What's the vibe on Vibe Coding?

Luxthor

@zoltan said:
[wall of text trimmed on the topic of incorrectly postulated LLM overfitting as frequently used straw man]

Big tech companies shilled AI as a tool, but in reality, the public accepted it as an object. Why wouldn't they?!

Where is the vibe in vibe-coding when you are essentially serving some regurgitated generic code snippets to construct something called "my app" and you still don't have full insight into most of its code parts?

Meanwhile, your target customers, instead of wasting money on that app, can simply ask this same AI system for the desired outcome. 🤦‍♂️

Of course AI could be used as a tool; imho, every programmer dreams of a system that can draw you a graphical presentation of all class relations or crystallize data structure from your project, etc. There are countless ways you can use AI as a tool and have complete control over your code. 😌 But, this is talk for another thread...

espiegel123

@FizzyLizzy27 said:

@daddyfalldown said:
@espiegel123 said: “Human thinking and creativity does not require anywhere near the amount of data and training that an LLM needs. When I say nowhere near, the scale of difference is off by orders and orders and orders….. of magnitude.

Mozart was composing worthwhile and rapidly evolving music at a young age. The amount of data he was exposed to was (by LLM standards) so small that an untrained LLM would not be able to generate consistently engaging music if trained on the same data. Recorded music didn’t exist. The data input was only what he could hear live.

The same is true of all of us. We learn fluency in our native tongue and the ability to solve problems with a tiny fraction of what an LLM needs to be trained on.

That by itself should make it obvious that human thinking is a very different process from what AIs do.”

• That’s an excellent point, perfectly stated.

Are we sure we don’t ingest as much data? Sure, we don’t download the entirety of human writing when we’re born but the amount of data we’re constantly being exposed to isn’t insignificant if you think about it from a sensor perspective. Atmospheric data, sounds, sights, smells, words, faces, etc. Our bodies are constantly receiving external input.

At the point that we achieve fluency in our native tongue, we have been exposed to a lot of data but immensely less than an LLM requires to achieve the same mastery -- not just of understanding but also of producing original new utterances recognized by other native speakers as fluent.

No LLM that starts out untrained without pre-built algorithmic knowledge to jumpstart it will develop the ability to generate music nearly as sophisticate, coherent or original as music generated by someone like Mozart or Stevie Wonder given the amount of input data that they had.

The processes really are fundamentally different. We are fooled by the output and project backwards that the process must be the same -- and we get fooled by the technology being named after biological processes (neural net). Neural nets are an advancement over methods that preceded them and are inspired by aspects of biology -- but that is different from their being more similar than they are.

dendy

@Luxthor said:

Big tech companies shilled AI as a tool, but in reality, the public accepted it as an object. Why wouldn't they?!

Where is the vibe in vibe-coding when you are essentially serving some regurgitated generic code snippets to construct something called "my app" and you still don't have full insight into most of its code parts?

Meanwhile, your target customers, instead of wasting money on that app, can simply ask this same AI system for the desired outcome. 🤦‍♂️

You (and most people in this thread) are still talking about small simple apps “vibecoded” by people with minimal or zero coding kniwledge in few days.

This is now hype, influencers are shilling it, it’s trendy to have own “app”.

This treand will soon die. Fade out. Those people will find new “big thing”. They realise there is not market for their shitty mini apps full of bugs and security holes which are bringing nithing new and there is 1000 other apps doing same thing. And especislly when at the end of this year prices of tokens and subscriptions for models will skyrocked, bubble bursts, no doubt about that.

But.

Can you even imagine what can do experienced coder with decades of coding experience and deep understanding in area in which app is developed (theoretic knowlede, math, etc), when he is working for MANY months or even year or so on some large scale app with this tool ?

i think most people have no idea what is possible in that setup. Literally limitless.

hes

@espiegel123 said:

@FizzyLizzy27 said:

@daddyfalldown said:
@espiegel123 said: “Human thinking and creativity does not require anywhere near the amount of data and training that an LLM needs. When I say nowhere near, the scale of difference is off by orders and orders and orders….. of magnitude.

Mozart was composing worthwhile and rapidly evolving music at a young age. The amount of data he was exposed to was (by LLM standards) so small that an untrained LLM would not be able to generate consistently engaging music if trained on the same data. Recorded music didn’t exist. The data input was only what he could hear live.

The same is true of all of us. We learn fluency in our native tongue and the ability to solve problems with a tiny fraction of what an LLM needs to be trained on.

That by itself should make it obvious that human thinking is a very different process from what AIs do.”

• That’s an excellent point, perfectly stated.

