"41% of all code on GitHub right now is AI generated."

[rproffitt]

This just in from an AI researcher.

Given the impact GenAI has had, it's been odd to see how unwelcome discussions about GenAI are here and other places.
Then again, it is an existential threat to new and old programmers that don't add these new tools to their repertoire or skill set.

Also, AI/ML has created no end of angst in Hollywood. Last month a company that makes audiobooks reduced its workforce because the AI tool to make audiobooks costs 20USD a month. Everyday voice actors at this point are going to have a rough time going forward except for the big names.

I have stories about the use of AI/ML/GenAI in the circles I run. All use has resulted in higher productivity. Which is good for the company, not so much everyone else.

[AndreRet]

I am all for AI, we use it quite extensively in our daily operations. You hit the nail on the head, either add it to your repertoire or get left behind IMHO.

[Dani]

"41% of all code on GitHub right now is AI generated."

I don’t see the positive in that.

[Reverend Jim]

I don’t see the positive in that.

I suppose that depends on whether it is good code, or bad code.

[Dani]

I have never seen ChatGPT be able to produce fully functioning code that does something useful and productive (eg a standalone tool or project, such as a GitHub repo) without any human editing in order to work.

[AndreRet]

I agree with Dani, the trick lies in the prompts you send it. Never take it's returned code as gospel, but it does give you a very solid foundation to work from.

I have noticed how many people take direct generated code and paste it hoping it will work. You need to still understand the language you are working with to see the pitfalls it creates and then correct that. We ran a lot of sample code through it and asked for better or improved code which it did return almost 90% of the time, especially functions we never thought of.

I suppose in time it will become more accurate based on the tons of articles around it, let's wait and see.

[rproffitt]

I'll supply two recurring uses that have resulted in lower cost and lower human capital. That will upset a few out there.

1. Company would call on a SQL consultant at times. GPT has replaced that need for the consultant.
2. Another area is to contract for code that used regular expressions. Again, GPT has filled that need.

As GPT progressed this year more and more people I know use GPT based products to be more productive. Adoption of AI/ML/GPT has been swift and impactful.

This has made a lot of people very angry and been widely regarded as a bad move.

• Douglas Adams

[Reverend Jim]

Another area is to contract for code that used regular expressions

Ooooh. I had not thought of that one. That would be awesome. At least it would be if more people knew how to properly express their wishes in plain English.

exp = rx.compile("It's like, you know a letter or whatever followed by a dot and some stuff, right?")

[pritaeas]

Fuel to the fire ;)

https://www.theregister.com/2023/08/07/chatgpt_stack_overflow_ai/

[Dani]

https://www.theregister.com/2023/08/07/chatgpt_stack_overflow_ai/

That's exactly what I've been saying from the beginning.

[rproffitt]

As to good/bad code, I doubt humans do much better at their first pass.

Our success rate so far is 100%. We measure success in that the staff is able to get the job done with the GPT assist and not having to bother me or other consultants.

However there are those that want to write "it writes bad code" for reasons I can't guess. Maybe it's along the lines of self preservation?

[Dani]

As to good/bad code, I doubt humans do much better at their first pass.

But that's the point of the article, right? I might post some iffy code and and preface it with, "This hasn't been tested yet so use at your own risk". ChatGPT will spit out iffy code and preface it with 5 paragraphs of intellectual-sounding technical jargon explaining why it's correct.

[rproffitt]

We've reduced costs, increased productivity and won't be going back to pre-GPT days.

As to use at your own risk, that's true of Human Generated Code (HGC?)

[Dani]

As to use at your own risk, that's true of Human Generated Code (HGC?)

That's my point?

[rproffitt]

So why do folk write such articles? Why not cover use cases such as folk and companies I know?

My answer as above by Douglas Adams: "This has made a lot of people very angry and been widely regarded as a bad move."

[Dani]

So why do folk write such articles? Why not cover use cases such as folk and companies I know?

Because no one disagrees that ChatGPT and AI has a ton of benefits towards improving efficiency, streamlining workflows, eliminating repetitive tasks, etc. There isn’t really a need for a bunch of articles to advocate for it to bring attention to it. It’s already proved itself as the fastest growing platform of all time, and everyone has already heard of it by now, with the vast majority of people trying it, companies embracing it and integrating it into their workflows, etc.

