You seem to be describing an answer engine, which isn't what I said. It's really no worse than the last five or so years of Google if you just want links to a topic that you can go and read yourself. It's fine at that, it's not good.

It takes work to turn them into useful search engines. Set up librewolf. Drop in at least a search and a fetch agent. Wrap it in a some prompts to make sure you get actual links.

On my extremely non-scientific scale I'd rate that setup somewhere between barebones Kagi and pre-enshitified Google in terms of results.

I'm hopeful that an actual agent framework will do a better job that's significantly more work. If you just log onto a SaaS LLM you'll get very unpredictable search results.

To be fair, ai summarys and using it as a search engine are two VERY fucking different tasks. Like night and fucking day.

It also doesn't help the ai models in charge of the summaries on most search engines are some of the stupidest lowest quality pieces of shit we have.

A model told to actually do a search properly is functionally no different than if you or I did it. And will actually be able to do the job properly. They are very good at this sorta work load.

You ask it to go though a bunch of summaries then summarize those summaries and do use in like 1 seconds. You of course are getting fucking slop.

You aren't giving it the time or information needed to do the fucking job right. They arnt magic machines they can't do shit instantly.

I'm sorry, but I'm certain that the average AI summary beats/matches what the average person gets from a normal search. You're acting like people before AI didn't just believe whatever the first or second link told them, for most cases.

If you've ever even read the summary of a study, you're in the top 10% of "people that research online".