ChatGPT already far outperforms google for heavily-SEO’d topics like recipe searches.

Try asking “what herbs and spices go well in a chilli?” in both. I get a sensible, rough answer from ChatGPT within seconds. From Google I get page after page of content farms hiding information in amongst ads and life-story filler.

I wanted to demo it for my wife who was writing a paper on digital media consumption during the pandemic.

She was trying to find poems that went viral during the pandemic. Googling it showed a bunch of mediocre articles or irrelevant news stories

I asked chatGPT and it gave me a list of 10 poems by relatively well known poets. A brief review showed that these poems were, indeed, viral during the pandemic. Saved at least 15 minutes of Googling.

This is going to take a long time. Longer than one would think if one is using the progression of the language model as a basis for this prediction.

There are a lot of pitfalls and erroneous assumptions built into both Google's current search and the information used to train AI/ML models. Two big ones are "assuming that a person searching for something and accepting the answer means it's a successful search" and "assuming a person searching/asking for something actually wants what they're asking for".

I'm a librarian with several years of reference experience under my belt and neither of those things are true. They're both good tools for a well considered and well informed information search, but that 'well considered and well informed' is doing a lot of heavy lifting.

Since you have domain expertise, I'd love to know your opinion on searching through "personal libraries" like Zettelkasten (or similar repositories), and perhaps linking that with Internet-scale indexes.

Are there tools that do that well right now? Do you know of (maybe niche) projects exploring such ideas?