[prev in list] [next in list] [prev in thread] [next in thread] 

List:       gentoo-user
Subject:    Re: [gentoo-user] has anybody use ChatGPT for programming?
From:       Wols Lists <antlists () youngman ! org ! uk>
Date:       2023-03-22 8:31:15
Message-ID: 1cd05158-728c-ba06-2d36-c5598f35012c () youngman ! org ! uk
[Download RAW message or body]

On 21/03/2023 14:09, Rich Freeman wrote:
> On Tue, Mar 21, 2023 at 9:25 AM Anna<navi@vlhl.dev>  wrote:
> > > Has anybody use ChatGPT for programming?  I think it would very very handy \
> > > (less bugs) and less questions on the mailing-list
> > At least on my experience, it's results are wildly inconsistent for
> > code, specially for something non-trival. It can be confidently wrong,
> > and thus would still require you to know what you want to do and fact
> > check the reply...
> > 
> I know Home Assistant banned posting AI-generated responses on their
> forums when some users decided to create bots that would take
> questions from newbies, run them through ChatGPT, and then auto-post
> the responses as replies.
> 
> The result was mass confusion as often the answers seemed plausible
> but contained subtle errors.  Dealing with the resulting frustration
> and chaos took more volunteer time from those hanging out on the
> forums than just dealing with the original questions.

The results I've come across suggest this is the norm.

So code APPEARS better, but is usually WORSE, than code you actually 
wrote yourself.

So it's probably good for study, or if you want a template to follow, 
but don't blindly use any code. It's only as good as the data it's 
trained on, and as a doctor I know said to me "We select doctors from 
the general population, and if half the population are below average 
what does that say about doctors?".

ChatGPT (code especially) is probably trained mostly on student output. 
Do you really want to be "writing" student grade code?

Cheers,
Wol


[prev in list] [next in list] [prev in thread] [next in thread]