Yes, I use search engine(s) constantly - namely Kagi, which really does feel like Google used to.
I tried using LLMs for a recent project of mine when I was trying to figure out if something was possible, and they were actively misleading, every time. My issue for this project was that what I was asking for did end up not being currently possible, but LLMs wouldn't tell me that and would make up incorrect ways to solve my problem, since they didn't want to tell me it couldn't be done.
Really, these days, either I know some resource exists and I want to find it, in which case a search engine makes much more sense than an LLM which might hallucinate, or I want to know if something is possible / how to do it, and the LLM will again hallucinate an incorrect way to do it.
I've only found LLMs useful for translation, transcription, natural language interface, etc.
It's the same for me. I've switched to DuckDuckGo about 2 or 3 years ago and it feels like Google used to. I'm always shocked to see how bad the results are and how cluttered the top section is on Google if I happen to search there on someone else's computer.
LLMs have mostly been useful for three things: single line code completion (in GoLand), quickly translating JSON, and generating/optimizing marketing texts.
Agreed, for resource location, Kagi feels like Google did 20 years ago.
I use LLMs as a sounding board. Often if I'm trying to tease out the shape of a concept in my head, it's best to write it out. I now do this in the form of a question or request for information and dump it into the LLM.
Fully agreed. I pay for Kagi monthly now, and it's totally worth it. I really hope they grow and become more well known because they doing what one of the biggest companies in the world is doing, and doing it better, and it seems so few people even know about them.
Really, these days, either I know some resource exists and I want to find it, in which case a search engine makes much more sense than an LLM which might hallucinate, or I want to know if something is possible / how to do it, and the LLM will again hallucinate an incorrect way to do it.
I've only found LLMs useful for translation, transcription, natural language interface, etc.