AI cannot write software

Matrices is why I dropped out of Engineering. true story.

My God this thread will not age well...and it will be obvious in about maybe , 9 months.lol.

tigger286 said:

Matrices is why I dropped out of Engineering. true story.

Click to expand...

I dropped comp sci because of algebraic structures (Rings, groups etc)

AI won't fully replace devs, but it could lead to companies thinking they don’t need them, causing layoffs. It might also replace junior devs, which raises the question—where will future senior devs come from? This could harm the dev ecosystem.

For a project, AI can write 50% or more codes, but Developers have to check all of them. It still saves a lot of time.

ProxyBird said:

My God this thread will not age well...and it will be obvious in about maybe , 9 months.lol.

Click to expand...

I don't mean to be offensive with this, and I have been a victim of it many many times myself, but..

In this world there exists really powerful techniques of manipulation. I know because I'm in groups where we study and apply them. Most people, including most marketers are not aware how deep it goes.

Carefully orchestrated manipulation tends to spread like a virus in society.

How you know you've been a victim of it is this :-

If you have a strong opinion of something and you don't know why you've got that strong opinion, then you're probably subconsciously repeating the energy of that wave.

I'm sure you can spot MANY examples of this today, but, one of them is in fact AI hype.

The problem with this one is it's near impossible to see the lies unless you go really deep into the inner workings of LLMs.

But, take a look at this..https://www.pcmag.com/news/is-openais-rumored-20000-ai-agent-good-enough-to-take-your-job

This should give you insight into the reality.

It's like I said in one of my replies, the only way we can attempt to simulate an AGI is with enough fine-tuned models working together, where each model is very good at a specific task, and even then you've only got simulated AGI within a fixed range of tasks.

A real Ph.D level worker you can literally just give them ONE LINE like "Build me the best AI content creator that exists. You have a $5mil budget and 2 years."

That's AGI.

The human needs nothing more. They can go and achieve the task from that.

We will never have an LLM that can act like that, and that's incredibly obvious once you dig into how transformers work.

A transformer is basically..

Your prompt is tokens. Each token is a high dimensional vector in high dimensional space.

While not super mathematically accurate, you can think of this like a big complex shape in higher multi dimensional space.

Picture it like the size of a galaxy. A big complex bendy shape.

This represents your prompt.

The transformer using math to bend and warp this big shape into another shape, and that shape you end up with is your output, which, when bended and warped properly gives you an output that contains a good "answer".

In training it's just learning how to bend these shapes.

That's all it does. That's the simplest intuition of it. A shape bender, based on how it sees shapes relating to each other.

A really simple shape bender might be, circles become triangles, and squares become rectangles. A simple machine can learn to do that. A really complex machine can learn to convert infinitely more complex shapes.

This isn't strictly mathematically valid, as shapes should be continuous, and to "connect" vectors we need some rules, but it's accurate enough.

If you want a more accurate way to think of it they would be like constellations. Ie, distinct points. Stars in the sky in a specific formation.

So the transformer is changing one star formation with billions of stars, to another formation.

Human intelligence isn't like this. We aren't computers, computing high dimensional vectors to convert one thing into another. We have a continuous awareness.

Read up on Helen Keller. She went blind and deaf at 19 months, and at age 6 she learned language.

How is that even possible? Imagine. You're 6 years old. You've never been able to see. You've never been able to hear. All you can do is smell and touch. How do you learn language and communication? Humans are vastly different creatures to transformer models. One is true aware intelligence and the other is just computation of vectors.

OriginWorld said:

I dropped comp sci because of algebraic structures (Rings, groups etc)

Click to expand...

That's abstract algebra

Linear algebra and calculus is what's used in machine learning. They are easier. You can learn the basics of linear algebra quite quickly.

usapvasells said:

AI won't fully replace devs, but it could lead to companies thinking they don’t need them, causing layoffs. It might also replace junior devs, which raises the question—where will future senior devs come from? This could harm the dev ecosystem.

Click to expand...

It won't replace junior devs any more than IDEs replaced junior devs. What it will do is make junior devs more effective if we use LLMs in the right way and not trying to make them "AGI".

Rather than needing a team with a team leader, 4 senior devs, 3 mid level and 3 juniors, you could get away with 2 senior, 2 mid level and 1 junior for example.

They won't replace though, as you still need humans in the loop. LLMs are more like calculators in that they allow humans to do things faster and better.

