My job uses a data science platform that has a special ai assistant trained on its own docs.
The first time I tried using it, it used the wrong language. The second time I used it, it was hallucinating its own functions, but after looking up the docs I told it what function to use and it gave me code that worked
I have not used it a third time. I don’t think i will.
What are you talking about? We mention this on a daily basis. That's the #1 complaint about ChatGPT when used for factual purposes
Reminder that all these Chat-formatted LLMs are just text-completion engines trained on text formatted like a chat. You're not having a conversation with it, it's "completing" the chat history you're providing it. By randomly(!) choosing the next text tokens that seems like they best fit the text provided.
If you don't directly provide, in the chat history and/or the text completion prompt, the information you're trying to retrieve, you're essentially fishing for text in a sea of random text tokens that seems like it fits the question.
It will always complete the text, even if the tokens it chooses minimally fit the context, it chooses the best text it can but it will always complete the text.
This is how they work, and anything else is usually the company putting in a bunch of guide bumpers to reformat prompts into coaxing the models to respond in a "smarter" way (see GPT-4o and "chain of reasoning")
I beg someone to help me. There is this new guy at my workplace, officially as a developer who can't write code at all. He has pasted an entire project I did into ChatGPT with "optimize this" and pull requested it. I swear.
Report up the chain, if it's safe to do so and they are likely to understand.
Also, check what your company's rules regarding data security and LLM use are. My understanding is that at many places putting private company or customer data into an outside LLM is seen as shouting company secrets out to the open internet. At least that's the policy where I'm at. Pasting an entire project in would definitely violate things for my workplace.
In general that's rude as hell. New guy comes in, grabs an entire project they have no background with, and just chucks it at an LLM? No actual review of it themselves, just an assumption that your code is so shit that a general use text generator will do better? Doesn't sound like a "team player" to me (management eats that kind of talk up).
Maybe couch it as "I want to make sure that as a team, we're utilizing the tools available to us in the best way possible to multiply our strengths. That said, I'm concerned the approach that [LLM idiot] is using will only result in more work for the team. Using chatGPT as he has is an explosive approach, when I feel that a more scalpel-like approach to address specific areas for improvement would be the best method moving forward. We should be using these tools to address specific concerns, not chucking everything at the wall in some never ending chase of an undefined idea of 'more optimized'."
Perhaps frame it in terms of man hours? The immediateness of 5 minutes in chatGPT can cost the team multiple workdays in reviewing the output, whereas more focused code review up front can reduce the man hour cost significantly.
There's also a bunch of articles out there online about how overuse of LLMs is leading to a measurable decrease in code quality and increase in security issues in code bases.
Such a great answer, thank you lots!
It depends upon what you use ChatGPT for and if you know how to use it productively. For example if I ask ChatGPT coding questions it is often very helpful. If I ask it history questions it constantly makes things up. You also again need to know how to use it, like people who claim ChatGPT is not helpful for coding you ask them how they use it and they basically just ask ChatGPT to do their whole project for them and when it fails they claim it is useless. But that's not the productive way to use it, the productive way to use it is like a replacement for StackOverflow or to provide you examples of how to use some library, or things like that, not doing your whole project for you. Of course, people often use it incorrectly so it's probably not a good idea to allow its use in the workplace, but for individual use it can be very helpful.
For coding it heavily depends on the language. For example, it's quite decent at writing C#, but whenever I try to ask it any question about rust, it's either flat out wrong or doesn't even fucking compile.
Also found it most useful when I know exactly what I want, just don't know the syntax. Like when I was writing C# code generation for the first time. Also unsurprisingly sucks at working with libraries.
And thank god it doesn't get them all the way there, because if it were able to completely do everything accurately with the level of ambiguous prompts the layperson gives it, anyone technical would essentially be out of a job.
And honestly, the world would be better off not making people complacent just being end users of everything, and instead have to have a modicum of understanding what they are doing.
I used to think its just neophobia having all these kids using smart phones and touch screens for everything at increasingly earlier ages, but its like they only know how to use/consume things, never an inkling of trying to tinker with things and understand how to repurpose the mechanisms , figure out how things work (tbf everything now is super integrated, much harder to repair).
It just doesn't bode well to me when it seems like the future labor force is so disconnected from the underlying systems they use.
I used it today to find out how to do something on my Juniper that would have taken 45 minutes of sifting bullshit documentation. One question and I figured it out in 2 minutes.