The problem is that a lot of people don’t understand or are not aware of the danger of the technology when used incorrectly, and the damage it’s doing within publications, within forums, within sites like DaniWeb and Stack Overflow, within GitHub repositories, etc. There is more need to inform the world about the dark side of AI. This is journalism, after all. News articles about how everything is roses and rainbows 99% of the time isn’t news. They report on the 1% when it’s not. Especially when that 1% has the ability to spew false information that could cause a lot of damage.

[Reverend Jim]

That's a problem when the news shows only allocate two minutes to cover the topic. You need much more than that to cover it fairly. And, honestly, most people would just tune it out (the same people who read only the headlines). The only person I trust to give it a fair hearing is John Oliver. If you ignore the crude asides, he does cover his main stories in great detail.

[Dani]

Well that’s with news about anything, right? Stories about all the great and innovative ways that a relatively new technology is being used, especially a technology that is already garnering as much praise and attention as ChatGPT, to where everyone already knows what it is and what it’s capable of, are really nothing more than just fluff pieces. What is newsworthy is when this technology being touted the world over as the best thing since sliced bread has a dark side that not many fully understand, that is capable of causing damage by spreading false information on a mass scale.

[Dani]

https://www.google.com/search?q=news+is+always+negative&ie=UTF-8&oe=UTF-8&hl=en-us&client=safari

[Reverend Jim]

I don't entirely agree with that. I think the reason bad news prevails is

1. It produces a stronger emotional response
2. It requires almost zero thought
3. There is just so much more of it

I can believe the negativity bias in the case where it confirms (or seems to confirm) something you already believe.

[Dani]

You don't agree with my post about positive articles about ChatGPT being fluff content, and online publishers not wanting to pour money into fluff content because it doesn't earn the views/$ (and that's why there aren't many of them), or you don't agree with anything in the Google results I linked to?

[Dani]

Given the impact GenAI has had, it's been odd to see how unwelcome discussions about GenAI are here and other places.

Again, discussions (in your own words) about GenAI are not, and never were, unwelcome here. Copying and pasting into posts what AI spits out is what's unwelcome.

[rproffitt]

And there it is again. If I were to share how to get GPT to give better results, the copy paste of AI/GPT output is unavoidable. As such, sharing what we learned is not possible here.

Weird choice but hey, your choice which I will abide by. Example: An AI whisperer is a specialist who excels in comprehending, communicating with and guiding AI systems.

How to share how to improve GPT performance is not possible without copy/paste from the AI/GPT session.

[Dani]

And there it is again. If I were to share how to get GPT to give better results, the copy paste of AI/GPT output is unavoidable. As such, sharing what we learned is not possible here.

It seems we are going around in circles. Per previous posts, you can copy/paste AI/GPT output if it's a part of a tutorial or post about GPT.

[rproffitt]

Yes, you only write that after I reply about what you wrote. In this thread there was no caveat until I complained again.

[Reverend Jim]

My problem was with the "negativity bias" statement although I suppose you could make a case based on evolution. If you pay more attention to negative news than positive news you are less likely to ignore potentially dangerous stimuli (is the rustling bush a predator or the wind).

[AndreRet]

Based on comments above, the good, the bad and the evil will always be there. If posters post exactly what they get as a result it normally points to them not understanding the actual output. I can see this a lot in another similar site on a regular basis. The trick is to analyze the output, take it as a pointer and correct the code - see my example below...

IMHO, ChatGPT is a very good tool to use, yes we have to adapt to it as I stated in my previous post. It all boils down to the prompts given. We have all seen how many times a poster here will state that they Googled an issue but found no resolution, based on their search criteria.

