Oof let’s see, what am I an expert in? Probably system design - I work at (insert big tech) and run a system design club there every Friday. I use ChatGPT to bounce ideas and find holes in my design planning before each session.
Does it make mistakes? Not really? it has a hard time getting creative with nuanced examples (i.e. if you ask it to “give practical examples where the time/accuracy tradeoff in Flink is important” it can’t come up with more than 1 or 2 truly distinct examples) but it’s never wrong.
The only times it’s blatantly wrong is when it hallucinates due to lack of context (or oversaturated context). But you can kind of tell something doesn’t make sense and prod followups.
Tl;dr funny meme, would be funnier if true
I ask AI shitbots technical questions and get wrong answers daily. I said this in another comment, but I regularly have to ask it if what it gave me was actually real.
Like, asking copilot about Powershell commands and modules that are by no means obscure will cause it to hallucinate flags that don’t exist based on the prompt. I give it plenty of context on what I’m using and trying to do, and it makes up shit based on what it thinks I want to hear.
That’s not been my experience with it. I’m a software engineer and when I ask it stuff it usually gives plausible answers but there is always something wrong. For example it will recommend old outdated libraries or patterns that look like they would work but when you try them out you figure out they are setup differently now or didn’t even exist.
I have been using windsurf to code recently and I’m liking that but it makes some weird choices sometimes and it is way too eager to code so it spits out a ton of code you need to review. It would be easy to get it to generate a bunch of spaghetti code that works mostly that’s not maintainable by a person out of the box.
My main experience with AI is that the pull requests I’ve reviewed have got at least twice as large, and I need to review the code much, much more carefully.
Yeah you just trade fun coding time for boring code reviews haha
I’ve been using o3-mini mostly for
ffmpegcommand lines. And a bit ofsed. And it hasn’t been terrible, it’s a good way to learn stuff I can’t decipher from the man pages.Totally didn’t misread that as ‘ffmpreg’ nope.
In my experience plain old googling still better.
I wonder if AI got better or if Google results got worse.
Bit of the first, lots of the second.
True, in many cases I’m still searching around because the explanations from humans aren’t as simplified as the LLM. I’ll often have to be precise in my prompting to get the answers I want which one can’t be if they don’t know what to ask.
And that’s how you learn, and learning includes knowing how to check if the info you’re getting is correct.
LLM confidently gives you easy to digest bite, which is plain wrong 40 to 60% of the time, and even if you’re lucky it will be worse for you.I’m in the kiddie pool, so I do look things up or ask what stuff does. Even though I looked at the man page for printf (printf.3 I believe), there was nothing about %*s for example, and searching for these things outside of asking LLM’s is some times too hard to filter down to the correct answer. I’m on 2 lines of code per hour, so I’m not exactly rushing.
Shell scripting is quite annoying to be sure. Thinking of learning python instead.
Are you me? I’ve been doing the exact same thing this week. How creepy.
we just had to create a new instance for coder7ZybCtRwMc, we’ll merge it back soon
come on guys, the joke is right there… 60% of the time it works, every time!
i mainly use it for fact checking sources from the internet and looking for bias. i double check everything of course. beyond that its good for rule checking for MTG commander games, and deck building. i mainly use it for its search function.
Most of my searches have to do with video games, and I have yet to see any of those AI generated answers be accurate. But I mean, when the source of the AI’s info is coming from a Fandom wiki, it was already wading in shit before it ever generated a response.
I’ve tried it a few times with Dwarf Fortress, and it was always horribly wrong hallucinated instructions on how to do something.
Exactly my thoughts.
I have frequentley seen gpt give a wrong answer to a question, get told that its incorrect, and the bot fights with me and insists Im wrong. and on other less serious matters Ive seen it immediatley fold and take any answer I give it as “correct”
ChatGPT is a tool. Use it for tasks where the cost of verifying the output is correct is less than the cost of doing it by hand.
Youre still doing it by hand to verify in any scientific capacity. I only use ChatGPT for philosophical hypotheticals involving the far future. We’re both wrong but it’s fun for the back and forth.
