Does ChatAI make StackOverflow obsolete ?

[david_navigator 12]

Thought I'd just start a little discussion about using ChatAI for code samples as it's proved useful on a couple of occasions recently.

 

[david_navigator 12]

[david_navigator 12]

Interestingly none of the code samples it's given me so far have worked "out of the box" & many have contained deprecated or obsolete units. From what I can see, looking at the ExpertsExchange data it seems to have used, it weighs older information higher than recent information, which is of course not what's wanted with delphi.

[Sherlock 663]

33 minutes ago, david_navigator said:

Interestingly none of the code samples it's given me so far have worked "out of the box"

Well, there's your answer. And also why SO has banned AI generated answers. So...actually no need to start a new thread. Just chime in one of the other threads:

 

 

[david_navigator 12]

Thanks. I did a search but nothing came up - maybe the search needs more AI 🙂

[Brandon Staggs 278]

As interesting as it is, it regularly provides inaccurate details for just about everything I have tried with it. I would not use it for code. It may be useful to get the mind jogging on something but it is not a replacement for a human expert.

[Stano 143]

I know/I read of two cases where programmers used it. AI reduced their time to create the right code from a week to a day. No, they didn't get the right result. But AI led them to it.

He who knows what he wants and what he is doing will help himself.

[david_navigator 12]

2 hours ago, Stano said:

I know/I read of two cases where programmers used it. AI reduced their time to create the right code from a week to a day. No, they didn't get the right result. But AI led them to it.

He who knows what he wants and what he is doing will help himself.

That's the experience I've had. Each sample of code has not compiled out of the box, but fixing it has given me an understanding of the code or pointed me in the direction of something I didn't know about.

[Rollo62 536]

Quote

Does ChatAI make StackOverflow obsolete ?

Nope.

I think that chatGPT is trained and powered from SO, beside many other sources.

So it should not be able to replace its original sources, since the global "brain" of millions of human developer brains is simply to valuable for new input and failure corrections.

[Brandon Staggs 278]

On 1/12/2023 at 9:28 AM, Stano said:

I know/I read of two cases where programmers used it. AI reduced their time to create the right code from a week to a day. No, they didn't get the right result. But AI led them to it.

He who knows what he wants and what he is doing will help himself.

I can see using it to get started in a direction, I guess, but my initial tests with it soured me on trusting it. I asked it to describe various things that are non-trivial but I know a lot about, and every single time, it included false information on those topics. For example, I asked it about my own software, and it listed compatibility with operating systems that are expressly not supported. I asked it to summarize the events of a Star Trek episodes and it got lots of details breathtakingly wrong. If I can ask it questions I already know the answer to, and it gives me wrong results, how am I going to trust it to help me on things I don't know?

[David Heffernan 2345]

I'd live to see it answer some of the questions that have been answered on SO, but without being able to refer to SO. In short, no, AI is still a long way from making human though obsolete. It can be a really power tool to use alongside it. 

[Brandon Staggs 278]

I think the term "Artificial Intelligence" is a misnomer. It makes people think there is some rudimentary form of "intelligence" at work in the software when it is nothing more than pattern matching and following models. ChaptGPT is just a trained language model. Nothing happens to "decide" what to tell you when you ask it a question, and it is trained on pre-existing massive blocks of text. Traversing its text data and pulling out chains of words that match it's training scores is not in any way "intelligent" response to your programming questions. Frankly, this whole thing is doomed to failure if there is an expectation that this will eventually produce consistently good material that can replace actual human consideration of a problem. Also, read this book, and stop being impressed by these tools:

 

https://amzn.to/3CUXWm4

 

And when I say "stop being impressed," I don't mean that there is nothing impressive or useful about "AI," but that it is nothing impressive compared to actual functioning intelligence and probably never will be. And if you think it will be, read that book, and see if you still think so. AI will always have uses, but I doubt it will ever go far beyond classifying things and pattern matching.

