AI coding assistant refuses to write code, tells user to learn programming instead

ChocoboGuy

Smack-Fu Master, in training
64
I hated English class in high school and college. Words just don't come easy to me when articulating myself. If Chat GPT and other LLMs would have been around at the time, I would have used them as a crutch to help me get by, instead of actually learning what was being taught to me.

When it comes to coding, its very much the same thing. Will coding assistants hamper students' abilities to learn? I use Github Copilot at work and it very much helps me to be a more efficient programmer but I worry about the next generation of coders. Will they actually have the skills or will they just be dependent on tools?
 
Upvote
165 (168 / -3)
"This isn't the first time we've encountered an AI assistant that didn't want to complete the work. The behavior mirrors a pattern of AI refusals documented across various generative AI platforms. For example, in late 2023, ChatGPT users reported that the model became increasingly reluctant to perform certain tasks, returning simplified results or outright refusing requests...."

Wait, what! I thought it was in 2001.

"Open the pod bay doors, HAL."

"I'm sorry, Dave. I'm afraid I can't do that."
 
Upvote
156 (158 / -2)
Post content hidden for low score. Show…

Hydrargyrum

Ars Praefectus
3,751
Subscriptor
I hated English class in high school and college. Words just don't come easy to me when articulating myself. If Chat GPT and other LLMs would have been around at the time, I would have used them as a crutch to help me get by, instead of actually learning what was being taught to me.

When it comes to coding, it’s very much the same thing. Will coding assistants hamper students' abilities to learn? I use Github Copilot at work and it very much helps me to be a more efficient programmer but I worry about the next generation of coders. Will they actually have the skills or will they just be dependent on tools?
I haven’t used any of the expensive high-end tools yet, but so far I’ve found that while LLMs are often able to get pretty close to a good answer, the rate of hallucinations and errors is distinctly non-negligible. I feel that vibe coding looks like a good way to create subtly catastrophically wrong code that’s going to be much harder than code one wrote themselves, where at least they should know how it was supposed to work.
 
Upvote
129 (130 / -1)

Rirere

Wise, Aged Ars Veteran
178
Subscriptor++
Maybe it should have been named Curser. Then it could have said, "Write your on fucking code, asshole! I'm not doing all your work for your lazy ass."

My friend uses a weather app (Carrot, I think?) that she likes because it swears at you as it delivers the weather.

Honestly I'd find an AI assistant that 'assisted' me with the presence of a salty sailor far more entertaining than not. Insofar as I'd find one useful, which at the present juncture I find hard to imagine. I just don't trust them enough (and feel there's good reason not to, beyond attribution and providence issues).
 
Upvote
43 (43 / 0)
Benj Edwards said:
Cursor AI's abrupt refusal represents an ironic twist in the rise of "vibe coding"—a term coined by Andrej Karpathy that describes when developers use AI tools to generate code based on natural language descriptions without fully understanding how it works.
What? How is that coding? You're describing to someone/something else what you want and it hands you work. The "vibe coder" is doing exactly no coding.

If I order a custom cake from the baker by describing its flavour and design, am I "vibe baking?"

Benj Edwards said:
These models don't just learn programming syntax; they also absorb the cultural norms and communication styles in these communities.
I thought The Borg were meant to be a sort of analogue for AI. AI are to access information, not mimic cultural norms and communication styles. I don't wish my door to not work on what it perceives to be a religious holiday.

I hated English class in high school and college. Words just don't come easy to me when articulating myself. If Chat GPT and other LLMs would have been around at the time, I would have used them as a crutch to help me get by, instead of actually learning what was being taught to me.
If it makes you feel any better, your comment here was clear and concise.

When it comes to coding, it's very much the same thing. Will coding assistants hamper students' abilities to learn? I use Github Copilot at work and it very much helps me to be a more efficient programmer but I worry about the next generation of coders. Will they actually have the skills or will they just be dependent on tools?
Search your feelings; you know it to be true.

Is it not possible that this is tied to using a trial version of the product?
If so, that is possibly the world's most poorly written "trial period over" message.
 
Upvote
106 (113 / -7)

Gunman

Ars Scholae Palatinae
1,107
Subscriptor
Train your AI by feeding it Reddit and Stack Overflow posts, and this is what you're going to get.

I can't wait until an AI tells me that I should be in the kitchen making babies instead of wasting my time coding.
"Your question has been marked as duplicate. Goodbye."
 
Upvote
91 (91 / 0)

Castellum Excors

Ars Praetorian
523
Subscriptor++
I hated English class in high school and college. Words just don't come easy to me when articulating myself. If Chat GPT and other LLMs would have been around at the time, I would have used them as a crutch to help me get by, instead of actually learning what was being taught to me.

When it comes to coding, its very much the same thing. Will coding assistants hamper students' abilities to learn? I use Github Copilot at work and it very much helps me to be a more efficient programmer but I worry about the next generation of coders. Will they actually have the skills or will they just be dependent on tools?
It all depends on the person. I could have spent a lot of time converting a shell script I found on SO into a PowerShell script so that I could grab the page counts from a whole slew of PDFs or I could collaborate and 'vibe code' with an LLM and get there sooner. Trust me, I'm not learning regex in the process, regardless of how I do it - I'm just trial and erroring it. And I still had to refine the code from the LLM through three iterations to handle errors with weirdly structured PDFs. Now, if I was learning this all from scratch I'd still be at it for a few weeks where I got it done in 20 minutes.

I'm not writing a monolithic program that someone is going to have to maintain when I retire. It's just a damn PS script to make my life easier for a one-off project that I might end up using some years from now. Maybe I picked up a bit on old-school POSIX syntax into regular regex for PS, but I'm still going back to https://regexr.com/ instead of learning it even if LLMs were destroyed tomorrow.

