ChatGPT discussion, and ChatGPT-generated post examples

I’ve been playing with these LLMs for a while, and find them pretty impressive. Sure, it’s a programming trick, but the results can be good. And, the results can be bad. Like the lawyer referenced above, I asked ChatGPT to write some powershell code for me, and it literally invented some functions that did not exist in the API I was using. It took me a while to figure it out – I could not see why it was not working. I finally realized what was happening, and challenged ChatGPT- it apologized, and said, “yes, I made up that function, it does not exist, but it should.” I asked it to try again, and again it made up a different function. Very strange, but clearly these tools need to be used carefully.

They will get better, and I agree with those that say that they will have a huge impact in the coming years.

how do you impersonate a frog

At least it didn’t come up with “you carry a baguette under your arm, and wear a beret and stripey tee-shirt, with a string of onions round your neck”.

Or “leesten verry karefully I vill say thees only wuns” "allo allo.😃

I have tried using chatGpT plus GitHub copilot to try to build an iPhone app quicker in SwiftUI with a php backend. I am familiar w app dev using UIKit and old php, but wanted to try to do it in the more modern framework.

Would have involved a lot of google, try and error, but with the two tools was amazingly fast, as it could generate a lot of the code skeleton in the right way. And with copilot plus of course quite a bit of manual typing, I got the app working much quicker than I think I would have.

It’s actually flying related, will probably start a new thread on it. But it’s a way to issue apple wallet boarding pass for your passenger :)

I have to say, I am an old timer for programming but very very impressed how useful it is for a semi real project (app open source and published on the App Store)

All the GitHub read me, App Store description and disclaimer are written using chatGpT (fully disclosed, not trying to hide it)… if you ever get to it, i think the humour in the disclaimer is quite cool …

HTTPS://GitHub.com/roznet/FlyFunBoarding

johnh wrote:

At least it didn’t come up with “you carry a baguette under your arm, and wear a beret and stripey tee-shirt, with a string of onions round your neck”.

Hilarious!

While the post in question may indeed be an output of ChatGPT or a similar algorithm, some forums have been full of such posts for years, especially Microsoft-owned ones. For virtually every obscure Windows problem reported there, there would be at least one irrelevant answer to the tune of “update your drivers, run system file checker, check for viruses, and don’t even think of using pirated software”, usually with some kind of Microsoft certification mentioned in the answerer’s signature.

Fortunately, the more wicked among AI developers are already working on AI tools aimed at descriminating AI-generated text from human one

A friend of mine is doing is masters in uni at the moment.

He was telling me that after they handed in their last assignment and the lecturer was talking to them about the results, the lecturer had to tell everyone that they are not to use ChatGPT for their work (they can use it to help research but not to write the assignment report).

He pointed that the answers were very bland and not very good anyway, so even if it’s not picked up, you aren’t going to get marked very well for it.

But then he went on to say “And if you do decide to take your chances with getting ChatGPT to write your report, then at least have the intelligence to remove the ‘regenerate response’ before copying and pasting it!”

For those that haven’t used ChatGPT, at the end of the answer there is a button “Regenerate response” where you can get ChatGPT to answer it again (useful in some situations). The students were selecting the text including that button and pasting it straight into their report without reading what they were doing!

Peter wrote:

in some cases they benefit by building up posting reputation points which enable a later post on something else

I think this is the real reason. Not necessarily a point system but a post count. They they do come out with some sort of commercial post, they look like a long time user.

Peter wrote:

ChatGPT is only doing what 99.35% of coders have always done: hit Github / Stackexchange / etc And in the best open source tradition you then spend 6 months making it work properly, because almost nobody posts fully debugged sources to anything (large OS projects like LWIP excepted).

Probably, but I found ChatGPT a much more efficient workflow than Google/stackoverflow, because it provides answers merging multiple query in a very targeted and customised manner.

Yes, if the project is large or a specific new library/optimisation/new approach, it won’t probably help much, and you’ll have to spend most of the time on the debugging/tweaking.

but for putting ideas together like “What if I could have an app that manages and issue boarding pass for apple wallet”, which is not rocket science to do, or require any type of advanced algorithm, just put together a lot of components in a specific and unique way, the tool helps you do it much quicker I think because it goes beyond each individual component (like google/stackoverflow would do) but can answer specifically for the overall goal.

I think to work the tool still requires the user to have good programming and domain knowledge on the topic though, I wonder if it will ever work and solve such problem for a complete newbie to programming…

I know this is going a bit off topic on the use of chatgpt for forum post, but tangentially relevant hopefully about the technology.