Discussion
What bothered me about LLM use growing, and why people would ever ask it anything at all when it only gives an answer that sounds like an answer might, had me missing something glaringly obvious until this week.
I realised it while on my eighth google for the purpose of diagnostic LEDs on a Mac Pro 4,1 CPU board - and getting links to bad forum answers, and videos (some AI generated themselves) I realised…
Search engines today also only give shitty approximations of what answers are like.
@NanoRaptor Google and Bing both replaced page-rank with IA intuition. Absolutely the worst way of doing it as the inference behind the search is the same quick and dirty you get with the free versions of ChatGPT and Copilot (and others). I have found the paid versions to give exponentially better answers with much less risk for slop (although the risk is always there)
@NanoRaptor
It is all about monetisation.
For example, Google pushes Youtube videos so that it can make money by showing you advertising.
"Sponsored" links are prioritised even when they are not relevant.
With AI answers disabled, searching is faste, but the search algorithms are still tainted by AI, and the sheer accumulation of stuff in today’s web drags down accuracy, mainly because of a huge accumulation of outdated info and increasing AI slop clickbait.
@NanoRaptor @cstross I gave in and decided you get what you pay for, and search is important. So now we have a family subscription to Kagi,and I’m so happy with it I’m now using their browser as well
@NanoRaptor Freya Holmer’s video on this is problem is one of the better reasoned ones I’ve seen.
The TDLW version is AI slop is ruining search by filling all of the results with AI hallucinations. And as a result, there’s no such little no -AI content left to crawl that the AI is only left consuming itself.
@NanoRaptor You are not alone in noticing this. I had bookmarked a very educational blog post by @pluralistic with links to further sources on the topic of search engines worsening. If you feel like reading up on that: https://pluralistic.net/2024/04/24/naming-names/#prabhakar-raghavan
Fair warning though, it's all a bit depressing (´＿｀)
@sev @NanoRaptor @pluralistic I think the (deliberate) worsening of search is half the problem - the other half is that the internet itself is filling up with low-effort slop (this started pre-LLMs, but LLMs certainly make it easier to do), so even a decent search engine has a harder problem to solve sorting wheat from chaff.
@NanoRaptor
So-called "AI" is just like the internet:
Garbage In, Garbage Out.
At least the name of the so-called "AI" is usually half-right, though there have been some notable exceptions even to that.
Sturgeon was an optimist.
@NanoRaptor the really crazy-making thing is that the _reason_ search engines have gotten so much worse is because of LLMs
both because generated shit is crowding out humans sharing actual first-hand knowledge on the internet, _and_ because search engines have become much more aggressive about indexing pages based on a language model's interpretation of the text rather than the text itself, so the specific words you use in your search query becomes kind of a vague suggestion
@NanoRaptor It has been unpleasant realising just how many people consider answer-shaped objects and answers to be the same thing.
@NanoRaptor It has been unpleasant realising just how many people consider answer-shaped objects and answers to be the same thing.
@NanoRaptor The 3,1 I know! I have two of them. With different lights: so I combined all the bits that work and now have one that’s “fine”.
Good luck with all “search” these days. And/or when I want to know something I’ll ask either you or @futzle !
So I held my nose and asked chat-glorified-predictive-text and it gave me a mildly wrong but far better answer than Google, DuckDuckGo and Bing.
I posted the above comments in a few places online and so far have had three people tell me I simply need to learn how to search. They confidently told me the same wrong answers.
Thus this is now my last day online, I’m moving into the woods to make potions from archaic tomes and maybe I’ll die from mercury poisoning or laudanum. Maybe both.
@NanoRaptor This is all sad but true. But also maybe not confined to the internet. I often wonder if humans aren’t also saying things that sound like something that would be true, perhaps checking to see if it’s also something that might have been part of their training set (memory).
The more outrageous AI capabilities get based on stupid correlations, the more inclined I am to believe our capabilities are based on similar stupidity.
@NanoRaptor “need to learn how to search”
Ah the people who haven’t noticed yet…
Doing a proper search used to be a skill indeed, but it’s a losing battle against enshittification and the skill is now much more in sieving through wrong and/or unrelated bullshit to find enough little kernels of close enough to true to piece it together.
"glorified predictive text" --lol perfect!
If one day I just quit posting — no goodbye note, no account deletion, just...gone — it is because I came to the same conclusion and am now happily hanging out in a desert cave, chilling with all the tame lizards, living off of cactus fruit and rabbits and scavenging the Samaritans' water and food drops left for migrants. I think you are on to something here.
@NanoRaptor if it makes any difference, I would miss your posts…
A space for Bonfire maintainers and contributors to communicate
Install bonfire.cafe
Get the full app experience