This is similar to gabe Newell's idea of piracy. This is a convenience issue. And GPT solves some of it.
Treat it like a janitor rather than an answer machine and you'll have a better time. I call it my bitch bot.
I just tried out Gemini.
I asked it several questions in the form of 'are there any things of category x which also are in category y?' type questions.
It would often confidently reply 'No, here's a summary of things that meet all your conditions to fall into category x, but sadly none also fall into category y'.
Then I would reply, 'wait, you don't know about thing gamma, which does fall into both x and y?'
To which it would reply 'Wow, you're right! It turns out gamma does fall into x and y' and then give a bit of a description of how/why that is the case.
After that, I would say '... so you... lied to me. ok. well anyway, please further describe thing gamma that you previously said you did not know about, but now say that you do know about.'
And that is where it gets ... fun?
It always starts with an apology template.
Then, if its some kind of topic that has almost certainly been manually dissuaded from talking about, it then lies again and says 'actually, I do not know about thing gamma, even though I just told you I did'.
If it is not a topic that it has been manually dissuaded from talking about, it does the apology template and then also further summarizes thing gamma.
...
I asked it 'do you write code?' and it gave a moderately lengthy explanation of how it is comprised of code, but does not write its own code.
Cool, not really what I asked. Then command 'write an implementation of bogo sort in python 3.'
... and then it does that.
...
Awesome. Hooray. Billions and billions of dollars for a shitty way to reform web search results into a coversational form, which is very often confidently wrong and misleading.
Idk why we have to keep re-hashing this debate about whether AI is a trustworthy source or summarizer of information when it's clear that it isn't - at least not often enough to justify this level of attention.
It's not as valuable as the marketing suggests, but it does have some applications where it may be helpful, especially if given a conscious effort to direct it well. It's better understood as a mild curiosity and a proof of concept for transformer-based machine learning that might eventually lead to something more profound down the road but certainly not as it exists now.
What is really un-compelling, though, is the constant stream of anecdotes about how easy it is to fool into errors. It's like listening to an adult brag about tricking a kid into thinking chocolate milk comes from brown cows. It makes it seem like there's some marketing battle being fought over public perception of its value as a product that's completely detached from how anyone actually uses or understands it as a novel piece of software.
Probably it keeps getting rehashed because people who actually understand how computers work are extremely angry and horrified that basically every idiot executive believes the hype and then asks their underlings to inplement it, and will then blame them for doing what they asked them to do when it turns out their idea was really, unimaginably stupid, but idiot executive gets golden parachute and software person gets fired.
That, and/or the widespread proliferation of this bullshit is making stupid people more stupid, and just making more people stupid in general.
Or how like all the money and energy spent on this is actively murdering the environment and dooming the vast majority of our species, when it could be put toward building affordable housing or renovating crumbling infrastructure.
Don't worry, if we keep throwing exponential increasing amounts of effort at the thing with exponentially diminishing returns, eventually it'll become God!
to fool into errors
tricking a kid
I've never tried to fool or trick AI with excessively complex questions. When I tried to test it (a few different models over some period of time - ChatGPT, Bing AI, Gemini) I asked stuff as simple as "what's the etymology of this word in that language", "what is [some phenomenon]". The models still produced responses ranging from shoddy to absolutely ridiculous.
completely detached from how anyone actually uses
I've seen numerous people use it the same way I tested it, basically a Google search that you can talk with, with similarly shit results.
Why do we expect a higher degree of trustworthiness from a novel LLM than we de from any given source or forum comment on the internet?
At what point do we stop hand-wringing over llms failing to meet some perceived level of accuracy and hold the people using it responsible for verifying the response themselves?
Theres a giant disclaimer on every one of these models that responses may contain errors or hallucinations, at this point I think it's fair to blame the user for ignoring those warnings and not the models for not meeting some arbitrary standard.
copilot did the same with basic math. just to test it I said "let's say I have a 10x6 rectangle. what number would I have to divide width and height by, in order to end up with a rectangle that's half the area?"
it said "in order to make it half, you should divide them by 2. so [pointlessly lengthy steps explaining the divisions]"
I said "but that would make the area 5x3 = 15 units which is not half the area of 60"
it said "you're right! in order to ... [fixing the answer to √2 using approximation"
I don't know if I said it then, or after some other fucking nonsense but when I said "you're useless" it had the fucking audacity to take offense and end the conversation!
like fuck off, you don't get to have fake pride if you don't have basic fake intelligence but use it in your description.
