this post was submitted on 05 Jun 2025
592 points (97.6% liked)
People Twitter
7251 readers
790 users here now
People tweeting stuff. We allow tweets from anyone.
RULES:
- Mark NSFW content.
- No doxxing people.
- Must be a pic of the tweet or similar. No direct links to the tweet.
- No bullying or international politcs
- Be excellent to each other.
- Provide an archived link to the tweet (or similar) being shown if it's a major figure or a politician.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
it only takes a couple times of getting a made-up bullshit answer from chatgpt to learn your lesson of just skip asking chatgpt anything altogether
I feel like a lot of people in this community underestimate the average person's willingness to trust an AI. Over the past few months, every time I've seen a coworker ask something and search it up, I have never seen them click on a website to view the answer. They'll always take what the AI summary tells them at face value
Which is very scary
That's what people get when they ask me questions too but they still bother me all the time so clearly that's not going to work.
But chatgpt always gives such great answers on topics I know nothing at all about!
Oh yeah, AI can easily replace all the jobs I don't understand too!
Gell-mann amnesia. Might have to invent a special name for the AI flavour of it.
I've only really found it useful when you provide the source of information/data to your prompt. E.g. say you want to convert one data format to another like table data into JSON
It works very consistently in those types of use cases. Otherwise it's a dice roll.
I was using it to blow through an online math course I'd ultimately decided I didn't need but didn't want to drop. One step of a problem I had it solve involved finding the square root of something; it spat out a number that was kind of close, but functionally unusable. I told it it made a mistake three times and it gave a different number each time. When I finally gave it the right answer and asked, "are you running a calculation or just making up a number" it said that if I logged in, it would use real time calculations. Logged in on a different device, asked the same question, it again made up a number, but when I pointed it out, it corrected itself on the first try. Very janky.
So it forced you to ask it many times? Now imagine that you paid for it each time. For the creator then, mission fucking accomplished.
ChatGPT doesn't actually do calculations. It can generate code that will actually calculate the answer, or provide a formula, but ChatGPT cannot do math.
It's just like me fr fr
You need multi-shot prompting when it comes to math. Either the motherfucker gets it right, or you will not be able to course correct it in a lot of cases. When a token is in the context, it's in the context and you're fucked.
Alternatively you could edit the context, correct the parameters and then run it again.
On the other side of the shit aisle
Shoutout to my man Mistral Small 24B who is so insecure, it will talk itself out of correct answers. It's so much like me in not having any self worth or confidence.
Or you could learn to use the tool better and ask better questions. It's pretty decent at some things, absolutely terrible for others.
Asking to explain something like shorting a stock is one of the better uses, since there are tons of relevant posts explaining exactly that.
It's a good tool so long as there are already better ways to get your answer
It's good if the answers exist, but you don't know how to find them. They're like search engines that can generated related terms. or regurgitate common answers.
I find LLMs help me use existing search engines a lot better, because I can usually get it to spit out domain-specific terms to a more general query.
Forget trying to say anything positive about LLM or AI. The Lemmyverse downvotes any positive comment related to AI.
I'm well aware, but I think clearing misconceptions is valuable, and since I'm getting a fair amount of votes in both directions and discussion, hopefully that means people have read and considered my point.
I'm not going to recommend people use LLMs for everything or even claim that they're perfect for everyone (in fact, I don't like using them), just that they do have valid uses and, if it comes up, can be used efficiently (i.e. not burn down the planet). I use them a handful of times in a given week, often less, and mostly to get more keywords to search on a traditional search engine.
So yeah, it's whatever. I very much dislike both extremes here, and am trying to drive home the point that there is a happy middleground.
Oooooooor, I could use something mostly reliable instead of something that's sketch.
Nono you see... why be unsure when you could be wrong? 🤔
Why not both? Use the LLM to refine what you're looking for, and better sources for details. It's like skimming summaries in search results before picking a web page, or asking friends/family before looking up actual sources.
LLMs are great for interactive information retrieval and figuring out what information you actually need. They don't do everything, but they do a lot more than detractors claim and lot less than proponents claim. Find that happy middle-ground and it'll be a great tool.
LLMs are great for tech bros and CEOs who want maximum profit with minimum effort all while stealing work that isn't theirs and poisoning the planet at the same time.
They're also great for non-tech bros who just want to get stuff done, and they don't have to poison the planet at all. We run a few in our office on a Mac Mini, which sips power.
Those tech bros and CEOs are mostly fleecing investors, so I guess I'm not very concerned about them.
Why though? If I can't trust them half the time and I have to figure out when that is, nah.
So asking family and friends about things you don't know isn't worth it? Reading personal blogs isn't worth it?
As long as you go in knowing what it offers, it can be a great tool, like those little summaries on search results. I use them to find information when the search engine isn't going to be a great option, or at my day job to generate code for a poorly documented library. I tend to use it a handful of times per week, if that, and only for things that I know how to verify (i.e. manually test, confirm w/ official sources, etc).
As I said in another comment, searching for what's right or wrong takes a lot of time. Using it as a tool that is untrustworthy isn't great. For simple shit, maybe? But then simple shit is covered by most other places.
Then you're using it wrong.
LLMs shouldn't be used to search for what's right or wrong, they should be used to quickly get a wide breadth of information on a given topic or provide a starting point on code or text projects that you can refine later.
For example, I wanted to use a new library in a project, and the documentation was a bit cryptic and examples weren't quite right either. So I asked an LLM tuned for coding tasks to generate some example code for our use case (described the features I wanted to use), and the code worked. I needed to tweak some stuff (like I would w/ any example), but it worked. I used the LLM because I knew there would be a bunch of public code projects using this library with a variety of use cases, and that the LLM would probably do a reasonable job of picking out a decent one to build from, and I was right.
On another topic, I needed to do research on an unfamiliar topic, so I asked the LLM to provide a few examples in that domain w/ brief descriptions. That generated a ton of useful keywords that I used to find more reputable sources (keywords that would've taken hours of searching to generate), so I was able to quickly find reliable information starting from a pretty vague notion of what I wanted to learn.
LLMs have a lot of limitations, but if they're used to accomplish common tasks quickly, they can be incredibly useful. I don't think they'll replace anyone's job (unless your job is already pointless), traditional search engines (as much as Google wants it to), or anything like that, but they are useful tools that can make some of the more annoying parts of my job more efficient.
LLMs have flat out made up functions that don't exist when I've used them for coding help. Was not useful, did not point me in a good direction, and wasted my time.
Sure, they certainly can hallucinate things. But some models are way better than others at a given task, so it's important to find a good fit and to learn to use the tool effectively.
We have three different models at work, and they work a lot differently and are good at different things.
You need to actively have the relevant code in context.
I use it to describe code from shitty undocumented libraries, and my local models can explain the code well enough in lieu of actual documentation.
Not everyone has a bank of experts waiting for whatever questions they have.
If you have better options, fine. But if you are spending two hours googling instead of just asking chatgpt to spend five minutes finding you mostly the same links, you are just wasting time. Its easy to have it pull sources that you can quickly verify and then base your actually documentation on. FYI, I personally do both (I search while it does).
Except these "experts" are wrong a lot, so you can't trust them. It's the confidently wrong that's problematic.
Yes, you are still expected to participate and verify what is said. I also dont copy paste stuff from websites without verification since god knows the internet in general isn't always right either.
It's a productivity tool meant to help you, not do the job for you.
Except, I'm constantly trying to figure out what's right or wrong. At least wikipedia has people fighting about the truth, LLMs just state incorrect shit as truth and then stares at you.
Ask chatgpt the question while outlining your need for sources and direct quotes. (30 secs)
The bot searches. (30 secs to 10 minute depending on if deep research is used. This is free time you can spend doing something else productive, even the same or similar tasks on your end)
Click the links and ctrl-f to find the quote or keywords if its a large document. Verify it isn't bullshit. (1 min)
4a. It is valid, the link and relevant quotes are added to a work document to be used later (1 min)
4b. It is not valid, the legwork must be done yourself (10-20 minutes, maybe more)
Maybe its because I'm coming from a research perspective, where you need to put sources on everything. I would never take something chatgpt gave me at face value and dump it into a doc. That being said, I feel like the argument boils down to "since the tool can be used stupidly, I won't try to use it properly".
And there are many uses for a variety of things. I have it build me summaries of papers when I make a bank of them for example. I just finished reading the paper, so I can verify and modify the summary. I could write it but I dont want to be bothered trying to figure out the best way to give the most info in one paragraph. Chatgpt already writes better hyper condensed blocks of texts then me anyways.
Its good at making tables too and its hard to make mistakes when I'm giving it all the data in the first place.
How does it help productivity if you have to redo all of its work?
How did you get "redo" from "verifying sources"? Even if only 10% is good in any case (very rare but it is better at some subjects then others), that is still 10% you don't have to do. What we currently have is also the worst it's going to be.
Keep burying your head in the sand but you will just become the laughing stock of your office. You guys are aiming to become this decades "boomer that types with one finger".
Just the amount of time I save when I ask it to build me tables I can drop into documents is worth it. It's my information, I just don't have to copy paste it one by one into excel. People that are anti-AI in professional contexts are actually nuts.
That happy middle ground is so effective, so sad to see people dismissing it as useless. Also sad that people use it for everything and believe it all the time.
Great example of a good use case was using it to spell check my thesis, asking it to explain why it was wrongly spelled and then using my own brain, think if it is correct. That way I am more sure I catch things that I may have missed (suprise, I missed a lot and got helped a lot). Yes 30% was plain wrong but I just ignored that.
What's the benefit of example code in a project if you have to change it all anyway? What's the benefit of an article summary if you end up reading the article afterward? What's the benefit of a college course that just teaches you how to learn on your own (i.e. most of them)?
LLMs are great for some things, terrible for others, just like any tool. Use them for what they're good at (generating example code, getting an intro to a topic, etc) and not what they're poor at (greenfield projects, hard questions, etc). As you use the tool more, you'll get a feel for what it's good at and what it's not.
This is it. I remember 15 to 20 years ago there were people who were GOOD at using Google and people who were bad at it. The difference was knowing what the search engine was good at and how to ask it the right questions.
People who just went in blindly and clicked on the first link didn’t get what they needed out of it. Did that make it a bad tool? Probably not.
Using a LLM wrong is just as bad as using the wrong size wrench. Would a giant crescent wrench get a tiny nut off a bolt? Probably, but it would make more sense to use the right size one in the first place.
I stopped using it when I asked who I was and then it said I was a prolific author then proceeded to name various books I absolutely did not write.
Why the fuck would it know who you are?
If you have an account, you can tell it things about yourself. I used my boss's account for a project at work (felt gross). I made the mistake of saying "good morning" to it one day, and it proceeded to ask me if I was going to do (activities related to my boss's personal life - and the details were accurate). I was thinking, "why does he tell it so much about himself?"
So it's working as intended.
and I'm apparently a famous Tiktoker and Youtuber.
I just read "The Autobiography of QueenHawlSera"!
Have I been duped?