As an example, high level based, I asked -

using php, write a small "hello world" landing page

The return, as expected, with no errors were -

<!DOCTYPE html>
<html>
<head>
    <title>Hello World</title>
</head>
<body>
    <h1><?php echo "Hello, World!"; ?></h1>
</body>
</html>

When I asked for the following, a huge difference can be seen, again as a high level output -

using php, write a small "hello world" landing page. The 'hello world must swirl in and have a popup effect using CSS. Make the 'hello world' clickable as a link to show page2.php. The link must have a 'decoration: none' property. Page 1 and 2 must have a modern colorful background. Make 'hello world' as a variable in php and transfer the string to page2.php. Display the string again on page 2 as a heading.

a Totally different output received where the first solution errored by not showing the animation or the link -

Page1.php
    <!DOCTYPE html>
<html>
<head>
    <title>Hello World Page 1</title>
    <style>
        body {
            background-color: #f39c12; /* Colorful background */
            display: flex;
            justify-content: center;
            align-items: center;
            height: 100vh;
            margin: 0;
        }
        .hello-box {
            position: relative;
            display: inline-block;
            cursor: pointer;
        }
        .hello-text {
            animation: swirl 2s ease-in-out infinite, popup 1s ease-in-out;
            text-decoration: none;
            color: white;
        }
        @keyframes swirl {
            0% { transform: rotate(0deg); }
            100% { transform: rotate(360deg); }
        }
        @keyframes popup {
            0%, 100% { transform: scale(1); }
            50% { transform: scale(1.1); }
        }
    </style>
</head>
<body>
    <a href="page2.php" class="hello-box">
        <span class="hello-text"><?php echo $helloMessage; ?></span>
    </a>
    <?php
        $helloMessage = "Hello, World!";
    ?>
</body>
</html>

    page2.php
    <!DOCTYPE html>
<html>
<head>
    <title>Hello World Page 2</title>
    <style>
        body {
            background-color: #3498db; /* Colorful background */
            display: flex;
            justify-content: center;
            align-items: center;
            height: 100vh;
            margin: 0;
        }
        .hello-heading {
            color: white;
        }
    </style>
</head>
<body>
    <h1 class="hello-heading"><?php echo $helloMessage; ?></h1>
</body>
</html>

In the above the 'hello-box' style were incomplete, the '$hellomessage' and a-link element were incorrect. By changing small things it worked as expected -

page1.php
<!DOCTYPE html>
<html>
<head>
    <title>Hello World Page 1</title>
    <style>
        body {
            background-color: #f39c12; /* Colorful background */
            display: flex;
            justify-content: center;
            align-items: center;
            height: 100vh;
            margin: 0;
        }
        .hello-box {
            position: relative;
            display: inline-block;
            cursor: pointer;
            text-decoration: none;
            color: white;
            animation: swirl 2s infinite, popup 1s;
        }
        @keyframes swirl {
            0% { transform: rotate(0deg); }
            100% { transform: rotate(360deg); }
        }
        @keyframes popup {
            0%, 100% { transform: scale(1); }
            50% { transform: scale(1.1); }
        }
    </style>
</head>
<body>
    <?php
        $helloMessage = "Hello, World!";
    ?>
    <a href="page2.php?message=<?php echo urlencode($helloMessage); ?>" class="hello-box">
        <span>
            <?php echo $helloMessage; ?>
        </span>
    </a>
</body>
</html>

page2.php
<!DOCTYPE html>
<html>
<head>
    <title>Hello World Page 2</title>
    <style>
        body {
            background-color: #3498db; /* Colorful background */
            display: flex;
            justify-content: center;
            align-items: center;
            height: 100vh;
            margin: 0;
        }
        .hello-heading {
            color: white;
        }
    </style>
</head>
<body>
    <?php
        $receivedMessage = isset($_GET['message']) ? $_GET['message'] : 'Default Message';
    ?>
    <h1 class="hello-heading"><?php echo $receivedMessage; ?></h1>
</body>
</html>

Unfortunately not everybody is following these methods because they copy and paste and believe that the code is 100% correct/compliable.

[Dani]

Unfortunately not everybody is following these methods because they copy and paste and believe that the code is 100% correct/compliable.

And therein lies the problem.

[Dani]

Full-stop, copying and pasting ChatGPT responses directly into UGC sites (sites comprised of user generated content like forums and Q&A sites), especially when there is no indication of if the post was composed by a human or not, is guaranteed to lower the overall quality of content on those sites.

[Hasan_25]

AI is made magical change after the covid if anyone will not be adopt this change so than they will face too much criticize in their work.