Its a perfect encapsulation of the corpo mindset:
Whatever I do is profound, meaningful, with endless possibilities for future greatness...
... even though I'm just talking out of my ass 99% of the time...
... and if you have the audacity, the nerve, to have a completely normal reaction when you determine that that is what I am doing, pshaw, how uncouth, I won't stand for your abuse!
...
They've done it. They've made a talking (not thinking) machine in their own image.
And it was not good.
You start a conversation you can't even finish it You're talkin' a lot, but you're not sayin' anything When I have nothing to say, my lips are sealed Say something once, why say it again?
Psycho Killer Qu'est-ce que c'est
Because of I haven't found anyone asking the same question on a search index, ChatGPT won't tell me to just use Google or close my question as a duplicate when it's not a duplicate.
The only reason i use ChatGPT for some quick stuff is just that search engines suck so bad.
Perplexity (or open source equivalents) are much better for this.
All tools get misused.
Depending on the task, it’s quicker to verify the AI response than work through the blank page phase.
chatgpt has been really good for teaching me code. As long as I write the code myself and just ask for clarity or best practices i haven't had any bad hallucinations.
For example I wanted to change a character in an array with another one but it would give some error about data types that were way out of my league. Anyways apparently I needed to run list(string) first even though string[5] will return the character.
However that's in python which I assume is well understood due to the ton of stackoverflow questions and alternative docs. I did ask it to do something in Google docs scripting something once and it had no idea what was going on and just hoped it worked. Fair enough, I also had no idea what was going on.
The reason why string[5] = '5' doesn't work is that strings in Python are immutable (cannot be changed). By doing list(string) you are actually creating a new list with the contents of the string and then modifying the list.
I wonder if ChatGPT explains this or just tells you to do this... as this works but can be quite inefficient.
To me this highlights the danger with using AI... sure you can complete a task, but you may not understand why or learn important concepts.
Yeah, it's a gift and a curse for exploring a new domain. It can help you move faster, but you'll definitely loose some understanding you'd get from struggling on those topics longer.
bold of u to assume there are docs
Or docs are far too extensive... reading imagemagick docs is like reading through some old tech wizard's personal diary.. "i was inspired to shape this spell like this because of such and such...." like, bro.. come on, I just want the command, the args, and some examples... 🤷♂️
I’m convinced people who can’t tell when a chat bot is hallucinating are also bad at telling whether something else they’re reading is true or not. What online are you reading that you’re not fact checking anyway? If you’re writing a report you don’t pull the first fact you find and call it good, you need to find a couple citations for it. If you’re writing code, you don’t just write the program and assume it’s correct, you test it. It’s just a tool and I think most people are coping because they’re bad at using it
They don't give you the answer, they give you a rough idea of where to look for the answer.
I've used them to generate chunks of boilerplate code that was 80% of what I needed, because I knew what I needed and wanted to save time.
There are ways of doing that which dont require burning an acre of rainforest
Yep. The overwhelming majority of IDEs have support for making templates/snippets.
VScode/VScodium has a very robust snippet system where you can set parts as "fill in the blank" that you can tab between, with optional drop down menus for choices. You can even link different "fill in" sections so you can do stuff like type in an argument name and have it propagate that same name through multiple places in your snippet.
If that's too much, how the fuck can any dev (or even someone hacking together scripts) survive without at least one file of common shit they made before that they can copy paste from? I really feel like that's bare minimum.
Either it's boilerplate you can already copy from somewhere else (documentation or previous work), or it's something you should probably review (at least briefly) and make into a template or snippet you can copy and paste later. That's part of the magic of programming: you get to build your own toolbox over time.
sigh people do talk about this, they complain about it non-stop. These same people probably aren't using it as intended, or are deliberately trying to farm a "gotcha" response. AI is a very neat tool which can do a lot of things well, but it's important to recognize its limitations. I don't use it for things I don't understand because I won't recognize if it's spitting out nonsense, but for topics I do understand it's hard to overstate how efficient and time saving it is.
The FuckAI people are valid for their concerns.
Unfortunately, their anger seems to constantly be misdirected at the weirdest things, instead of root issues.
People tweeting stuff. We allow tweets from anyone.
RULES:
