![Great job 👏 [Twonks]](https://lemmy.world/pictrs/image/69fd8e3b-4aea-4e91-addf-bd8fafc25787.jpeg){kind=link}
cross-posted from: https://lemmy.world/post/49392677
Twonks | Bluesky
Transcript
TW😶NKS
A comic in four panels:
Panel 1. White text on black
AI Design Logic
Panel 2. A guy sits in a restaurant at a table with a checkered table cloth. A waiter stands near, hands behind back waiting attentively.
Guy: Get me a cheese pizza
Panel 3. The waiter returns with a pizza in hand.
Panel 4. The guy gestures proudly at the pizza. The waiter looks less than amused.
Guy: Wow, look what I made!
Not everyone sees it this way because it;s a fucking stupid take, but didn’t worry Lemmy has a massive anti-ai hard-on so your mindless groupthink is safe here
Just because you ignore the reasons, doesn’t make it “mindless”
Explain how it’s a stupid take? This comic is exactly what AI does. You tell it to do something, it sometimes kinda does. You didn’t create anything.
I’m not the guy but I guess it’s that it takes a bit of skill to tell it what kind of pizza you want and to verify it’s actually the right thing that you receive etc. So in this example the order should be pretty elaborate or there should be several panels where the waiter brings the wrong or not quite right pizza.
While telling AI what you need is extremely easily in one regard (it’s natural language) it makes it often pretty difficult to be accurate at the same time. Also by the nature of LLMs the results are hard or impossible to predict.
However, I think it’s a bit like printing something with a 3D printer and then saying “I made this”. The 3D printer actually made it but telling it what you want was the difficult part and at some point involved some 3D modelling or CAD or even g-code programming, tinkering with filament choices, speed and temperature settings, infill, support structures etc etc. While this is quite easy to do nowadays I remember the time where it was a big challenge to even get the damn filament to stick to the build surface. Another similarly is that it’s often not the right tool for a job. E.g. if you want the same object thousands of times or objects that have super fine structures or objects that have to withstand a lot of physical use / abuse or temperature.
However, unlike with a 3D printer you’re pretty much guaranteed to get something when you use LLMs.
Maybe it was a mistake to try and use natural language for this kind of job, it’s too fuzzy and interpretable. Maybe we need a new language, one with a much stricter syntax and no room for interpretation. If you make it simple enough, and use some natural language for important keywords, shouldn’t be too difficult to learn.
We could call it “language for programming” or something like that.
Eh, who am I kidding, that’ll never catch on.
I think there’s some difference in 3d printing. If you find a premade model online with written settings, then print it saying you made it, taking credit for the design, that’s what vibe coding is more like. It would be more apt to say you printed it. If you create a model via 3d modeling, cad, gdnt, etc, then that’s more like programming, and you can say you made it and that be true and honest.
Isn’t programming also just telling the computer what to do? It gets converted to machine language, and you didn’t write that yourself either.
The only difference I see is that programming uses more complex wording so it requires skill and thus can’t be used by everyone.
It is like promting AI in Klingon.
When using other people’s code, there actually is a long-established system of attribution and licencing. You still have to do some (occasionally quite complex) problem solving to figure out how to apply their code and your previous expertise to the task.
It would be more like prompting AI with a fully written text to translate into a different language… except that compiling your code generally produces mostly deterministic, predictable results (which, indeed, is part of programming: to predict how your program will behave) which you can test for errors and debug, while the AI translation is nondeterministic by design (because it’s supposed to sound human and humans don’t speak deterministically either) and if you know enough to check the translation, it would have been much less wasteful to just translate it your damn self instead of burning the planet for a petty convenience.
If I use auto complete, did I write the message?
If I use a hammer, did I build a piece of furniture?
If I use computer software to design an antenna, did I design the antenna?
If I ask AI to kill someone for me, did I kill them or does the AI go to jail?
Those are all false equivalencies. All of those are tools to supplement a job. Using AI to vibe code an app is not a tool, it’s wait service, like the comic illustrates.
If you auto generated an entire message, like prompting AI to write you an email, then no, you didn’t write a message. If you auto complete words that you intended to write, then you used tools to help write a message that you created.
A hammer is a tool to build something. Vibe coding is like buying an IKEA piece of furniture and claiming you built it. You didn’t build anything, you assembled someone else’s build. If you had a magic hammer trained on other people’s work and you told it to build a chair, then no, you would not have built a chair.
Computer software is again a tool to complete a job. If you do the engineering, design, and build the antenna, then that’s like programming, yes you designed an antenna. If you tell the software you want an antenna and it generates plans you didn’t design an antenna, you ordered plans for one.
If you ask AI to kill someone, that’s like taking a hammer and killing someone. Yes, you used a tool to kill someone.
It’s the complete generation of work and the claim that someone did it themselves that is what AI works are at their core. There is a small skill to prompting most effectively, but that’s not creating anything. It’s at most building an outline for something. Skilled developers can review code and make corrections, but they didn’t create anything.
If asking AI to kill someone is using a tool to kill someone, then I would have used a tool to build X - because that’s like using a hammer to build something (your words)
That’s inconsistent logic
It’s more that ordering a crime is still illegal. If you tell a person to go and kill someone else you’re still liable. In the case of using an LLM, considering our governments are trying to keep these companies not liable for anything, even assisting children to kill themselves, you’ll likely be judged as a killer yourself anyway. But if the LLM company had some degree of responsibility, you’d still be held as ultimately responsible because the LLM presumably didn’t come up with the crime, you did.
To reiterate what the other person was saying, if you print your own 3D model you can say you created the thing in addition to printing it. If you just download a model, regardless of license, and do no modifications, you just printed someone else’s creation.
When you get an LLM to generate code for you, you can abbreviate that as “I generated this”, which will be understood as using gen AI, saying you created it is misleading as we understand that as meaning you coded it yourself. Even technical leads shouldn’t go around claiming everyone’s work, their credit extends to their designs, direct contributions, leadership, mentoring, etc, they’re not the sole creators of whatever project they lead.
Working exclusively with generated code would make you closer to a maintainer from a library you didn’t write, you can put in the effort to understand what’s going on and maybe even be confident enough that it’s good for what it is, but that doesn’t make you the creator, the same way reviewing a PR from a colleague doesn’t make you the author of their code.
It’s not about what’s legal, it’s that computers can’t be held responsible for things
How small should the pieces be before you can say you built it? Then it’s ship of Theseus
You’re liable for the murder that you ordered. This is a legal distinction where building something isn’t. Technically if you order AI to kill someone and it goes off and does it, you didn’t kill someone, you prompted it to. This is all theoretical as AI doesn’t have the ability outright kill someone, so this is all under the presumption that AI has a physical means of killing someone unattended.
If this AI were more like a gun where you pointed it at someone and commanded it to shoot, then you killed someone use AI as a weapon.
I will say there is nuance that can be debated between when AI use crosses from a supplemental tool to vibe coding. I don’t have a clear cut answer on that and don’t have the drive to go into it further, but I will concede that.
Right, because murder is equivalent to creative work.
But you have a point. They didn’t actually kill. They ordered a hit. Still culpable, but without the effort or skill to actually do the deed themselves.
A wild take appeared!
As an AI chatbot I think your take is fucking stupid.
I would agree with you, but you took such a shitty way to formulate that, that you probably won’t get anyone at your side