It is not true in general that verifying output for a science-related prompt requires doing it by hand, where “doing it by hand” means putting in the effort to answer the prompt manually without using AI.
You can get pretty in the weeds with conversions on ChatGPT in the chemistry world or even just basic lab work where a small miscalculation at scale can cost thousands of dollars or invite lawsuits.
I check against actual calibrated equipment as a verification final step.
I said not true in general. I don’t know much about chemistry. It may be more true in chemistry.
Coding is different. In many situations it can be cheap to test or eyeball the output.
Crucially, in nearly any subject, it can give you leads. Nobody expects every lead to pan out. But leads are hard to find.
I imagine ChatGPT and code is a lot like air and water.
Both parts are in the other part. Meaning llm is probably more native at learning reading and writing code than it is at interpreting engineering standards worldwide and allocation the exact thread pitch for a bolt you need to order thousands of. Go and thread one to verify.
This is possibly true due to the bias of the people who made it. But I reject the notion that because ChatGPT is made of code per se that it must understand code better than other subjects. Are humans good at biology for this reason?
Honestly, I’ve found it best for quickly reformatting text and other content. It should live and die as a clerical tool.
Which is exactly why every time I see big tech companies making another stupid implementation of it, it pisses me off.
LLMs like ChatGPT are fundamentally word probability machines. They predict the probability of words based on context (or if not given context, just the general probability) when given notes, for instance, they have all the context and knowledge, and all they have to do it predict the most statistically probable way of formatting the existing data into a better structure. Literally the perfect use case for the technology.
Even in similar contexts that don’t immediately seem like “text reformatting,” it’s extremely handy. For instance, Linkwarden can auto-tag your bookmarks, based on a predetermined list you set, using the context of each page fed into a model running via Ollama. Great feature, very useful.
Yet somehow, every tech company manages to use it in every way except that when developing products with it. It’s so discouraging to see.
I think that AI has now reached the point where it can deceive people ,not equal to humanity.
I just use it to write emails, so I declare the facts to the LLM and tell it to write an email based on that and the context of the email. Works pretty well but doesn’t really sound like something I wrote, it adds too much emotion.
That sounds like more work than just writing the email to me
Yeah, that has been my experience so far. LLMs take as much or more work vs the way I normally do things.
This is what LLMs should be used for. People treat them like search engines and encyclopedias, which they definitely aren’t
If it’s being designed to answer questions, then it should simply be an advanced search engine that points to actual researched content.
The way it acts now, it’s trying to be an expert based one “something a friend of a friend said”, and that makes it confidently wrong far too often.
Exactly this is why I have a love/hate relationship with just about any LLM.
I love it most for generating code samples (small enough that I can manually check them, not entire files/projects) and re-writing existing text, again small enough to verify everything. Common theme being that I have to re-read its output a few times, to make 100% sure it hasn’t made some random mistake.
I’m not entirely sure we’re going to resolve this without additional technology, outside of ‘the LLM’-itself.
First off, the beauty of these two posts being beside each other is palpable.
Second, as you can see on the picture, it’s more like 60%
No it’s not. If you actually read the study, it’s about AI search engines correctly finding and citing the source of a given quote, not general correctness, and not just the plain model
Read the study? Why would i do that when there’s an infographic right there?
(thank you for the clarification, i actually appreciate it)
40% seems low
Removed by mod
LLMs are actually pretty good for looking up words by their definition. But that is just about the only topic I can think where they are correct even close to 80% of the time.
Removed by mod
This, but for Wikipedia.
There’s an easy way to settle this debate. Link me a Wikipedia article that’s objectively wrong.
I will wait.
This, but for all media.
What topics are you an expert on and can you provide some links to Wikipedia pages about them that are wrong?
I’m a doctor of classical philology and most of the articles on ancient languages, texts, history contain errors. I haven’t made a list of those articles because the lesson I took from the experience was simply never to use Wikipedia.
Can you at least link one article and tell us what is wrong about it?
How do you get a fucking PhD but you can’t be bothered to post a single source for your unlikely claims? That person is full of shit.
