The issue with direct LLM integration with web search is: They serve two different purposes. I dont search for things and want a GPT response. Likewise, I dont go to cahtgpt and want search results.
It might seem like a weird distinction but I use them differently and when you mush them together they become less useful overall.
Posting an error message into search may or may not get me a root cause or fix, but pasting it into chatgpt will very likely get me on the right track very quickly. Searching for a product I know exists is a pita on chat GPT, but a web search will pull it up pretty quickly.
If I search for a product, I absolutely DO NOT WANT A GIANT WALL OF GPT BULLSHIT before meaningful search results.
They are different products and have different use-cases. Stop trying to blend them! /rant
It's just a different way of searching. You can just say "give me links only"
It'll take time for mass adaptation
Yet I feel like there are uses for a blend:
BinGPT took a list of restaurants gave me all their hours and then formatted it into a nice markdown table for me.
Only issue is that at least one of the restaurants had the wrong hours (though I believe this was because I included notes with each restaurant and they confused it)
Still, it was nice not having to do 20 inividual searches and do the formatting manually
The issue with direct LLM integration with web search is: They serve two different purposes. I dont search for things and want a GPT response. Likewise, I dont go to cahtgpt and want search results.
It might seem like a weird distinction but I use them differently and when you mush them together they become less useful overall.
Posting an error message into search may or may not get me a root cause or fix, but pasting it into chatgpt will very likely get me on the right track very quickly. Searching for a product I know exists is a pita on chat GPT, but a web search will pull it up pretty quickly.
If I search for a product, I absolutely DO NOT WANT A GIANT WALL OF GPT BULLSHIT before meaningful search results.
They are different products and have different use-cases. Stop trying to blend them! /rant
It's just a different way of searching. You can just say "give me links only"
It'll take time for mass adaptation
Yet I feel like there are uses for a blend:
BinGPT took a list of restaurants gave me all their hours and then formatted it into a nice markdown table for me.
Only issue is that at least one of the restaurants had the wrong hours (though I believe this was because I included notes with each restaurant and they confused it)
Still, it was nice not having to do 20 inividual searches and do the formatting manually