I believe it's just making it easier for Junior devs to become mids, it will speed up the learning curve. I wish we had this 15 years ago when I wrote my first lines of code, would have saved me from so much struggle and hours spent digging for basic stuff on forums & on Stack Overflow. Now all the junior devs can have an AI tutor next to them 24/7, they just have to learn to use it the right way (instead of vibe coding, they should ask for explanations and actually learn what it's doing and why is it doing the way it's doing. Question AI on each step, and it will probably give you another 2 or 3 better solutions).

katona44 said:

I believe it's just making it easier for Junior devs to become mids, it will speed up the learning curve. I wish we had this 15 years ago when I wrote my first lines of code, would have saved me from so much struggle and hours spent digging for basic stuff on forums & on Stack Overflow. Now all the junior devs can have an AI tutor next to them 24/7, they just have to learn to use it the right way (instead of vibe coding, they should ask for explanations and actually learn what it's doing and why is it doing the way it's doing. Question AI on each step, and it will probably give you another 2 or 3 better solutions).

Click to expand...

LLMs are beyond incredible for learning.

I can learn concepts in a few hours that would have taken months of study to grasp because i can frame my current understanding and get immediate feedback on where i am right and wrong, then i can pose another question like “ok so its like x and if you do y blah blah”.

Or i can ask it to create examples, or create visuals. Its just beyond amazing for learning.

Youre spot on with this. We are on the wrong path. The real power is learning, helping us with problem solving, helping is to visualize and providing other fine tuned specialist tasks to assist us.

splishsplash said:

I just don't see how AI coding is going to replace devs AT ALL.

I'm trying my absolute best, but maybe the problem is I just have too much experience?

I can see all the absolutely wacky shit it's doing and I'm like WTF.

Aider for example. It's an AI dev system, with 80% coded with itself.

It doesn't even work with Azure deepseek. It has a bug. So I decided to use windsurf and claude 3.5 to try to fix the bug to see how it handles it.

Here's one of the things it wanted to do.

It wants to add this in model-settings.yml


- name: azure_ai/deepseek-r1
edit_format: diff
weak_model_name: gpt-4o-mini
use_repo_map: true
lazy: true
reminder: sys
examples_as_sys_msg: true
extra_params:
api_type: "azure"
api_key_env_var: "AZURE_API_KEY"


Pay attention to the last 2 extra_params.

For starters, wtf is with the string "AZURE_API_KEY"? That's just WRONG. That needs to be an env variable set in http://models.py.

Now, let's go to http://models.py and see what it simultaneously wants to change in there.

if http://self.name.startswith("azure_ai/"):
api_base = os.environ.get("AZURE_API_BASE")
api_version = os.environ.get("AZURE_API_VERSION")
api_key = os.environ.get("AZURE_API_KEY")
if not api_base or not api_version or not api_key:
raise ValueError("AZURE_API_BASE, AZURE_API_VERSION, and AZURE_API_KEY must be set for Azure AI models")
kwargs.update({
"api_type": "azure",
"api_base": api_base,
"api_version": api_version,
"api_key": api_key,
})

Ok, so it's getting the env variables, that's right.

But, it's now just adding api_type: "azure" again in the code, after adding it in model-settings.yml

And you know the best bit?

THATS NOT A VALID PARAM!

These are kwargs for litellm.completion. There is no param called api_type.

I can tell everyone that, people using AI to develop software are in for one heck of a treat dealing with bugs and maintainability.

The AI has absolutely no understanding of what's going on. It's hallucinating and missing important things. When it fixes one bug, it creates 10 more.

The only thing AI is good for is analysing codebases, and creating quick one-time small pieces of software.

You might think it saves you time, but what's happening is you're sprinting off when the race is 150 mile ultramarathon. You'll be tired 500m into the race, while the really slow jogger will just keep going and absolutely obliterate you.

I get that people are building stuff with it.. Rotating boxes and all that cool stuff.

But if anyone tries to actually build something that people are going to use, that needs to be maintained, updated and bug-fixed, they're going to get a massive reality check.

Software is NOT write once and forget.

Look, here's the codebase for vscode.

https://github.com/microsoft/vscode


Go visit issues.

"5k+"

And this is enterprise grade software written by humans at Microsoft.

Now imagine a few years down the line with all this AI slop.

I will say this.. There's going to be an absolute mountain of gold for experienced devs to fix shit.

And before anyone says AI is getting better..

It isn't.

It's the same next token prediction with the same problems, the same lost in the middle context window issues, the same lack of understanding of what it's doing. It just gets more sophisticated at pattern matching.

The only way to get it even close to the competence of a human is throwing a ton of compute at it, and passing everything back and forth, running 100's of inferences using an array of different fine-tuned models that do different tasks.

Yes it will improve, but it's still not going to have abilities that humans do.

Click to expand...