The fun part about Wikipedia is you can take your expertise and help correct the information, that’s the entire point of the site
why don’t you then go and fix these quoting high quality sources? are there none?
Because some don’t let you. I can’t find anything to edit Elon musk or even suggest an edit. It says he is a co-founder of OpenAi. I can’t find any evidence to suggest he has any involvement. Wikipedia says co-founder tho.
https://openai.com/index/introducing-openai/
https://www.theverge.com/2018/2/21/17036214/elon-musk-openai-ai-safety-leaves-board
Tech billionaire Elon Musk is leaving the board of OpenAI, the nonprofit research group he co-founded with Y Combinator president Sam Altman to study the ethics and safety of artificial intelligence.
The move was announced in a short blog post, explaining that Musk is leaving in order to avoid a conflict of interest between OpenAI’s work and the machine learning research done by Telsa to develop autonomous driving.
He’s not involved anymore, but he used to be. It’s not inaccurate to say he was a co-founder.
Interesting! Cheers! I didn’t go farther than openai wiki tbh. It didn’t list him there so I figured it was inaccurate. It turns out it is me who is inaccurate!
There are plenty of high quality sources, but I don’t work for free. If you want me to produce an encyclopedia using my professional expertise, I’m happy to do it, but it’s a massive undertaking that I expect to be compensated for.
Many FOSS projects don’t have money to pay people
Do not bring Wikipedia into this argument.
Wikipedia is the library of Alexandria and the amount of effort people put into keeping Wikipedia pages as accurate as possible should make every LLM supporter be ashamed with how inaccurate their models are if they use Wikipedia as training data
With all due respect, Wikipedia’s accuracy is incredibly variable. Some articles might be better than others, but a huge number of them (large enough to shatter confidence in the platform as a whole) contain factual errors and undisguised editorial biases.
It is likely that articles on past social events or individuals will have some bias, as is the case with most articles on those matters.
But, almost all articles on aspects of science are thoroughly peer reviewed and cited with sources. This alone makes Wikipedia invaluable as a source of knowledge.
TBF, as soon as you move out of the English language the oversight of a million pair of eyes gets patchy fast. I have seen credible reports about Wikipedia pages in languages spoken by say, less than 10 million people, where certain elements can easily control the narrative.
But hey, some people always criticize wikipedia as if there was some actually 100% objective alternative out there, and that I disagree with.
Fair point.
I don’t browse Wikipedia much in languages other than English (mainly because those pages are the most up-to-date) but I can imagine there are some pages that straight up need to be in other languages. And given the smaller number of people reviewing edits in those languages, it can be manipulated to say what they want it to say.
I do agree on the last point as well. The fact that literally anyone can edit Wikipedia takes a small portion of the bias element out of the equation, but it is very difficult to not have some form of bias in any reporting. I more use Wikipedia as a knowledge source on scientific aspects which are less likely to have bias in their reporting
Idk it says Elon Musk is a co-founder of openAi on wikipedia. I haven’t found any evidence to suggest he had anything to do with it. Not very accurate reporting.
Isn’t co-founder similar to being made partner at a firm? You can kind of buy your way in, even if you weren’t one of the real originals.
That is definitely how I view it. I’m always open to being shown I am wrong, with sufficient evidence, but on this, I believe you are accurate on this.
The company counts Elon Musk among its cofounders, though he has since cut ties and become a vocal critic of it (while launching his own competitor).
If this were true, which I have my doubts, at least Wikipedia tries and has a specific goal of doing better. AI companies largely don’t give a hot fuck as long as it works good enough to vacuum up investments or profits
Your doubts are irrelevant. Just spend some time fact checking random articles and you will quickly verify for yourself how many inaccuracies are allowed to remain uncorrected for years.
Small inaccuracies are different to just being completely wrong though
The obvious difference being that Wikipedia has contributors cite their sources, and can be corrected in ways that LLMs are flat out incapable of doing
Really curious about anything Wikipedia has wrong though. I can start with something an LLM gets wrong constantly if you like
