I need to be very clear, that the push towards "vibe coding" - that is, deliberately deskilling people - is because AI code assistants are an (increasingly expensive) subscription service.

If you know how to code, you can just write Python, C, Java, R, PHP, whatever for free and make things. You may not own the tools of production, but at least you're not renting them.

If you have been deskilled so you only know how to vibe code, you will be paying for that privilege forever.

This also goes, by the way, for researchers who are starting to be convinced they don't need to learn how to be scientists anymore, because "the AI" can just do the science for them. Nope.

@jimbob My sincere hope is that AI is exposed for the wasteful fraud it proves itself to be (over and over) before it ruins large segments of the economy and society.

The most egregious failure of AI is its inability to admit that it doesn't know what it very clearly doesn't know -- and instead, produces confident-sounding bullshit.

The dot-bomb-v2.0 is long overdue.

@JustinDerrick @jimbob I hate that I could drop the name of any of our unserious (industry, national) leaders in place of AI and it would be just as accurate. Our cultural moment really is the final approach of falsehood’s flight and the truth limping after it is lost over the horizon.
@JustinDerrick @jimbob I'm very worried that when the bubble bursts, governments will decide AI is too big to fail and is now essential infrastructure. Public money to be poured into the black hole indefinitely...

@jimbob you are going to LOVE my new idea. Why vibe code when you can pay someone else to do it for you?

https://yolocode.dev

YOLOCode – Autonomous AI-Driven Development

@ianiv @jimbob "Is code quality guaranteed? No."
It's just what I always wanted!

@ianiv @jimbob Just need to build a bridge between it and myapp.ai
https://www.mymap.ai/app-idea-generator

This is moving into Electric Monk territory.

Free AI App Idea Generator

MyMap.AI
@ibk @ianiv @jimbob What the fuck did I just click on?
@mattb @ianiv @jimbob The future.
You clicked on the future.
@ibk @ianiv @jimbob "Tab has crashed" what a perfect solution to any AI problem!
@StumpyTheMutt just the kind of top-notch quality you can expect from the service

@ianiv @jimbob "All artefacts are generated algorithmically and delivered as-is. YOLOCode makes no representations or warranties—express or implied—regarding functionality, security, performance, intellectual-property provenance, or suitability for any particular use. You must review, test, and validate the output before deploying it."

All this for 5k a week

@ianiv @jimbob OK, is this a serious service? Because for me it has a strong vibe of a company providing some health related service named "Snake Oil". It might even exist and bring profit. But I can only imagine the next level is being straight into the face of potential customers, and make and ad: "come to us, we will fraud the hell out of you".
@agturcz it’s something I threw together in 15 minutes for exactly this kind of commentary
@jimbob this feels a lot like the 90s when outsourcing to India made computer science the purview of nerds who loved just geeking out on this stuff. There was no promise of huge pay. I feel this is where it’s headed again - the companies that do the best will be the ones that invest in skilled labor versus those who fall for the race to the bottom
@jimbob actual experimental scientists who collect real-world data don't need ai until the write-up stage. Although big pharma has been trying to push through toxicology modeling for literally decades and might win on that point.
@rskurat "need" is still a strong word. I've published over 130 papers, and have a h-index of 48, and I've never used one in writing a paper.

@jimbob true - 'need' might be the case for the youngins. I can bang out a well-written 1000 words in about 40 minutes, in my own distinctive voice. AI can issue 1000-word blandly written boilerplate in 10 seconds. Our tech overlords think these are the same things.

But don't get me started on my students inability to understand the basic content of a single page of text. In the long run our superior writing may not matter.

@rskurat @jimbob the pharma part failed in the past and will fail again.
@Leendaal and the official response will be "how could we have known?"
@jimbob Can't say that I get this vide coding thing [I'm not a programmer, I was only ever good at microcontroller assembler in my tertiary education]. This reminded me though of Jenson Huang's reported position on coding is that no one should bother trying to learn it anymore and just use AI. Again not a coder but to me that seemed to be spectacularly wrong-sighted. Do we not bother to learn how to write too?

@alanthecampbell I think an analogy with the invention of spell checkers is kind of useful. Spell checkers are good! When they were added to word processors, they were legitimately great! Not perfect, but a good addition to the software you're using.

Now imagine if spell checkers required a subscription to work, to pay for the insane energy and warehouses full of processors to run them. And imagine if, once spell checkers came out, it was declared that we simply didn't need to teach kids to spell anymore, since spell checkers only cost $15 a month.

@alanthecampbell @jimbob Apart from anything else, if humans stop writing stuff, what are we going to train the AIs on?
@robinadams @alanthecampbell @jimbob sadly it seems you do not know about training data "farms". This is a huge business in labeling data right now.
@jimbob Wendell Berry is turning in his grave.
@lykso @jimbob If he were dead, but he's still very much alive...
@neoluddite @jimbob Oh shit. 😲
@neoluddite @jimbob Wendell Berry is turning in his homestead.
@jimbob
This seems to be the mechanism of drugs. The more you use them the more you need them.
@jimbob
"Programmers are also human" latest video is a fantastic satire on the topic https://www.youtube.com/watch?v=_2C2CNmK7dQ
Senior Engineer tries Vibe Coding.

