Ask Lemmy

I’ve only ever tried ChatGPT and that’s what I’ve stuck with.

Most of the time it does what I ask. My two main uses are editing my writing and answering random questions. I also use it to bounce ideas off, and honestly it’s the “adult” I talk to when I’m about to tear my hair out trying to have civil conversations online. I feel like an LLM is tuned better to my autistic wavelength than most people are. I rarely get to talk about my deepest interests with others because they’re usually not the kind of things your average Joe spends time thinking about - so rather than boring people to death or sharing thoughts with friends who won’t even respond, I find myself turning to ChatGPT more and more.

It’s interesting that I genuinely enjoy my interactions with an LLM more than with most people online. I’m not sure what that says about me, other people, or the LLM itself. Out of everything I do online, YouTube and ChatGPT give me by far the fewest regrettable minutes.

I use chatgpt for day to day stuff , grok and qwen for image generation

Ohh no you posted something semi positive about AI on Lemmy. Looks like everyone already did their duty and downvoted you for it. God forbid you have a use for it.

Personally I like the Qwen models mostly when I'm at home. They've been proven fairly competent. I run a sled hosted Tabby copilot alternative for vs code too that I've come to enjoy.

none and nothing.

Generate my personal firehose of falsehoods

Hey, if putin is allowed to have a firehose, I should get an even bigger firehose

Gemini, because I don't need a separate account.

Looking up things that may be hard to find otherwise, or at least attempting to. Hell, it once cited me a blog on website that didn't even have a domain name.

And it can be useful to quickly either remember or find how to use some program. Get the command from LLM, check what it does in the man page and use what's needed. Faster than just reading through a man page from scratch.

It's like 90% accurate, so it's often faster to just verify the information than research on my own (possibly due to how shitty search engines become).

On the other hand, with certain more obscure things it will just claim that such a thing doesn't exist, saying that there is a misunderstanding on human side. For example when I tried searching for a certain cheap data plan.

I think the only ethical and worthwhile LLM is Neuro-sama. It serves me by encouraging animators to make cartoons of its stream highlights.

I started like most with gpt and used it until I had no more hair left to pull out. Then recently I switched to using Claude and was blown away with how much better it was at everything compared to gpt.

The closest I get to using an LLM is Futo voice input.

Gemma 2.5 on ollama is all my lowly laptop can run.

It's good enough for what I want - summarize, translate, generate bullshit.

Depends on what you need it for. There's no single LLM model that's a superset of all others today.

CoPilot/Gemini Agent for mundane things like writing a test or a summary of a function. Honestly though with how often it's wrong, if you averaged the time saving against the time wasting, (either reviewing it's borked PR or scanning the description it's written in the IDE) it's probably breaking even. Meh over all.

I use different ones for different things. Perplexity answers my tech questions best (what is the dax formula for ___). Gemini is good if I'm making something that needs more casual language or some help brainstorming. Copilot is good for corporate jargon. It plays nice with other MS tools, so if I need to connect it to a PowerAutomate or Power app, I can do that. I used ChatGPT to make a lifetime movie plot generator with great success.

Running Gemma or Hermes through ollama mostly. I don't really use llms much but I'd much rather self host one when I do need it. I tried deepseek locally but even though it was super fast it the "reasoning" bs was enough to make me get rid of it.