If LLMs used very little RAM and processing power, was light weight even for Mythos or Sol, would people be as opposed? Like no data centers, no memory shortages, like it was able to run on a calculator with the same performance.
If LLMs used very little RAM and processing power, was light weight even for Mythos or Sol, would people be as opposed? Like no data centers, no memory shortages, like it was able to run on a calculator with the same performance.
This is pretty much my thoughts as well. My unsolicited rant:
The technology is good and anyone who says it’s “just autocorrect” are wrong. Having said that, I agree that LLMs are useful really for nothing more than low-risk, personal research and/or signposting to other sources. It’s in no way a replacement for a human (unless your definition of what a human does is that it’s just a thing that produces words at you - in which case, you’re probably a middle manager - very myopic, regardless). It’s also not appropriate for anything financial, professional, or confidential. Here, the first L in LLM is a hindrance. Local, small language models are more appropriate and cost-effective here.
Anyway. I don’t “use AI” like some people but over the last year I’ve used my access to Proton’s Lumo (already included in my subscription) a handful of times and been pleased with the results. The way that I see it is that all the LLM libraries are trained almost entirely on Reddit or websites that stole their content from Reddit anyway, so use it like you would Reddit: “LGBT horror movies before 2000,” “Summarize and cite criticisms of Camille Paglia,” “Was Bioshock 2 well-received at launch?” All very low-stakes. Doesn’t really matter. Could’ve been a Reddit thread, but it’s likely that any response there would’ve been a middleman for an AI answer anyway.
Anyway. More: Call it cope but I’ve been developing a theory/outlook that the internet as a thing is basically dead and that the Pandora’s Box of LLM tech, which can produce a basically infinite amount of what was once the internet’s currency (read: information, but really: text), has killed it. The Fediverse is utopian but I don’t think anyone here really believes it will ever be anything more than a cluster of fringe communities. Having said that, I justify good LLM use (like mine, lucky that) as being the “New Game Plus” mode of the internet. Everything is already there, everything’s unlocked. There is no challenge. It’s kind of cheating and you’re not really supposed to play the game like that. But at the same time, there is no going back to the real beginning, I just have the option of nostalgically re-creating it by going to old.reddit or postfixing before:2022 to all my
GoogleDDG searches. Fun, and game-ish in it’s own right, but also artificially limiting for no reasons beyond that./rant
A lot of people like it this way but I really don’t. What I wished happened during the last few reddit exodus was what happened back in around 2010 when digg users ran away to back-then-new reddit. The good thing about the fediverse is that people can actually choose instances they like that defedrates from content they don’t like. I’m a hypocrite if I say I don’t agree with that.
But somehow something must have changed with internet users between the digg-to-reddit exodus to newreddit-to-fediverse exoduses (weird that my auto-complete says this is the correct plural form) and not a lot of people want to contribute for making content. I’m pointing at myself as well so sometimes I do ponder why I don’t be the change I want.
Gist of it is I want reddit circa 2010-2015 on the threadiverse specifically (as I don’t use the micro-blogging side) but it’s having trouble attaining good people that are willing to contribute time and energy to make worthwhile content.
Except a large part of the reason that people used reddit is that you tended to get decent information. The big/good answers would often be some dude who is waay to into football having a 6 post page about who the GOAT player is.
Now, AI is basically so prolific, that a lot of what AI is being trained on…is other AI. Its slop badly reproducing slop, and the answers that AI gives you are either wrong, or are driven largely by somebody making 50 AI chatbots to swarm reddit with their opinion (or more often, to sell their product).
I was already sold on it, you don’t have to keep buttering me up on the idea.
When everything is so searchable and indexable, obscurity is almost a feature