buttplug.io 🍑🔌I really cannot stress enough how much of an unexpected marketing gift the term "vibe coding" has been for us.
sinister@buttplugio what the heck is vibe coding?
Patrizia@sinister @buttplugio Throwing prompts at generative AI until it does the thing you want. It's explicitly not about understanding what it does or worrying the code quality.
@patrizia @buttplugio at this rate should i even learn how to program? Its been years ive not made much progress lol
@sinister @buttplugio Only you can answer that, but I don't expect programming jobs to go away - a lot of trouble gets caused when nobody understands or cares about the code :)
Some open source projects are be plagued by users submitting AI-generated patches and security reports that are invalid, because the user didn't read, understand, or investigate what the AI spat out.
Alan Miller 
🇺🇦@patrizia @sinister @buttplugio it's been a long time since I considered myself a coder, but my feeling is that AI generated code will be acceptable for the 90% that's not edge cases. It'll have problems with the 90% of code written to deal with the edge cases.
Cassandrich@fencepost More like 10%.
Klaus Frank@fencepost @patrizia @sinister @buttplugio
But the 90% case is done in 10% of the time you spend on the code. IntelliSense and tab completion do already everything you need. AND you don't get into a situation where you're wasting hours to find a bug in a section of the code that you don't even know what it is supposed to be doing.
Frank Heijkamp@fencepost @patrizia @sinister @buttplugio Programming is mostly about proper handling of edge cases. Everything else is just straight through. The straight ends are easy, it is the corners where you spend most of your time figuring out how to make it work.
Ludovic Archivist Lagouardette@fencepost @patrizia @sinister @buttplugio that would mean that a project with 10 AI patches would have 65% chance of having a critical error that could have been caught by understanding the code, and with 30 ai patches that goes to 95%. If a human coder that understood what they were doing had a 3% error rate, 10 patches would mean a 25% chance of a critical error slipping in, and after 30 patches that would be 60%. That would be if the human didn't reflect on their previous patches and use unit tests (which people that know what they are doing do)
vampirdaddy@fencepost @patrizia @sinister @buttplugio
Debugging is 90% of the time you are spending. The first raw program text usually is the fast part, unless you work really hard on optimisation problems - which is another weak point for AI.
AI is trained to only shows the most common text in the context given - neither the most elegant, nor the fastest, most robust nor most correct program.
Myrakel@fencepost @patrizia @sinister @buttplugio
I tried it a few times now but I never(!) got a result that was usable. Maybe it's possible to let the AI write a well-known mathematical algorithm - but everything else seems to be completely non-working nonsense.
It starts with not compilable code, continues with hallucinated libraries and enumerations and ends with agreeing to be unable to solve the given problem in a working way (or just repeating the steps again and again until the user stops).
I tried it a few times now but I never(!) got a result that was usable. Maybe it's possible to let the AI write a well-known mathematical algorithm - but everything else seems to be completely non-working nonsense.
It starts with not compilable code, continues with hallucinated libraries and enumerations and ends with agreeing to be unable to solve the given problem in a working way (or just repeating the steps again and again until the user stops).
Simon Brooke@patrizia @sinister @buttplugio when I was a school child more than fifty years ago, we were advised by our careers advisors not to go into software, because within five years (then) computers would be so clever they would program themselves.
Reader, I did not follow this advice.
Tim 🛥️🌈@simon_brooke @patrizia @sinister @buttplugio I am a similar age - one teacher told me "data entry" would pay more. I also ignored this advice.
Alex, the Hearth Fire@simon_brooke and you were correct not to
Eric Miller@sinister @patrizia @buttplugio Personally, I say yes. Learning how to program in Python is supremely useful, even if you don't take an actual programming job. Its not that hard to write a script to automate some dumb, repetitive computer task that saves you time... and also is kinda fun.
@sinister @patrizia @buttplugio
Well as a hobby it probably becomes quite useful when you can just exploit 1980s vulnerabilities in the software of some "fancy startup"...
