This is why I use SearXNG; locally hosted on a container. It collates and sorts out the crap from all the engines at once. I get a useful list of results normally. My personal instance is configured to shotgun several engines at once and uses Wolfram Alpha or Wikipedia for informational boxes over other engines; if those services present a result.
My experience:
- Infobox (if applicable), Source is reliable; usually Wolfram Alpha or Wikipedia. May not always be immediately relevant but it's usually a close enough match.
- Most relevant results (3 to 7 of them)
- Relevant results containing any terms (Many)
- Less relevant results (Usually on page 2 or 3 by this time)
- Nonsensical results; may be slightly relevant (Usually you're 7 to 10 pages deep by then)
Yeah this seems like a non-issue to me as well; the source material for the models is probably the cause of this bias.
I also don't think there's a lot of sources for this manner of speaking. Let's also not forget that there's oftentimes instructions given to the LLM that ask it to avoid certain topics which it will in fact do.