They are just stochastic systems and no matter how much you fine-tune them with RL or scale up data and compute, they’re still just advanced pattern matchers, remixing old training data without real understanding or creativity.

AI will not replace dev, but its hand in hand. Im coding using AI atm, and i am like 50%++ more productive than before...

osasinc said:

They are just stochastic systems and no matter how much you fine-tune them with RL or scale up data and compute, they’re still just advanced pattern matchers, remixing old training data without real understanding or creativity.

Click to expand...

I wouldn't go to the other extreme here.

DeepMind created an RL trained model called AlphaTensor which found a more efficient way to multiply 4x4 and 5x5 matrices together than the current way discovered by humans 53 years ago.

LLMs and AIs trained with RL are *incredibly powerful* and *incredibly useful*.

They just aren't going to lead to AGI.

It's pattern matching, sure, but incredibly sophisticated pattern matching with room to generalize.

Most things in the universe are just patterns.

Language is patterns. Math is patterns. Software engineering is patterns.

The hard part is a general model that's like a human. We are pretty special and not easy to replicate, but we can be complimented well with AI.

For example, that matrix optimization.

A couple of Austrian mathematicians took AlphaTensor's optimized version, and improved upon it.

And it took human ingenuity to create AlphaTensor.

The combination of humans + AIs is where the magic is.

3 pages of this guy ranting.

I'm currently using AI to develop simple scripts of ~700 / ~800 lines of code.
It takes some time and reiteration, but it closes the gap between us and a beginner developer.
Definitely better than paying $1,500 a disgusting overpriced developer that will leave me with a low quality product.
It will take probably another 3-4 years to cover most the use-cases needed in my agency, and I'm happy with that.

I should have guessed by the first 4 sentences that this topic was a waste of time, my bad on keep reading.
No idea why it's pinned.

Yes, AI can't write software, but it can code, and very well. It's a good helper, but not a replacement.

Are you familiar with the ARC-challenge?

The answer to the flaw of LLM's is actually a search. A Alpha-Go kind of approach.
I can tell you, AI code will become better and better, it will be full of bugs, but it will work.

The ARC-challenge is amazing, we will have new approaches for code, which probably will not be 100% LLM based.

EDIT: Additionally we will se also a lot of no-code solutions.
So guys if you want become rich, make a no-code platform

lol, as a programmer, your post made me happy . I think its going to be another tool in the hands of programmers. I think it will increase efficiency of the coders. Those who are thinking that non coders can code facebook like apps. Thats not going to happen.
Think about it, we write code once and read 10 times. If you want to add one feature you might have to tweak 10 files buried under 100 files. Thats where it gets tough.

The Scarlet Pimp said:

AI won't replace coders but there are still plenty of hucksters on social media selling information on how you can get rich by prompting AI to write software apps and then selling them.

What they never mention is that the software will need to be upgraded and AI won't be able to fix it easily, if at all. I recently read an article that said coders spend a lot of time removing the garbage code that AI always adds to their existing software when they prompt an agent to fix some code.

One article said 'bots tend to add as much as 40% junk code to existing scripts that needs to be tracked down and removed by the humans. This takes up LOTS of time and increases the workload.

Click to expand...

Thats my experience as well. It takes lot of time to fix the shit ai coded. Its fantastic to have q and a , creating functions , classes etc. If you think you will write one line like "code me scrapebox", it wont work.

This is simply a skill issue with prompting. I have coded an entire desktop application based on python in a day with 5000 lines of code using nothing but gpt o1. I have been working with AI for at least 3-4 hours daily ever since late 2023 though, so i have a lot of experience and know exactly how to prompt AIs.

Also, you simply cannot expect to one-shot applications with AI (at least for the next year or so). You need to bug test, come back to it, fix it, test it again, prompt again.

Read the paper "Situational Awareness" by Leopold Aschenbrenner, he used to work for OpenAI.

People are not ready for what's coming. Mark my words, by Late 2026 the best coder on the planet will no longer be human.

And in another 5 - 10 years we will have Superintelligence.

All of the smartest people on the planet agree with this and are getting ready for it. Ray Kurzweil, biggest AI visionary from google has been warning it for decades.

Garou_jpg said:

3 pages of this guy ranting.

I'm currently using AI to develop simple scripts of ~700 / ~800 lines of code.
It takes some time and reiteration, but it closes the gap between us and a beginner developer.
Definitely better than paying $1,500 a disgusting overpriced developer that will leave me with a low quality product.
It will take probably another 3-4 years to cover most the use-cases needed in my agency, and I'm happy with that.

I should have guessed by the first 4 sentences that this topic was a waste of time, my bad on keep reading.
No idea why it's pinned.

Click to expand...

