Ask Lemmy

I think its great for inspiration but your final product should never be raw AI/LLM output

I've always said I think it's fine in filler content, it can allow small teams to quickly populate their world with background stuff that you never notice. Except when it's not there.

But with great power comes great responsibility. And I don't necssesarily think most can handle that.

I have used copilot a couple times to be like "I have this scenario and want to do this. What are my options?". I'd rather have a good Internet search and real people, but that's all shitted up.

The answers from the LLM aren't even consistently good. If I didn't know programming I wouldn't be able to use this information effectively. That's probably why a lot of vibe coding is so bad.

GenAI is a plagiarism machine. If you use it, you're complicit.

Ethics aside, LLMs in particular tend to "hallucinate". If you blindly trust their output, you're a dumbass. I honestly feel bad for young people who should be studying but are instead relying on ChatGPT and the likes.

It speeds up my dev time dramatically. I know what I want to do, I have an idea of how I want to do it. LLM generates boilerplate code I review. I tweak it. I fix the bug. If there is something I don't understand, I ask sources to review the output. I test it. Then I'll submit it for peer review once I'm happy with the code and the output.

If it truly helps you, I think that might be enough for me. I say truly because you need to use an AI with responsibility to not ruin yourself. Like, don't let it think for you. Don't trust everything it says.

I use it a lot when applying for jobs, something I've struggled with on and off for 12 years. I suck and writing the cover letter and CV. It takes me 2-3 days to update a cover letter for a job because it takes so much energy. With AI that is down to 1-2 days.

It's also great for explaning things in other words of if you're trying to look up something that's hard to search for, I don't have any examples tho.

I used to use it to help me formulate scentences since english isn't my first language. Now I instead use Kagi Translate.

It's as useful as a rubber duck. Decent at bouncing ideas off it when no one is available, or you can't be bothered to bother people about dumb ideas.

But at the moment, no, it's not justifiable as it directly fuels oligarchies, fascism in the US, and tech bros. Perhaps when the bubble pops.

I read that they're not terrible when used to power NPC's in games.

https://www.pcgamer.com/software/ai/company-that-makes-generative-ai-powered-npcs-reports-that-95-percent-of-players-enjoy-their-generative-ai-powered-npcs/

Not my personal take, mind you, but thought it relevant.

The best use of AI I've seen thus far is reading legislative bills. Those monstrosities are so fucking long and filled with earmarks that it's next to impossible to understand what is in them.

Having an AI not only read the bill but keep a watch of it as it goes through Congress is probably the best use of AI because it actually helps citizens.

I am on record saying we need an AI that can track prices of various things that can then predict when the best time it is to buy something.

I want an AI bot that saves me money or gets me a good deal or extracts money from the capital class.

LLM's have their use, there is no doubt about that. I'm in the middle of creating a home brew campaign for my D&D group and unfortunately I'm a lousy artist and I wanted a few things visualized. Well, I used a photo generating AI to create something that had the visual I wanted. I'm going to use it for my campaign and it will probably just sit on my hard drive after I'm done.

My employer is rolling out AI and is asking us to find places to insert it into our workflows. I am doing that with my team, but none of us are really sure if it will be of any benefit.

The problem right now is we're at the stage where idiots are convinced it is something that it is not and they have literally thrown 10's of billions of dollars at it. Now... They are staring at the wide abyss that is the amount of money they invested vs the amount of money people are willing to pay for it.

I've seen arguments for and against the presence of an AI bubble... Personally, I think it's a bubble that's so large that it will take down several long established computer industry manufacturers when if pops. Those that are arguing its absence probably have large investments that they do not want to see fail.

1. The sciences obviously

2. For me personally, data collation

3. Learning

4. Assisting with Linux sysadmin stuff (used to be a "how do I X" meant hours of scouring online forums and asking questions that might be deleted because draconian forum rules or get answered weeks later if at all, now I can get shit done in minutes)

   5. I also use it a lot to explore ideas and arguments, like a sort of metaphysical sparring partner.

My current list of reasons why you shouldn't use generative AI/LLMs

A) because of the environmental impacts and massive amount of water used to cool data centers https://news.mit.edu/2025/explained-generative-ai-environmental-impact-0117

B) because of the negative impacts on the health and lives of people living near data centers https://www.bbc.com/news/articles/cy8gy7lv448o

C) because they're plagiarism machines that are incapable of creating anything new and are often wrong https://knowledge.wharton.upenn.edu/article/does-ai-limit-our-creativity/ https://www.plagiarismtoday.com/2024/06/20/why-ai-has-a-plagiarism-problem/

D) because using them negatively affects artists and creatives and their ability to maintain their livelihoods https://www.sciencedirect.com/science/article/pii/S2713374523000316 https://www.insideradio.com/free/media-industry-continues-reshaping-workforce-in-2025-amid-digital-shift/article_403564f7-08ce-45a1-9366-a47923cd2c09.html

