A loosely moderated place to ask open-ended questions
Search asklemmy π
If your post meets the following criteria, it's welcome here!
Looking for support?
Looking for a community?
~Icon~ ~by~ ~@Double_A@discuss.tchncs.de~
What are you talking about? I just tried two test queries on DDG, and neither one had LLM-generated nonsense, and the one that was in double-quotes returned only five results, all of which had the double-quoted phrase and one of which was the thing I was challenging it to find.
Can you give an example of a query where DDG returns LLM results or doesn't respect your double-quotes?
I think they are referring to the search engines returning LLM content farm websites.
Maybe Iβm a little out of the loop, what are llm content farm websites?
Low effort websites made easier by LLM generated text. Itβs not new, just made easier with the ubiquity of LLM tools. Think of it as the latest generation of spam websites π
Ah I see. Junk βnewsβ and other regurgitated blah. Yeah, Iβd guess any free search engine will probably be bloated with that. Not to mention that itβs google, bing, and orange bing. Not a ton of crawlers out there indexing everything is there?
The only other (not absolutely tiny) one I'm aware of is brave, but it has its own issues
Ironically one of DDG's early selling points, before they fully jumped on the privacy bandwagon, was that they would filter out results for low-effort content farms (this was pre-LLM stuff).
I had used DDG since almost the beginning and it was one of the things I was originally sold on. It's difficult to find a source for it now but I did find this: https://web.archive.org/web/20110608072253/https://www.technologyreview.com/blog/post.aspx?bid=377&bpid=25532
Forbes for example
Think recipe websites that take forever to get to the recipe but it's for other topics. Like a simple question, "what is the release date for X new game?" And then there will be like 5+ paragraphs of jibber jabber about the game and then finally the last article will say when it releases.
This sort of site has been around for a while but supposedly they're more common nowadays. Personally I think people just have a better eye for things not written entirely by humans. Either way it's annoying to deal with them.
Ugh I feel like I have been seeing more of that. Asked how many ml in a wine pour and got like 5 sites that wouldnβt just come out and say it. All kinds of gobbledegook dancing around the topic but no one would just freaking say it. 140ml in case you needed it
Sites that mass-generate garbage using llms
https://github.com/laylavish/uBlockOrigin-HUGE-AI-Blocklist is supposed to remove AI slop from the results of various search engines. Wants the uBlacklist browser addon though.
uBlacklist is an excellent add on anyway
Posted this previously:
yes. use any of the following, in no particular order:
Also, Stract
Are you (or is anyone here) daily-driving Stract yet? I discovered it a few months ago and thought it was everything I was looking for in a search engine, but also concluded that its search results aren't up to the standard I can use for now, so I filed it as one to look out for. Would be interested in hearing others' experiences.
Unfortunately not. I'd like to, but as you say it's not quite there yet. I probably should try it more frequently.
I think Searx is a good suggestion. Can be a bit slow to return results because it runs the search on a bunch of search engines and compiles the results, but that helps to make sure better stuff rises to the top.
The other suggestions aren't suggestions at all. They are obsoleted by searx.space
DDG ... obsolete
startpage.com ... obsolete
Browsers have default search engines. Curse everytime, DDG is accidentally queried.
DDG is a curse word!
Any centralized site, with privacy claims, is treated as lying thru their teeth. Front run future news.
Just tried it and it actually worked surprisingly well
Microsoft invests in Ecosia
I use Qwant sometimes but it's sourced from Bing. Searx is better if you can self-host. Kagi is better if you can afford to pay (but you asked for free).
There's also Startpage, also pretty good. That has search results from Google and a bit of Bing though, so if that's a dealbreaker, then yeah.
Doesn't Kagi offer X amount of free searches per month before you pay?
No. They have a trial of 100 one-time searches, but that's it.
Brave Search is pretty nice
I enjoy using Brave Search. But it sucks for images for some reason. I use DDG for images.
Yeah, agree. I donβt see many LLM results but I canβt say I never see any.
Kagi if you are willing to pay for the service. I think that's reasonable but your needs may vary.
I might pay for quality search if my searches werenβt linked back to my credit card and therefore identity.
SHUM
If it wasn't something I use every day and even for work, I wouldn't bother. But yeah, Kagi result are definitely better. Being able to rate sites high/lower in results and blocking some all together really has helped filter out the nonsense that a lot of search engines give
Not sure if its untrustworthly or not, but switched over to Startpage and been liking the results. Just wished it would implement the !bang system from ddg
They use Googles results with a bit of Bing mixed in. Bangs should work too, like !wiki for Wikipedia or !d for DeepL They're partly owned by an adtech company though (and say they dont share anything).
Startpage.com works well enough
I've been using brave search on my pc and phone for maybe 6 months now. i still use google like 10% of the time if i'm searching for something that isn't in english, but otherwise, id even say for many things brave returns better results than google
udm14.org is about as close as you can get
startpage
Startpage uses Google.
You can go old school like me. I just keep a list of trusted sources. Like I work a lot with SpringBoot, so I add "baeldung" to my search. If I want to look up a vacation spot, I usually add the name of national television's vacation review show, because they always give good advice and have been doing so for over 20 years.
shitty LLM-generated nonsense sites
It's kind of impossible to filter these out computationally, regardless of brand. You pretty much need to run them through a bigger LLM than generated them, and the economics of doing that for every indexed site are obviously bad. Doing it by hand may or may not be workable either, depending on how quickly you can detect bad domains versus how quickly a new domain can be put up.
It's on us to figure out who's trustworthy, and who just sounds authoritative, unfortunately.