This, I think, is the real promise of vibe coding tools—that you can learn how to code without a CS degree.
I learned to code without a CS degree. I used a for Dummies book, W3schools, Stack Overflow and the good guidance of a senior developer. Learning to code was never the issue. And I think poking around in the code, experimenting, stumbling on unrelated but helpful answers, before finding your problem, are all great ways to become experienced that are prevented by the use of a tool like Bolt. If Bolt produces code that confuses experienced developers, how is the vibe coder supposed to learn anything useful from it?
Badabinski@kbin.earth
on 07 Aug 20:43
nextcollapse
I learned to program by shitting out God awful shell scripts that got gently thrashed by senior devs. The only way I've ever learned anything is by having a real-world problem that I can solve. You absolutely do NOT need a CS degree to learn software dev or even some of compsci itself, and I agree that tools like Bolt are going to make shit harder. It's one thing to copy stack overflow code because you have people arguing about it in the comments. You get to hear the pros and cons and it can eventually make sense. It's something entirely different when an LLM shits out code that it can't even accurately describe later.
peoplebeproblems@midwest.social
on 07 Aug 22:31
nextcollapse
Or that it can produce repeatedly. That’s something that bothers me. Slight changes in the prompt and you get a wildly different result. Or, worse, you get the same bad output every time you prompt it.
You can use that to your advantage! Slight prompt changes can give you different ideas on how to proceed, give you some items to evaluate. But that’s all they’re good for, and while they can be solid on getting you past a block, I’m horrified to think anyone in the IT space thinks an LLM can output safe, working code.
LLMs are great for getting around specific stumbling blocks, might even present a path you hadn’t thought of or knew about. And that is it. Stop right there and you’ll be fine.
I completely understand how an ignorant bystander would believe AIs pump out working code. I cannot understand anyone with any experience thinking that.
vacuumflower@lemmy.sdf.org
on 14 Aug 12:14
collapse
The only way I’ve ever learned anything is by having a real-world problem that I can solve.
Same thing here except I’m still not a developer. Just from time to time can do something if it’s less boring than going another way.
I’ve even played through the “Turing Complete” game once, because I can’t force myself to repeat it. And it was very interesting, absolutely cool, except that gun has fired. It appears the game changed enough though, maybe it’s a sufficiently different gun to fire again. It’s a game for entertainment, not even talking about real life.
And when there’s a direct incentive, nothing is hard, for real. The hardship is in eyes tiring, time passing, time to render (in case of POV-Ray), migraines. But the task itself just takes it all as a payment, not as an effort.
And I sincerely don’t get why my diagnoses are ASD and BAD despite describing this many times, that is, that happens with ASD too, but honestly ADHD seems the most intuitive abbreviation here.
chicken@lemmy.dbzer0.com
on 08 Aug 04:06
nextcollapse
To be fair, in the quoted passage the author is explicitly not referring to Bolt, but asking for explanations of specific, probably small scope issues from non-agentic AI tools.
Personally even while in school for CS I spent a large number of hours staring at a screen being totally unsure how to proceed to figure out what I didn’t understand, most of it trivial details, or making random edits and hoping it would fix something. I’m sure there are advantages to learning by stumbling around, but I really don’t think it’s the ideal way unless you’re already a very methodically curious person.
The temptation to jump directly to asking the AI to just do everything for you without yourself understanding it is definitely going to be a stumbling block for people learning, and I’m not sure if there’s a good way around that one, but otoh something available 24/7 that can mostly accurately answer beginner questions in context and as you have worded them seems like it would be crazy helpful, so many times I just wasn’t able to progress until I could get some attention from someone.
I’m in this picture! Started taking Harvard’s free CS50 class and quickly bailed, didn’t want to learn actual computer science.
Swedneck@discuss.tchncs.de
on 08 Aug 21:27
collapse
just python and looking up every single thing i want the code to do
“python print text”
“python get input”
“python loops”
“python iterate through list”
and this is why i’m such a diehard python shill, it doesn’t require you to write 5 billion lines of boilerplate code to print some fucking text, it’s very approachable.
threaded - newest
This is the endgame, isn’t it? Just a matter of time.