E) because people who use AI show significant cognitive impairments compared to people who don't https://www.media.mit.edu/publications/your-brain-on-chatgpt/ https://time.com/7295195/ai-chatgpt-google-learning-school/

F) because using them might break your brain and drive you to psychosis https://theweek.com/tech/spiralism-ai-religion-cult-chatbot https://mental.jmir.org/2025/1/e85799 https://youtu.be/VRjgNgJms3Q

G) because Zelda Williams asked you not to https://www.bbc.com/news/articles/c0r0erqk18jo https://www.abc.net.au/news/2025-10-07/zelda-williams-calls-out-ai-video-of-late-father-robin-williams/105863964

H) because OpenAI is helping Trump bomb schools in Iran https://www.usatoday.com/story/opinion/columnist/2026/03/06/openai-pentagon-tech-surveillance-us-citizens/88983682007/

I) because RAM costs have skyrocketed because OpenAI has used money it doesn't have to purchase RAM from Nvidia that currently doesn't exist to stock data centers that also don't currently exist, inconveniencing everyone for what amounts to speculative construction https://www.theverge.com/news/839353/pc-ram-shortage-pricing-spike-news

J) because Sam Altman says that his endgame is to rent knowledge back to you at a cost https://gizmodo.com/sam-altman-says-intelligence-will-be-a-utility-and-hes-just-the-man-to-collect-the-bills-2000732953

K) because some AI bro is going to totally ignore all of this and ask an LLM to write a rebuttal rather than read any of it.

Medicine.

Evidence shows that some highly specialised models are better at things like detecting breast cancer in scans than human doctors.

Properly anonymised automatic second scans by an AI to catch the markers that human doctors miss for another review by a specialist is an excellent potential use case for an LLM AI.

Transcription services can save doctors huge amounts of admin time and allows them to focus on the patient if they know there's a reliable system in place for typing up notes for a consultation. As long as it's treated as a "please review these notes are accurate" rather than treated as a gospel recording and the data is destroyed once it's job is complete and the patient has been able to give informed consent.

The way these things are being used in actual medical contexts right now is frankly terrifying.

I feel like self hosting LLMs and GenAi is slightly better for the environment, definitely less environmental impact than gaming.

It just these massive datacenters and models. If people can just be a little more patient and specialized with their AI usage it saves so much electricity

I think anything with text generation is fine. Your multiple Google searches are highly likely to eat more resources than that. Also, fuck Google, use Ecosia. But when I suspect an answer isn't one quick search away, I happily rather use Le Chat for answers, than give Reddit traffic, or have to wade through the shite that is Fandom, Wikia or whatever. Not to mention using AI helps me get past the issue of having to check multiple sites for an answer, just to find that the answer is "Google it" or "Nvm, solved it". Some of you fuckers did this.

However people need to understand that an AI is exactly as fallible as any person. Yes, it has access and capability to handle way more data but between trying to please you and just it getting it's wires crossed, it's going to make mistakes. YOU need to be able to assess the accuracy of the output. The more important the topic, the more careful you need to be and always assume that the possibility of error is there no matter how hard you try - JUST LIKE WITH ANY BIT OF INFORMATION. I see so many people cite academic articles like they prove whatever claim they are making, just to see that the study in question was funded by The Company That Wants to Prove The Claim and sample size was 3 people who work for The Company That Wants to Prove The Claim. At least AI has a small chance of pointing the issue out if YOU yourself tell it to be critical - and I actually suspect this is part of the reason some people hate AI. They don't like that it absolutely can be more intellectually rigorous than a person with an emotional investment in whatever they want to be true. Yes, you can have an AI asspat your grandest delusions but if you actually try to get it to be critical, it will be. You can use a hammer to hit people, or you can use it on a nail as intended (and how many times you hit your own fingers is on you, not the hammer).

I would draw a line on artwork, videos, music. While I'm not going to crucify actual artists using AI assistance to take out some tedium from a project, I still wouldn't encourage it. Stolen artwork to train AI is one thing and the environmental impact is VASTLY greater than just text. Generating one AI image can use as much energy as even a 1,000 text responses. I would also really like to be able to completely opt out of AI slop in media sites. I fucking hate that Soundcloud allows it.

And a last point on AI text responses: if you saw the rise of alt-right and the anti-vaxx stuff, you probably are familiar with gish galloping and Brandolini's Law. If not, you really fucking should be. AI can make it so much easier to debunk misinformation. YES it can make it easier to perpetuate too but this is where we see the AI weapons race. Bad actors can AND WILL use AI to fill any void with their rhetoric. If you value truth and facts and want to prevent misinformation from spreading you are gimping yourself if you're not using AI.