Posted on 03/24/2025 9:24:07 PM PDT by SeekAndFind
Way back in 2023, Andrej Karpathy, an eminent AI guru, made waves with a striking claim that “the hottest new programming language is English”. This was because the advent of large language models (LLMs) meant that from now on humans would not have to learn arcane programming languages in order to tell computers what to do. Henceforth, they could speak to machines like the Duke of Devonshire spoke to his gardener, and the machines would do their bidding.
Ever since LLMs emerged, programmers have been early adopters, using them as unpaid assistants (or “co-pilots”) and finding them useful up to a point – but always with the proviso that, like interns, they make mistakes, and you need to have real programming expertise to spot those.
Recently, though, Karpathy stirred the pot by doubling down on his original vision. “There’s a new kind of coding,” he announced, “I call ‘vibe coding’, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It’s possible because the LLMs … are getting too good.
“When I get error messages I just copy [and] paste them in with no comment, usually that fixes it … I’m building a project or web app, but it’s not really coding – I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.”
Kevin Roose, a noted New York Times tech columnist, seems to have been energised by Karpathy’s endorsement of the technology. “I am not a coder,” he burbled. “I can’t write a single line of Python, JavaScript or C++ … And yet, for the past several months, I’ve been coding up a storm.”
(Excerpt) Read more at theguardian.com ...
The code changes for the same request. Still need a human.
I asked GROK to make a simple webpage of top header and bottom footer and left and right column.
I made 2 requests for each line.
I used the same directions below and GROK came up with the same layout but the code was slightly different.
GROK -
1st line: This can best be viewed on a pc and does not collapse into a single column for cell phone. First attempt used vh, px, flex. The second attempt used vh, %, flex.
make a simple webpage with top and bottom div and left and right div
2nd line. This one will create a webpage for pc that can also be viewed on a cell phone.
First attempt used vh, px, flex. The second attempt used vh, %, rem, px, flex
make a simple responsive layout webpage with top and bottom div and left and right div
#6 “When I get error messages I just copy [and] paste them in with no comment, usually that fixes it … I’m building a project or web app, but it’s not really coding – I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.”
Mostly works....
While Microsoft still sending updates for bugs in their windows OS around since 1995 version Windows 95.
Or use an ATM run on their code.
Yeah, try "mostly works" with a payroll system.
I knew the article was going to be trash before I read it - and I wasn’t surprised.
I’ve tried several models and many write poor code - sure, it works, but it doesn’t scale well. The C++ code they churn out is great to illustrate how to solve a problem, but they generally won’t work well (or fast) in complex systems. At this point I would say it works as well as a novice programmer, not quite a junior programmer. Programming by humans isn’t going anywhere anytime soon. AI is a great tool, but it doesn’t have the experience to make larger projects work. I’m not sure you can shove that into a LLM yet.
Putting it in the hands of inexperienced developers is only kicking the can down the road - there’s a guy who built a small SAAS and had to take it down after the code that was written by AI had bugs and security flaws. His story is on X.
Apologies if this sounds like I am insulting you (I promise that is not my intention), but I seems you haven't been in the industry very long.
Your remark about designing software is spot on, that is what you are SUPPOSED TO DO, even with agile, etc you should have some form of requirements and and an initial design to implement.
But for whatever reason there has always been a deep mistrust (or some might even say jealousy) between management and the technical staff, and it seems MBA types are always trying to force developers to use shortcuts and "magic bullets' instead of sound engineering practices.
Methinks this is because these MBA types see everything a cost which needs to be reduced, which translated to English means "headcount" (other than their own positions which are strangely not classified as an expense).
This business culture is only concerned about cranking out code ASAP, to them design, unit tests, etc are just a developer indulging in their passion which wastes time and money.
This is why C-suite types love outsourcing to India, for the most part these people are just carbon based generative LLMs who write basic code by googling for the closest example of what they think is wanted and slightly modifying it to meet what they think is needed.
Programing is, has, and always will be hard to do correctly, since it requires time, expertise, and several iterations to get right
LLMs are just one in a long line of failed attempts to ignore this.
probably except for cobol...
CASE tools promise. Been hearing it since the 1970’s.
I use AI tools almost every day to help me write code. Yes it helps me do things much faster. But it does not take away the need to understand all the details End to End and fix bugs when the AI tool gets it wrong.
Its not eliminating programmers. But it is greatly reducing the numbers needed, especially of those who are not versatile and able to pick up new tools quickly.
And software continues to eliminate many more jobs as automati9n & integration is such a game changer in almost all industries.
I just spent a couple of days working with Grok 3 writing a program to simulate sessions of gambling at craps.
Grok first presented me with code in Python, with which I am not acquainted. I asked for Turbo Pascal and he obliged with a comment suggesting that it's quite outdated. That might be, but at least I was able to work with it.
Grok made surprising errors in logic, making a modification once which looked like a beginner's oversight. It should have been easy to get right. I'm amazed that he got it wrong.
Grok also didn't structure the program as tidily as I would have. This slowed trouble-shooting as the logical flow wasn't as apparent as it could have been. It was like me trouble-shooting a beginner's code.
I will probably do something like this again. I would ask Grok for small pieces of the code and describe the structure that I want. Then Grok and I will prove that the pieces work. Then we will link them together, proving that the whole thing works.
One example of Grok's shortcomings is that he properly evaluated the probability of making a point of 10 in craps. Unfortunately, he failed to properly weight that outcome by the probability of facing the need to make a 10.
When the shortcoming was pointed out, Grok recognized the mistake. Grok uses phrases such as "Great catch!" to acknowledge the user's ability to correct Grok's error.
After quite a few "Great catches" one's confidence in Grok decreases quite a bit. Unfortunately, my understanding of higher level statistics calculations involving variances of very large samples was insufficient to check Grok's work. Our simulations failed to match Grok's calculations; not by a whole lot but by an amount which appeared to be indicative of an error somewhere in our analysis.
I think Grok might be able to help me in the future. I once worked with a software engineer who was much more proficient than I. He could code a program on the fly using a combination of top-down and bottom-up design with minimal errors. It's harder for me to imagine that Grok would be a help to him; at least not at Grok's present level of skill.
I’m a retired engineer.
“Apologies if this sounds like I am insulting you (I promise that is not my intention), but I seems you haven’t been in the industry very long.”
True, I was also an electronics engineer. I worked for twenty years at Boeing, and Ten years at the Keck Observatory in Hawaii.
I designed systems that involved both Hardware and software.
Most all of it Science oriented Prototype “stuff”. Nothing for commercial sales. I was one of those real “scientists” that they keep hidden away.
We cost money, wrote papers, but never made a tangible profit, we just facilitated the company or institution in the pursuit of their goals.
The people I worked for thought I was worth that cost.
How else do you retire on acreage, with an ocean view,
in Hawaii, at age 55?
You are a very good designer.
You are a lucky man, it sounds like you really enjoyed your career, and you were well rewarded for it, so you are way smarter than myself.
Nah. Too much of coding is error handling. You can’t trust users, or networks. AI can give you the basic run through, maybe, if it doesn’t accidentally use an API that doesn’t exist, but that’s only 10%. And AI can’t wrap itself around all the exception.
“You are a lucky man” No I am not, I’m a blessed man.
“you are way smarter than myself.” I would not make that conclusion!
May you be blessed on your trip through life.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.