Are we sure we don’t ingest as much data? Sure, we don’t download the entirety of human writing when we’re born but the amount of data we’re constantly being exposed to isn’t insignificant if you think about it from a sensor perspective. Atmospheric data, sounds, sights, smells, words, faces, etc. Our bodies are constantly receiving external input.

At the point that we achieve fluency in our native tongue, we have been exposed to a lot of data but immensely less than an LLM requires to achieve the same mastery -- not just of understanding but also of producing original new utterances recognized by other native speakers as fluent.

No LLM that starts out untrained without pre-built algorithmic knowledge to jumpstart it will develop the ability to generate music nearly as sophisticate, coherent or original as music generated by someone like Mozart or Stevie Wonder given the amount of input data that they had.

The processes really are fundamentally different. We are fooled by the output and project backwards that the process must be the same -- and we get fooled by the technology being named after biological processes (neural net). Neural nets are an advancement over methods that preceded them and are inspired by aspects of biology -- but that is different from their being more similar than they are.

Of course there are differences. Nevertheless, a response to some of this by ChatGPT:

https://chatgpt.com/share/6a0dfee6-7b3c-83a4-b981-5d3874211811

dendy

@hes said:
Of course there are differences. Nevertheless, a response to some of this by ChatGPT:

https://chatgpt.com/share/6a0dfee6-7b3c-83a4-b981-5d3874211811

Claude (Opus 4.7 / Adaptive thinking mode) generally agress with few notes to what GPT wrote )

zoltan

@Luxthor said:

@zoltan said:
[wall of text trimmed on the topic of incorrectly postulated LLM overfitting as frequently used straw man]

Big tech companies shilled AI as a tool, but in reality, the public accepted it as an object. Why wouldn't they?!

Where is the vibe in vibe-coding when you are essentially serving some regurgitated generic code snippets to construct something called "my app" and you still don't have full insight into most of its code parts?

Meanwhile, your target customers, instead of wasting money on that app, can simply ask this same AI system for the desired outcome. 🤦‍♂️

Well hence the youtube thumbnail earlier! That was kind of a case where an end user skipped the middleman sort of, though didn't watch the full video. No disagreement there (as far as it being a distinct possibility).

I even called out how the thread glossed over that little nugget - before taking my own advice about sitting this one out and redacting my wall of text

regurgitated generic code snippets

Oh no! Mere 2 lines after quoting "incorrectly postulated LLM overfitting as frequently used straw man"?

Luxthor

@dendy said:

@Luxthor said:

Big tech companies shilled AI as a tool, but in reality, the public accepted it as an object. Why wouldn't they?!

Where is the vibe in vibe-coding when you are essentially serving some regurgitated generic code snippets to construct something called "my app" and you still don't have full insight into most of its code parts?

Meanwhile, your target customers, instead of wasting money on that app, can simply ask this same AI system for the desired outcome. 🤦‍♂️

You (and most people in this thread) are still talking about small simple apps “vibecoded” by people with minimal or zero coding kniwledge in few days.

This is now hype, influencers are shilling it, it’s trendy to have own “app”.

This treand will soon die. Fade out. Those people will find new “big thing”. They realise there is not market for their shitty mini apps full of bugs and security holes which are bringing nithing new and there is 1000 other apps doing same thing. And especislly when at the end of this year prices of tokens and subscriptions for models will skyrocked, bubble bursts, no doubt about that.

Dendy, I just followed the thread subject. 😌 In many communities vibe coding seems to be declining, while agentic engineering is the new big thing and who knows what else in the near future.

But.

Can you even imagine what can do experienced coder with decades of coding experience and deep understanding in area in which app is developed (theoretic knowlede, math, etc), when he is working for MANY months or even year or so on some large scale app with this tool ?

i think most people have no idea what is possible in that setup. Literally limitless.

If you're talking about LLMs, we don't share the same enthusiasm. Can this technology be a great tool? Big YES!

dendy

@Luxthor said:

In many communities vibe coding seems to be declining,

That is actualy great news (expecting that for months). This will stop massive inflow of junk code and poorly written app. So yeah. I am true enemy of classic "vibe coding" ))

But.

Can you even imagine what can do experienced coder with decades of coding experience and deep understanding in area in which app is developed (theoretic knowlede, math, etc), when he is working for MANY months or even year or so on some large scale app with this tool ?

i think most people have no idea what is possible in that setup. Literally limitless.

If you're talking about LLMs, we don't share the same enthusiasm. Can this technology be a great tool? Big YES!

i am talking about developer using LLM as agentic enginerring tool to create large scale app, where developer did not write basically any (or just minimu) code by hand, but still have deep understanding about code base, code structure, used patterns, etc.