrate with it, that would could evolve into something different from Search engines and yet eat Search engines.

I give ChatGPT a question on how to do X in some programming environment. So far same as googling. But then ok, for me that solution gives errors, or function doesn't exist, 2nd question to ChatGPT, bam half the time it gives something that puts me on the right track. I can't tell google that my python/postgres problem is special, so please do That example, but with such and such restrictions.