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

[Richard O'Keefe]

I can assure you that ChatGPT's advice about Smalltalk (which is
arguably more "mainstream" than Mercury) has little to commend it.
There's actually a book out there on using AI in programming which has
an appendix listing three sample sessions solving problems
in basic probability.  Two of them are right.  One of them, with a
wealth of convincing detail, is drastically wrong.  The author(s) of
the
book completely failed to notice.  ChatGPT is not really an Artificial
Intelligence system.  Large Language Models are *Information
Retrieval* systems.  Give them a cue and they regurgitate stuff, but
they have no actual *understanding* of what they mash up and
spit out.  If other people often have your problem, you may be in luck.

On Sat, 29 Jun 2024 at 01:39, Anders Lundstedt
<mercury-users at anderslundstedt.se> wrote:
>
> Dear all,
>
>
> I tried ChatGPTs "Code Copilot" (link to interaction log below) with
> the intent to see if it could tell me how the fix the following
> problem that took me embarrassingly long time to resolve. When I code,
> I mainly do so in Python where the string type is named ‘str’. Thus I
> typoed ‘uint_to_hex_str’ instead of ‘uint_to_hex_string’, and got the
> thus expected error message.
>
>
> I cannot say I am very impressed with Code Copilot's Mercury competence:
>
> https://chatgpt.com/g/g-2DQzU5UZl-code-copilot/c/61c0abd5-8ab0-4e25-b2ab-0ca6d64673f4
>
> (While my intent was to see if Code Copilot could help me with the
> problem just mentioned, I got a bit carried away already at the start
> and asked for advice on how to implement a parser for a markup
> language. For its answer to the above problem, see the end of the
> interaction log.)
>
> (I have almost no prior experience interacting—for any purpose—with
> large language models so I do not know how Code Copilot's Mercury
> competence compares with its competence in more mainstream programming
> languages.)
>
> (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.)
>
>
>
> Best regards,
>
> Anders Lundstedt
> _______________________________________________
> users mailing list
> users at lists.mercurylang.org
> https://lists.mercurylang.org/listinfo/users