[dummzeuch 1505]

1 hour ago, Brandon Staggs said:

I think the term "Artificial Intelligence" is a misnomer. It makes people think there is some rudimentary form of "intelligence" at work in the software when it is nothing more than pattern matching and following models. ChaptGPT is just a trained language model. Nothing happens to "decide" what to tell you when you ask it a question, and it is trained on pre-existing massive blocks of text. Traversing its text data and pulling out chains of words that match it's training scores is not in any way "intelligent" response to your programming questions. Frankly, this whole thing is doomed to failure if there is an expectation that this will eventually produce consistently good material that can replace actual human consideration of a problem. Also, read this book, and stop being impressed by these tools:

 

https://amzn.to/3CUXWm4

 

And when I say "stop being impressed," I don't mean that there is nothing impressive or useful about "AI," but that it is nothing impressive compared to actual functioning intelligence and probably never will be. And if you think it will be, read that book, and see if you still think so. AI will always have uses, but I doubt it will ever go far beyond classifying things and pattern matching.

Sorry, I couldn't resist letting ChatGPT write a reply to this:

Quote

Thank you for sharing your thoughts on the term "Artificial Intelligence." It is true that the term can be misleading and that current AI technology is primarily based on pattern recognition and following pre-existing models. However, this does not mean that AI is not useful or impressive in its current form. AI has a wide range of practical applications such as image recognition, natural language processing, and decision making. While it may not be able to fully replicate human intelligence, it can still be a valuable tool for automating certain tasks and making important decisions.

As for the book you have recommended, I am an AI model and i do not have the capability to read books. But I would like to remind you that AI is constantly evolving, and while it may not currently be able to fully replicate human intelligence, there is still potential for it to become more advanced in the future.

It is important to have realistic expectations and understand the limitations of AI, but it is also important to recognize its potential and uses in our society.

(I'll try to resist in the future though because I don't want this to escalate into a thread that consists of ChatGPT generated stuff only.)

[Brandon Staggs 278]

49 minutes ago, dummzeuch said:

(I'll try to resist in the future though because I don't want this to escalate into a thread that consists of ChatGPT generated stuff only.)

😄

 

Also wanted to recommend this other book:

 

The Myth of Artificial Intelligence: Why Computers Can’t Think the Way We Do 

https://amzn.to/3CWFOIP

 

The author has impressive experience and credentials on the topic. Parts of the book should have been left out because they detract from the stated premise of the book, but he does a good job explaining why there is essentially zero chance that what we regard as AI today could ever develop into something remotely approaching actual intelligence.

 

If anything, ChatGPT has only solidified in my mind the belief that machine learning is never going to be useful beyond what it already does well -- classification. I am just not impressed with the fact that it can create a lot of text on a topic strung together with passable grammar.

[David Heffernan 2345]

I think the answer to the original question is no

[aehimself 396]

54 minutes ago, David Heffernan said:

I think the answer to the original question is no

I don't know which one is funnier. The actual punchline or the ChatGPT answer.

Well played AI, well played.

[Juan C.Cilleruelo 12]

I think that OpenIA needs Stackoverflow as the feed of his responses. 
Because OpenIA is not able to investigate the solutions. Only can locate and infer what is the most accurate response.

To be sure that the response is the most accurate, it needs many wrong responses, which are the most frequent in StackOverflow.

Edited January 19, 2023 by Juan C.Cilleruelo

[David Heffernan 2345]

What AI could do is the websearch that so many askers on SO don't do, and thus avoid there being so many duplicate questions posted. That would be a plus. 

Edited January 19, 2023 by David Heffernan

[David Heffernan 2345]

deleted (why can't we delete posts)

Edited January 19, 2023 by David Heffernan

[Brandon Staggs 278]

I had better luck with asking it for a programming joke.

 

[Brandon Staggs 278]

21 hours ago, David Heffernan said:

I think the answer to the original question is no

ChatGPT agrees.

 

[aehimself 396]

15 minutes ago, Brandon Staggs said:

ChatGPT agrees.

Ask if it has any affiliation to / what it knows about SkyNet as it already seems to be self-aware 😮