I've had developers look me in my face and say "It's impossible to do this" and then Grok figured out the answer within a few minutes

We're already 80% towards AGI, and AI is already genuinely smarter than most people that you will interact with on a day to day basis.

The only problem is, we get used to things fast as humans. 7-8 years ago the entire concept of AI itself was Sci-fi, it was foreign. It was only stuff that you would see in the movies. You wouldn't even dream about actually talking to a REAL ai.

A few short years later and people are coding entire applications with AI and they are STILL not convinced that it will completely dismantle 99.9% of industries in the near future.

splishsplash said:

I just don't see how AI coding is going to replace devs AT ALL.

I'm trying my absolute best, but maybe the problem is I just have too much experience?

I can see all the absolutely wacky shit it's doing and I'm like WTF.

Aider for example. It's an AI dev system, with 80% coded with itself.

It doesn't even work with Azure deepseek. It has a bug. So I decided to use windsurf and claude 3.5 to try to fix the bug to see how it handles it.

Here's one of the things it wanted to do.

It wants to add this in model-settings.yml


- name: azure_ai/deepseek-r1
edit_format: diff
weak_model_name: gpt-4o-mini
use_repo_map: true
lazy: true
reminder: sys
examples_as_sys_msg: true
extra_params:
api_type: "azure"
api_key_env_var: "AZURE_API_KEY"


Pay attention to the last 2 extra_params.

For starters, wtf is with the string "AZURE_API_KEY"? That's just WRONG. That needs to be an env variable set in http://models.py.

Now, let's go to http://models.py and see what it simultaneously wants to change in there.

if http://self.name.startswith("azure_ai/"):
api_base = os.environ.get("AZURE_API_BASE")
api_version = os.environ.get("AZURE_API_VERSION")
api_key = os.environ.get("AZURE_API_KEY")
if not api_base or not api_version or not api_key:
raise ValueError("AZURE_API_BASE, AZURE_API_VERSION, and AZURE_API_KEY must be set for Azure AI models")
kwargs.update({
"api_type": "azure",
"api_base": api_base,
"api_version": api_version,
"api_key": api_key,
})

Ok, so it's getting the env variables, that's right.

But, it's now just adding api_type: "azure" again in the code, after adding it in model-settings.yml

And you know the best bit?

THATS NOT A VALID PARAM!

These are kwargs for litellm.completion. There is no param called api_type.

I can tell everyone that, people using AI to develop software are in for one heck of a treat dealing with bugs and maintainability.

The AI has absolutely no understanding of what's going on. It's hallucinating and missing important things. When it fixes one bug, it creates 10 more.

The only thing AI is good for is analysing codebases, and creating quick one-time small pieces of software.

You might think it saves you time, but what's happening is you're sprinting off when the race is 150 mile ultramarathon. You'll be tired 500m into the race, while the really slow jogger will just keep going and absolutely obliterate you.

I get that people are building stuff with it.. Rotating boxes and all that cool stuff.

But if anyone tries to actually build something that people are going to use, that needs to be maintained, updated and bug-fixed, they're going to get a massive reality check.

Software is NOT write once and forget.

Look, here's the codebase for vscode.

https://github.com/microsoft/vscode


Go visit issues.

"5k+"

And this is enterprise grade software written by humans at Microsoft.

Now imagine a few years down the line with all this AI slop.

I will say this.. There's going to be an absolute mountain of gold for experienced devs to fix shit.

And before anyone says AI is getting better..

It isn't.

It's the same next token prediction with the same problems, the same lost in the middle context window issues, the same lack of understanding of what it's doing. It just gets more sophisticated at pattern matching.

The only way to get it even close to the competence of a human is throwing a ton of compute at it, and passing everything back and forth, running 100's of inferences using an array of different fine-tuned models that do different tasks.

Yes it will improve, but it's still not going to have abilities that humans do.

Click to expand...

I get it, AI coding can be a mess, and your experience lets you see the flaws others miss. But that’s your advantage. AI isn’t replacing devs; it’s a tool that needs expert guidance.

The real opportunity? Master AI-assisted development. Build better workflows, fix its mistakes, and lead the way in using AI effectively. The best devs won’t be the ones AI replaces—they’ll be the ones who know how to orchestrate it.

splishsplash said:

I just don't see how AI coding is going to replace devs AT ALL.

Click to expand...

You exactly nailed the point. That’s when disruptive technologies come in.

It is definitely possible to code a full blown app - as someone who has less / zero dev skills.
This fact is already replacing a lot of staff. Later on the senior’s are affected as well.
Just think of the layoffs.

I remember when it started when loan departments said exactly the same, well - unfortunately it came faster than thought.