No, nothing would change: those who can’t code will continue to not coding or write some useless gibberish.
AI: for expert novices.
I learned to code without a CS degree. I used a for Dummies book, W3schools, Stack Overflow and the good guidance of a senior developer. Learning to code was never the issue. And I think poking around in the code, experimenting, stumbling on unrelated but helpful answers, before finding your problem, are all great ways to become experienced that are prevented by the use of a tool like Bolt. If Bolt produces code that confuses experienced developers, how is the vibe coder supposed to learn anything useful from it?
I learned to program by shitting out God awful shell scripts that got gently thrashed by senior devs. The only way I've ever learned anything is by having a real-world problem that I can solve. You absolutely do NOT need a CS degree to learn software dev or even some of compsci itself, and I agree that tools like Bolt are going to make shit harder. It's one thing to copy stack overflow code because you have people arguing about it in the comments. You get to hear the pros and cons and it can eventually make sense. It's something entirely different when an LLM shits out code that it can't even accurately describe later.
Or that it can produce repeatedly. That’s something that bothers me. Slight changes in the prompt and you get a wildly different result. Or, worse, you get the same bad output every time you prompt it.
And then there are the security flaws
You can use that to your advantage! Slight prompt changes can give you different ideas on how to proceed, give you some items to evaluate. But that’s all they’re good for, and while they can be solid on getting you past a block, I’m horrified to think anyone in the IT space thinks an LLM can output safe, working code.
LLMs are great for getting around specific stumbling blocks, might even present a path you hadn’t thought of or knew about. And that is it. Stop right there and you’ll be fine.
I completely understand how an ignorant bystander would believe AIs pump out working code. I cannot understand anyone with any experience thinking that.
Same thing here except I’m still not a developer. Just from time to time can do something if it’s less boring than going another way.
I’ve even played through the “Turing Complete” game once, because I can’t force myself to repeat it. And it was very interesting, absolutely cool, except that gun has fired. It appears the game changed enough though, maybe it’s a sufficiently different gun to fire again. It’s a game for entertainment, not even talking about real life.
And when there’s a direct incentive, nothing is hard, for real. The hardship is in eyes tiring, time passing, time to render (in case of POV-Ray), migraines. But the task itself just takes it all as a payment, not as an effort.
And I sincerely don’t get why my diagnoses are ASD and BAD despite describing this many times, that is, that happens with ASD too, but honestly ADHD seems the most intuitive abbreviation here.
To be fair, in the quoted passage the author is explicitly not referring to Bolt, but asking for explanations of specific, probably small scope issues from non-agentic AI tools.
Personally even while in school for CS I spent a large number of hours staring at a screen being totally unsure how to proceed to figure out what I didn’t understand, most of it trivial details, or making random edits and hoping it would fix something. I’m sure there are advantages to learning by stumbling around, but I really don’t think it’s the ideal way unless you’re already a very methodically curious person.
The temptation to jump directly to asking the AI to just do everything for you without yourself understanding it is definitely going to be a stumbling block for people learning, and I’m not sure if there’s a good way around that one, but otoh something available 24/7 that can mostly accurately answer beginner questions in context and as you have worded them seems like it would be crazy helpful, so many times I just wasn’t able to progress until I could get some attention from someone.
Indeed. A huge proportion of Gen X professionals were self taught children who learnt on home computers and then grew up.
CS degrees don’t teach you how to code. They teach you computer science.
I’m in this picture! Started taking Harvard’s free CS50 class and quickly bailed, didn’t want to learn actual computer science.
just python and looking up every single thing i want the code to do “python print text”
“python get input”
“python loops”
“python iterate through list”
and this is why i’m such a diehard python shill, it doesn’t require you to write 5 billion lines of boilerplate code to print some fucking text, it’s very approachable.
am I missing it or did they not actually post the link to their github?
edit: the link to github has been added. github.com/bunnywapen/dont-go-in-there
The ai didn’t do it for her 🫠