[m-users.] An interaction with ChatGPTs "Code Copilot"

[Sean Charles (emacstheviking)]

I tried CoPilot for two weeks before throwing it out. The amount of code it offered was growing ever larger (Python, for Django) and taking me longer to read and make sure it made sense. A lot of times it didn't or relied on a context I didn't have at that point in my code.

As for the FORTH code I asked it for one day over lunch... oh dear.

I wish the media would realise that LLM, as Richard states, are just parrots trained on a lot of raw code, both correct and incorrect.

Squawk!


> On 29 Jun 2024, at 07:31, Zoltan Somogyi <zoltan.somogyi at runbox.com> wrote:
> 
> 
> On 2024-06-28 23:27 +10:00 AEST, "Anders Lundstedt" <mercury-users at anderslundstedt.se> wrote:
>> (To avoid a future dead URL in the mailing list archives, I tried
>> saving the above URL with the archive.org Wayback Machine, but it
>> seems that does not work.)
> 
> I am afraid I have to say that for many people, including me,
> the link is *already* effectively a dead link.
> 
> LLMs cannot ever be trustworthy assistants to programmers,
> for exactly the reason Richard described: they do not even *try*
> to understand the texts they scan. I know enough about the
> variouts subfields of AI to have known this since the hype
> started around LLMs. I therefore have neither a copilot
> account, nor any desire to get one, and I don't want to get
> on any mailing lists about them either, which is why my
> response to the sign-in request that you get at the link
> you posted was to close the browser tab.
> 
> You may want to check out the email thread started by this post
> 
>  https://lists.mercurylang.org/archives/users/2023-October/010224.html
> 
> which turned to be about an LLM confidently sprouting incorrect info
> about Mercury.
> 
> Zoltan.
> _______________________________________________
> users mailing list
> users at lists.mercurylang.org
> https://lists.mercurylang.org/listinfo/users