No Stupid Questions

@rabiezaater@piefed.social @nostupidquestions@lemmy.world

generating ideas

LLMs don't generate ideas, stricto sensu. They do, and I find it useful for esoteric (gnosis through chaos magick) purposes, output names and words unbeknownst to the user (this is how I, as an ESL person, learned some words I didn't know before).

But if we consider hard determinism, do we as biological automatons, though?

learn to code

As someone who codes since my childhood, I wouldn't suggest relying on LLMs for that. They could be used to output a descriptive text about some function or library, but you must know LLMs are statistical machines, the output text is a chain of "which token is the most probable next?", an auto-completing only slightly "better" than, say, Gboard's auto-complete. They "hallucinate" precisely because they rely on statistics and randomness.

Again: extremely useful as an "Ouija board", not very useful for blindly relying for learning something, definitely not reliable for "vibe coding".

Wanna learn how to code? Do the Elliot Alderson (Mr. Robot TV series) approach: find an existing "Hello world" project/source-code, tinker with it, change things here and there, try to compile/run, Google the exception that the compiler/interpreter thrown at you, change more things, break things, then fix the things you broke... This is exactly how I did. Let go of any hurry and you'll likely going to master it eventually.

d&d [...] I need a character [...] it makes it up quick

Yes, this is one of the use cases where LLMs can thrive, as a dice with hundreds of billions of sides.

You may want to roll real dices, convert the number into the respective letter (A=1,B=2,...) then append it as a source of real entropy, because the randomness you get from LLMs is likely to be pseudorandom.

Ideally, you'd tune (using a RTL-SDR) to a blank radio frequency and digitize the (true noise) spectrum into ASCII, and voila: free randomness, straight from the Cosmic Womb to your computer!

get upset about AI “stealing” work with regard to code or other stuff that people willingly put out there for free for others to consume

Totally agree with you in this regard. Throughout the history, humans relied on other humans' "ideas". Most of the novelty stemmed from "what if I were to take this flamey thing that consumed the tree I used to sit on, and put it under this food?", mashing up existing things. If we really were to appeal, evolution is that, merging two genetic sequences in an approximate manner while trying to replicate, still I don't see humans accusing newborn of "stealing genetic work from their ancestors".

definitely useful in a lot of ways, [..] if [...] developed on a more localized and decentralized scale

I totally agree in this regard, too.

To answer the main question: IMHO, people hate AI because it has been pushed and used by corps to further enshittify this world. I'm not Anti-AI, but I'm not pro-AI either. There can be nuance from both.

AI is great, LLMs are a waste. This has been the case for years before LLMs.

LLMs which the current hype calls AI are the equivalent of a scammy car salesman. To your example of have AI teach you to code - AI is awful at coding. It produces code that is the average of a junior developer's output. It will look awesome from the outside because it will often mostly work at first, but in reality it's going to be an unmaintainable mess. An experienced engineer could use one and produce a good outcome, in some cases may be faster than without and in others slower - but the experienced engineer requirement is a must. What this means is your AI teacher by itself is a junior engineer, whose output wouldn't be trusted by themselves. That's the level you'll reach and may even learn and pick up terrible habits that'll set you back.

It will do all that and consume a ridiculous amount of resources for it compared to following a YouTube course.

I imagine a similar case is true for most industries, people who work in the industry see the absolute garbage coming out of it in large quantities and have to listen to people from the outside who don't know what good looks like in that context keep saying "oh you are now redundant cuz look how good ai is".

Meanwhile, it is trained on data stolen from the people who are now losing their jobs because the idiotic decision makers are on the side of believing how good the output looks like. AND there is more, it's doing it wasting a massive amount of resources, which drives up the prices for everyone (think all electrical devices needing computers, electricity prices). But what what money are they using for it? Oh yes! The money generated out of thin air by the corporations generating this massive AI bubble, which is most likely going to end with a crash that will decimate the market (and therefore the investments and pensions of people). And if the past is any indication, the government will prop the companies up with tax money - so people will pay for it twice.

Judging by the comments, I would say that most Lemmy users are aware of the downsides of LLMs. The average GPT user probably hasn't heard of half the points mentioned in these comments.
Judging by the downvotes, I would say that many Lemmy users are also very passionate about it. The average GPT user might think of LLMs like any other tool.

Unfortunately, I get the feeling that Lemmy isn't a suitable place for having a serious conversation about AI in general (not just LLMs). I would love to have that conversation, but this just isn't the place for it, as you can see. The people here seem to be too focused on LLMs, how they're developed and how they're forcibly implemented in places where they provide zero value etc. AI in general is such a broad category, and this kind of biased conversation misses 90% of it.

When you say AI, people hear LLM, and that's a genuine problem. When people say they hate AI, they probably aren't thinking of things like image search, optical character recognition, automatic categorization of the events of your bank account, signal processing in audio and video, image upscaling, frame generation, design of 3D structures, route planning etc. There's so much you can do with AI, but Lemmy users rarely mention those.

I don't, not in general.

There are good and bad uses of AI. For example I used AI to generate my profile picture here on Lemmy (would you have noticed?). In general the creation of art is one of the best uses of AI I can think of; it doesn't have serious consequences if it goes wrong, and it can easily be reviewed by a human whether it looks as it should.

But using AI to make actually meaningful business decisions without any human review at all? Using AI for customer service? Any company that does that deserves VERY negative consequences.

I don't agree with talking points like "AI companies should be required to pay copyright holders of their training data" or "AI is bad because of the environmental impact" or "AI is bad because of RAM prices" or "AI companies should be legally responsible for any mistakes the AI makes (such as libel or encouraging users' suicide)" or such things; I think all of these are nonsense.

I believe in general that AI gets too much attention in the media. It's really not that impactful.