I'm working on something of an article to that effect, but... I also have other things going on, and worry that by the time I've hit all the key points and polished 'em up, it will be completely outdated. Obligatory lol.

I guess a couple key points for that general LLM article would be:

• Going in, it's pretty dang important to understand exactly what a modern LLM is; what it's strengths and weaknesses are. How they can vary widely in quality and usefulness depending on their engines and platforms you access.
• Just like people who sound extremely sure of themselves, I think it's geberally helpful to think of LLM's as highly-opinionated friends who sometimes have the ability to be incredibly incisive and useful on certain matters, but can also be incredibly flawed, even upon very simple stuff.
• Also hugely important based on service involved to get a feel for where they're most reliable, and where they can be downright disastrous to rely upon.

Vibe coding silly Python projects.

When Google/DDG fails. Then inevitably it fails too, but at least it begs for forgiveness.

Nothing. ChatGPT is terrible.

I used to ask it a question here or there if I couldn't find the answer anywhere else, but I wasn't too satisfied with what I got. I tried to use it for coding but I kept getting funky answers so I stopped using it for that too.

The last time I used it, I was doing a repair job and asked it a question I essentially knew the answer to because I wanted to verify a step. It gave me an answer that was not only wrong, but legitimately straight up dangerous. I haven't used it since.

Same-ish. I'm fluent in Perl, but whenever I write a web backend I use ChatGPT to help me cook the JavaScript needed by the front end. No way in hell I'm learning Javascript without a gun to my head.

I don't use it.

I used it to help me with my hobby projects. I have an idea, and it tells me what parts to buy with what spec (like a 3000 rpm gearmotor, a 1:10 ratio gearbox, a camshaft, etc.) which I have no idea where to even start without it and without mechanical engineering skills.

nothing.

To drink out the waters of the land and darken the skies with the smoke of a thousand burning fires, to conquest and snare the hearts and minds of men until they are but slaves of the machine, thus I become your sole master in the ever burning Earth.

I also check for fun gift ideas.

Only interaction I usually have with llms is when my cookies get cleared and my search result has the AI overview thing turned on. It never makes it through to actually presenting a result before I turn it off, though. Hate that shit being on ecosia. Really defeats the purported mission purpose, imo.

I played with a free image generator back when they first came out, found them fairly unimpressive, and moved on.

I’ll wait until they can proc gen vr environments without causing nausea. That’s about the only use I personally have for what presently passes for ai.

To experiment and just use it out of curiousity. I know it's capabilities and being able to manage, manipulate and change any writing, content or data I send it ... but that's the problem ... I don't trust any of their damned systems to have access to any amount of my personal data or information.

Lately I've been using ChatGPT in rotation with a bunch of other LLMs, since I don't want to habitually use just one and miss out on developments by others. I've found that I am gravitating towards using ChatGPT for language-related stuff - "what does slang term X mean", "could you write up a speech for a character to say", "create a detailed description of a magic item", stuff like that. I also sometimes ask it to generate images, though not so much now that the GPT image tool is available through Bing's image creation interface. I mostly use local AI image tools nowadays, they're much more controllable, but ChatGPT's images are often a great starting point.

I just recently came across Jules, an autonomous coding agent by Google, and was impressed with its free access level. I dumped a whole directory full of haphazard Python scripts I've been accumulating over the years into it and asked it to refactor them into less of a mess and it did a remarkable job turning the folder into somewhat of an actual application. You can hook it up to your Github account too, if you want, and it'll submit its changes as a new branch.

Just bear in mind that it'll make mistakes, I did have to do a thorough debugging run to make sure everything still worked the same. But the amount of grunt work it saved me was huge.

When I need to come across as someone from marketing or HR. The silliness works out in my favor then

I don't really use LLMs.

Today at work I was working on someone with something, and we realized our list of tuples was backwards. He was like "oh I'll have chatgpt fix it"

I was like, [(x[1], x[0]) for x in stuff]. Took about zero seconds. Delegating to the chatbot might have felt impressive if it got it right, but it was also such a trivial task I wouldn't think to use it.

Nothing. I hate it. It lies about easily found stuff, then is obnoxious and just. Eirgh. I'd rather a human lie to me about things, at least they're like. Human.

Boy I hope you really, really enjoy debilitating security holes if you're using GPT to write your backend JS

I use Kagi chat all the time for random shit:

• scheming about food forest ideas
• learning about Linux
• helping me get inspired to write DnD campaigns
• journaling about stressful shit and getting encouragement to relax and breathe through it
• inspirations for cool projects to do

it's like a great resource to just see like an average of what people would say in that situation. I think of it like a faster more focused reddit thread full of good and bad ideas, and I sort through them

Nothing, I use Gemini for personal stuff and Claude at work.

It's super helpful for figuring out whom I should block on here. When anybody says something like "I asked shatgpt and…" I immediately know that all of their opinions henceforth are to be discarded without further consideration.

name ideas for various things and a last resort after hours of troubleshooting

I only do cosmetic stuff in JS. Anything that needs to be secure is handled by perl.

Good man.

As a discipline challenge for when I search something. Do my eyes go up to the sparkle answer? Will I resist if they try to? Would I even expand the prompt? Will I cross-reference?
Sure, if I'm gonna stick around at the article anyway, it might be easy to resist the clanker's reply, and in bold-highlighted airing dates for Pluribus. That information is also too important to risk getting wrong anyway.
But if it's just an innocuous activity like foraging mushrooms, I might get weak and just sneak a little peak at sparkle's advice.

I use the name for rolling my eyes.

Why the hell would I want to use ChatGPT?