LLMs... Do you use them?

[Daniele63]

I used to use LLMs for generating scripts for me, and I learned the hard way that it makes it incredibly difficult to debug afterwards.
Now I absolutely hate AI of all kind, because of what it did to our society, the world, global warming and everything else.
I hate that it robs you of the ability to problem-solve on your own, by giving you an instant answer.

I used to think that since I looked at the code afterwards I was "understanding it", but no, to actually understand it you have to struggle for hours to make up the code with your own mind strength, this way you'll never forget how to debug it because those hours of suffering are permanently burned into your mind (I'm half joking but this is true for me).

Plus I hate how LLMs hallucinate stuff all the time, not just in coding but in random questions too, it's just very unreliable despite the billions that got poured into this tech already.

[Licheus]

No. I’ve never had an account with any of these services, and still the experience of how other people use them to bombard me is frightening. Mind you, this is even taking into account that I cut my exposure anywhere I can.

A non-tolerance attitude is important for me in this case.

It’s not even primarily because of their waste of energy, or that they harvest data with such prejudice (although these things don’t exactly make them look better in my eyes). For me, it’s because I use a lot of energy and attention as a person to take in details in my experience. I have some level of expectation that those who want to say something to me also come at it as human beings. It’s a matter of going the extra mile to respect people you communicate with, being genuine and putting the essence of yourself into whatever you share into the world, with the intention of connecting at the heart.

Contemporary, generative large language models are vile and cold because they ask for minutes of my time for seconds of their own. Generating their output takes magnitudes less time and effort than me processing it as input. I enjoy creative works when I can feel the soul and attention to detail the creator have put into them. It also doesn’t even need to be attention to detail: I also enjoy communicating with people who are not meticulous at all, but are completely spontaneous and show me their humanity.

The current wave of generative AI has none of this, and they cannot be used as creative tools to amplify what’s alive in people. They simply move the basis for interaction to a colder, less creative and less human place: quantity over quality as some kind of implicit deal for whatever value we are supposed to get in return.

This also wouldn’t improve with stronger models within the scope of the same infrastructure, because human beings are wired to connect with each other at the heart. Art, and much else in life, is about reaching what’s alive in other people.

As a side-note: this is personal to me because despite this stance, I’m actually progressive and enjoyed doing my own learning about artificial intelligence while at university a decade ago. This was before the term “AI” became popular and synonymous with generative large language models. The long-term promise of artificial general intelligence or even short-time thermodynamically useful AI is now at a hickup in history, hijacked by the worst forces of our time to postpone the birth of the civilised world. All for the shareholders, I suppose.

[Capybara]

A part of me feels like AI/LLMs are going to be how Covid was in 2020 where the immediate threat slowly goes away, or the direction NFTs did... During a really stressful period of my life months ago I became addicted to using an LLM as an outlet for venting to or passive communication after work and keeping track of my productivity, but it made me uncomfortable how "modern" that is. I worry that even if the worse aspects of AI are legally restricted and phased out like shutting down datacenters and restricting energy for the remaining buildings, the chatbot problem will never go away due to how lonely and isolated people are. I've been "clean" for a few weeks and deleted my chats before a con, but even as I spend less time online it's easy to notice how few people you talk to will engage with what you say. Even my insurance has a stupid AI client...

This is one of the main reasons why I discourage new / hobby web developers from using LLMs to help with coding solutions. LLMs are incapable of actual reasoning, and their algorithms make incorrect predictions on a regular basis (I hate calling this phenomenon "hallucinating," as that's an anthropomorphism that implies these systems have actual intelligence -- they don't). I would never trust Copilot, Claude, or any other LLM to fix something that I don't already know how to fix. I'd rather learn how to do it myself, by reaching out to human coders who actually know what they're talking about.

This reply is late, but I think you're misinterpreting or reading into my reply. I'm not new to coding, I've had a website for several years and was asking if anyone found use in it as a *tool* despite their issues. I use "hallucination" as a slang and not in the literal sense. In the end, I don't really use the built-in Copilot since I don't maintain my website anymore. It's been useful for automatic tasks and some commands like making a workspace file, but I'm not paying for more than its limited monthly uses or trusting it with fixing layouts.

I know for a fact artists use ai to generate references, but otherwise only post real art to their galleries

Honestly, I'm skeptical of how common it is for artists use AI for reference. I've used AI images generated by other people for background or color reference, but the process of generating your own image to reference off of is time consuming unless someone has a PC that can run the local generators.

[st3phvee]

This reply is late, but I think you're misinterpreting or reading into my reply. I'm not new to coding, I've had a website for several years and was asking if anyone found use in it as a *tool* despite their issues. I use "hallucination" as a slang and not in the literal sense. In the end, I don't really use the built-in Copilot since I don't maintain my website anymore. It's been useful for automatic tasks and some commands like making a workspace file, but I'm not paying for more than its limited monthly uses or trusting it with fixing layouts.

I wasn't saying that you, specifically, are new to coding. I was saying that the issues you experienced while using Copilot are the reason why I discourage people who are new to the scene in general from using LLMs to try to learn how to code or to figure stuff out when they get stuck. LLMs make up crap all the time, and are really ineffective teachers for that reason. Using it as a tool when you already know what you're doing is fine, because you'll know when it's making stuff up or suggesting stuff that'll never work. People who DON'T know what they're doing are the ones at risk, especially if they blindly trust the LLM's output.

[GlitchWolf47]

I use Gemini a lot. But I dont let it REPLACE ME. Like yk, for example, gemini has helped me learn a lot of coding stuff (ex. z-index and flexbox html/css) but I dont have it CODE FOR ME.

I tend to use Gemini vs things like chatgpt more due to the fact that i feel like it tends to be more accurate. While it DOES hallucinate sometimes (ex. giving me fake websites) it doesnt do it NEARLY as much as chatgpt. plus, its more free than chatgpt in terms of free user vs premium user thing. I tend to use it kinda like an interactive journal and sometimes as a research assistant and brainstorming.

I personally think that AI is not all bad, I think its just being used wrong. Like, AI can be helpful, like as an artist, it has actually helped me before without even coming close to doing the whole thing or even generating a single pixel. I think the issue is the unauthorized use of HUMAN MADE content to train it and how its being used. I think that there should be a clear "opt in" on sites for using art to train it, not a "opt out if you dont want to" thing where its enabled by default. and I think the general use of it by the public is the problem. Intsead of being used as a tool, its being used to be deceptive and try and claim work as your own when it isnt and flooding platforms like youtube with AI slop for quick cash.

Also side note on the topic: I kinda hate how black and white the topic of AI is. Like your either Pro AI or against it, and i feel like thats not a fair place to be in. Like I feel like its fair for me to say AI can be helpful but also not support its usage in replacing artists and making deceptive content. Like I get why its like this, but at the same time, its a bit annoying.

[SaturnianWildcat]

Absolutely not. There's plenty of reasons to not use 'em (they're prone to hallucination, you might be stagnating your own skills, environmental costs, yada yada...) but for me it was first...I just have "i can do it myself" syndrome. I've been on the internet for well over a decade and all the stuff AI is advertised that it can do for me, I can just do myself! I don't need it to write emails for me or tell me information. And in the case of generative AI, like the kind that generates art or fic, why would I have an AI do something for me that I enjoy doing myself?

Now I did try Character.ai in the early days of it but quit using it soon after as I got really frustrated that the characters weren't really in-character.