Fabian【ファビアン】🏳️🌈@sinister right now, yes. While the LLM is relatively good with code not connected to a bigger codebase, it's bad if there is a codebase. And don't let me start with debugging. I think it's beneficial to be able to code-review the LLM generated thing, also in future.
++ The time will come when training data includes mostly LLM own generated stuff. So lets see how this will end.
pixx@sinister @patrizia @buttplugio
If you want to be good at it and find it interesting, absolutely it's worthwhile. Hell, even pretending for a second that the AI shit was any good which, it's not
"Other people are so good at it, should I even bother trying?"
Uh, yeah, that's how you become one of those people you're looking up at!
Is there a value in cooking when you can just go the restaurant?
Or in sketching when you can just go to a museum?
Or in going outside when you can just look at a picture of it?
Yes. To all of these.
There's a value to skill. There's a value to learning skills. There's a value to finding the things you love and loving the things you find, and no technological "advance" or mass-produced sweatship-product will ever change that.
Aroop Roelofs@sinister Programming isn't just "making it work", part of it is also having to deal with user stupidity and protecting against that.
AI doesn't quite know how stupid users can get.
Patrick Morris Miller@patrizia @sinister @buttplugio Until *you think* it does what you want. Rice's Theorem has some bad news for those would-be techbros.
enoch_exe_inc@patrizia @sinister @buttplugio I can’t believe that’s a real thing, and what’s more, that my COBOL-programmer father thinks this is a viable way to code.
guenther@enoch_exe_inc i mean cobol programmers are probably pretty desensitized with regards to bad code and awful design decisions.
@guenther @enoch_exe_inc @patrizia @sinister @buttplugio Coming from a guy who knows how to program in x86 assembly, COBOL is fuсking incomprehensible to me and I have no idea how my dad writes code in it. I’m fairly certain he just maintains it, though.
Zeeshan Ali Khan 
🇺🇦
🅱🅻🆄🅴🅱️It's another spat between people who use AI and people who don't.
It's dumb. Tools are tools, use what works, don't use what doesn't. Be careful about any dogma.
Short and sweet(after writing this, what a lie), 'AI' is a buzz word for a new kind of search. LLM's have their ups and downs just like traditional search types (exact match, semantic, regex) and Google search. The search can also be reversed to produce instead of search.
Which is causing trouble. So are scams being put into advertisement on Google and Facebook and every business, so were email scams, and so were phone scams. New tech, new attacks. Same old story.
AI is a bit dangerous as a tool because it's not local unlike most tools historically, which means it's nature could change. Though historically we have depended upon external systems like pip for Python, or GitHub, or mdm etc, or even phone infrastructure. Most of those won't all of a sudden change unless you install the update. But you do see rumblings of a trust breakdown with things like GitHub. Where people no longer trust that their private code is actually private. However, 'AI' (which is a terrible name for it) is very powerful and definitely has many use cases. Many of which have not been discovered yet.
Because 'AI' is built on data found everywhere and anywhere, and because in some cases it is polluting data sources (in some cases this is a huge problem), people are not happy, they feel, perhaps correctly that they have been stolen from. Simultaneously, most of us now have a new powerful search tool. It's a difficult moral landscape. Is vibe coding actually a thing that's super common? I don't know. I hear people use the term more often as an attack on AI. So my guess is that in reality, it is more a memetic dis on those who use AI.
Learning code will always be useful. To learn to code is to learn the structure of systems. Systems have some common truths about how they evolve, and attempting to build complex software teaches you about traps that systems run into that cause them to fail as they grow or age. Find a thing you want to do, and try to do it. Language models can help, and you can ask questions about the code itself. You won't always get the answer, but we still have other tools like traditional search, and books, and YouTube videos, etc. Use all of them.
Oh, and other people. People are still the most powerful source of information. Find smart people.
João Santos@BlueBee @sinister @buttplugio a tool is a tool, and a tool that you don't understand is a dangerous thing. Excavators, arc welders, and circular saws are all tools too, should we encourage people to use them without proper training too?
Nils 'Kojote' Hitze
aeva@sinister @buttplugio it's when you code with a plug in.
Robin