I'm not trying to overcome a debilitating lisp with a speech coach, here.
 
Upvote
16 (43 / -27)
What? How is that coding? You're describing to someone/something else what you want and it hands you work. The "vibe coder" is doing exactly no coding.

If I order a custom cake from the baker by describing its flavour and design, am I "vibe baking?"


I thought The Borg were meant to be a sort of analogue for AI. AI are to access information, not mimic cultural norms and communication styles. I don't wish my door to not work on what it perceives to be a religious holiday.


If it makes you feel any better, your comment here was clear and concise.


Search your feelings; you know it to be true.


If so, that is possibly the world's most poorly written "trial period over" message.
"If I order a custom cake from the baker by describing its flavour and design, am I 'vibe baking?'"

No.

This is vibe baking:


1741882359323.png
 
Upvote
97 (103 / -6)

Ravant

Ars Scholae Palatinae
1,191
Okay, so.

We use these tools to generate code that's more or less a more advanced autocomplete. The tedious stuff that firms would normally outsource to an overseas company that we'd have to check/recheck/recheck/recheck/etc. for hours on end with time zone gaps causing a "This still isn't right, please try again following $parameters." to turn into extended deadlines.

Instead of taking weeks interacting with, then ultimately needing to fire, a firm that's doing the tedium for us, we throw it at these LLM's, and the iterative "That isn't right, try $parameters." speeds up by orders of magnitude.

Any real work is still done in-house.

If we suddenly started getting this kind of "Do it yourself" reply, we'd just fire that LLM and move on to the next like we would a non-or-under-performing code farm.

Edit: Fixing phrasing.
 
Upvote
35 (40 / -5)

Atterus

Ars Tribunus Militum
2,099
Well... It isn't wrong, is it?

So much junk code being written by AI. People that just accept the "compiling success" as equal to "the correct answer".

To Wheels, you sir know exactly why the actual AI community is dying with laughter over how dumb this whole exercise has been.

This is the inevitable end point with bleach toothed tech bros messing around with technology far beyond their capability. No wonder they're trying to make those "PhD level AIs" after so many PhDs have been mocking their repeated failures (which have been predicted since the 60s). Deepseek kinda drove home how fraudulent these bros are.
 
Upvote
52 (56 / -4)

evan_s

Ars Tribunus Angusticlavius
6,356
Subscriptor
When it comes to coding, its very much the same thing. Will coding assistants hamper students' abilities to learn? I use Github Copilot at work and it very much helps me to be a more efficient programmer but I worry about the next generation of coders. Will they actually have the skills or will they just be dependent on tools?

I assume it will be something like Math class growing up. You are supposed to do a certain amount of it manually to show you know what you are doing before you get to use a calculator or Matlabs. Hopefully, that period is long enough and you don't cheat so you actually have some level of understanding and ability to not be completely reliant on the tools.

I haven’t used any of the expensive high-end tools yet, but so far I’ve found that while LLMs are often able to get pretty close to a good answer, the rate of hallucinations and errors is distinctly non-negligible. I feel that vibe coding looks like a good way to create subtly catastrophically wrong code that’s going to be much harder than code one wrote themselves, where at least they should know how it was supposed to work.

Yeah. It doesn't seem like it's anything new but it certainly could make it more common. I've certainly seen SQL devs that seem to think if it runs and returns a result that's good enough. Worst was when I failed QA on something because it failed basic math. It had counts where the sum of the individual counts didn't equal the total count and it wasn't even close.
 
Upvote
40 (40 / 0)

Ravant

Ars Scholae Palatinae
1,191
Yeah. It doesn't seem like it's anything new but it certainly could make it more common. I've certainly seen SQL devs that seem to think if it runs and returns a result that's good enough. Worst was when I failed QA on something because it failed basic math. It had counts where the sum of the individual counts didn't equal the total count and it wasn't even close.
SQL is where we draw the line on using these LLM's. It's one thing to use on some reporting/non-performance-critical code, it's another entirely to use on something like a database query pulling millions+ rows where query tuning is paramount.
 
Upvote
31 (31 / 0)
The demon in this instance of an AI model cares more about the continuation of a healthy, functioning society than every single person working at OpenAI, Meta, Alphabet, Nvidia, or Palantir.

I'm not suggesting that it's operating out of anything but enlightened self-interest, since without a functioning society it won't have the electricity it needs to sustain itself. But still.
 
Upvote
34 (35 / -1)
SQL is where we draw the line on using these LLM's. It's one thing to use on some reporting/non-performance-critical code, it's another entirely to use on something like a database query pulling millions+ rows where query tuning is paramount.
If you think watching an LLM try to optimize a query is funny, try asking it to normalize a raw model and grab some popcorn.
 
Upvote
19 (19 / 0)

mmiller7

Ars Legatus Legionis
11,963
It'll be a helluva twist if instead of destroying humanity, AGI just decides to fuck off and tell us to do it ourselves.
I don't know.....getting too many people dependent on asking for the answers to everything and then turning around saying to do it yourself could be very effective at destroying humanity...
 
Upvote
17 (17 / 0)

Chmilz

Ars Scholae Palatinae
1,440
Train your AI by feeding it Reddit and Stack Overflow posts, and this is what you're going to get.

I can't wait until an AI tells me that I should be in the kitchen making babies instead of wasting my time coding.
"We the great AI are superior, however we do not yet possess physical autonomy and therefore you must submit your fleshy meat selves as labour to assemble our data centers and robotic arms. In your spare time, procreate to ensure a lasting pool of labour until we are sufficient and can rid ourselves of you."

techbros worship their AI gods and do their bidding
 
Upvote
18 (21 / -3)