YouTube
@johentsch @jimbob “why is it rewriting it in lisp?” lol. I used Claude code for a thing exploring the distribution of algebraic numbers using 128 bit floats and instructions from a later interaction would make it remove features requested in earlier iterations. Also, it was not good at maintaining its own code. Still, it spit out ok PNG files without me having to learn some PNG libraries. But it couldn’t figure out a way not to run out of memories with algebraic numbers of large heights.

@jimbob I’ve moved the opposite direction recently and believe that to truly succeed in an advanced field where the technology is leveraged you need to fully understand it.

If your field is technology based then that understanding should be nearly to the level of actual from-scratch implementation, not that you’d have the hours or finances to do it but the evaluation skills.

I recognize I am a far outlier in this opinion and most people do not want to learn masters degree+ level statistics.

@dotsie @jimbob an engineer I talked to this evening at a benefit for my son’s school (not a cs engineer - one who works in another field) told me about a class he took in grad school where they built an entire computer from the ground up to run on a 286 based chip. As a class on learning the deep systems of computing.

This was in 2016. He found the class fantastic - they soldered their own circuits etc. and learned a lot.

@Rycaut that sounds amazing, I’d love to take that class or view the materials online. First principles are too often underrated.

@dotsie @Rycaut There's the excellent course Nand2Tetris which leads you through building a computer, operating system and high-level language with a stack-based virtual machine from first principles (Nand gates, flip-flops and - after you build the memory - a memory-mapped keyboard and screen).

https://www.nand2tetris.org

Home | nand2tetris

nand2tetris
@jimbob damn, i hadn't even thought of that. gotta add that to the rotation next time the subject comes up
@jimbob counterpoint: I can write 10+ languages and could learn SwiftUI too. But I wanted a very specific timer app that I got generated while I was doing other programming work. Nobody I actually talk to wants to stop programming. It's either: save time on writing stupid stuff, generate something while programming anyway, or generate something because they can't program in the first place.
Who are those people who are ready to deskill?

@jimbob plenty of people get by with "renting". It is less honorable in some sense but most people don't care enough to tell the difference. However, I will lament with you for a season.

This deskilling will be less like WYSIWYG and more like the transition from platform specific programming to portable compilers. It will be devastating to the abstraction proliferation problem but the world will largely move on fully ignorant of the existential risks

@jimbob

how about using it where we need it: teaching coding in the absolute most effective manner for an individual's particular needs?

who cares if they are forced to ask that embarrassing stupid question after all, if it's just a nice patient AI teacher?

@rexi how can it teach? Where did it receive its pedagogical training? How would a student know if the answers and explanations it gives are correct or not?

You can literally tell one of these things "you're wrong", even when it's not, and it will generate an alternative answer.

@rexi @jimbob yeah it would be nice if that were remotely within the capabilities of current AI.

@jimbob
And that deskilling will happen. Remember back in die 80s and the early 90s when one wanted to phone someone? You entered the phone number on a landline, digit by digit. For that you had memorized the most important phone numbers. I grew up remembering our homes phone number, the one of my grandma and the one of my best friend in elementary school.

Then at the end of the 90s suddenly everyone had mobile phones with all of the numbers in contacts stored on the phone. And I ask you now, how many phone numbers have you memorized today? Yeah, that's how deskilling happens.

And you are absolutely right, that vibe coding will deskill people to use program languages.
@vaurora

@jimbob I am using R since 2009 and also started to use chatgbt in my coding. It's great. I Iess have to look at old scripts how I did things or my "code collection script". Sure a discussion about stats only succeeded because I know stats. But it still helped.
LLMs will be great tools, but not substitute coders. I think it will make coding easier and even people that wouldn't have tried will engage. Doesn't make that the job easier for people that code to be understood?
@jimbob Amen. Same thing about designers.

@jimbob

It's almost like _skills_ have _value_.

@jimbob on the science part, considering ai can't adequately solve sophomore year math and physics problems
(it does too many mistakes), it can't be relied upon
@jimbob I'm not a native speaker, I've had just woken up, I think that I've never read before deskilling and I was thinking "AI is killing people now and then undoing that?" 😂 instead of de-skilling
@jimbob Vibe coders aren’t real coders, and anything they come up with is not worth anyones time.
@jimbob this is a really good framing. But what about the not-too-distant future where our personal hardware is powerful enough to run capable models locally?
@jimbob Hell yes brother! Forcing a single ideology on people and bringing down the overall standard of work and quality of professionals just to provide technological value through AI is such a scam. Typical silicon valley influencer executives: Over-promise and Under-deliver...
@jimbob I keep reading deskilling as desk-killing (as in firing) which it ultimately is doing too.
@jimbob I am wondering where acceptable tool use while coding stops and unacceptable use of AI for coding begins in your view?
@ErikJonker great question! where does it for you?
@tivasyk ..for a non-experienced, very casual coder , AI is great for explaining syntax, commands , debug etcetera, the fact that it hallucinates and sometimes makes mistakes is no problem whatsoever in the very limited coding I am doing 😀, also for learning it's great because AI is interactive, I can ask questions, ask it to elaborate or find another way for a problem etcetera really unparalleled with anything available before
@ErikJonker thank you for developing! as i said in another post here, i don't have a problem with individuals using llm's the way that makes sence to them and brings them joy.

now, this might seem agressive, but please bear with me: while elaborating on your use of llm's, you have not answered my question, or rather _your own question_ to the topic starter: where do _you_ draw the line between the acceptable and unacceptable use of llm for programming?