Path: news.eternal-september.org!eternal-september.org!.POSTED!not-for-mail
From: Richard Heathfield <rjh@cpax.org.uk>
Newsgroups: comp.lang.c
Subject: Re: "C" code from bots (was Re: Oh d-ai-ry d-ai-ry me)
Date: Sat, 31 May 2025 12:53:01 +0100
Organization: Fix this later
Lines: 38
Message-ID: <101eqj0$13q67$1@dont-email.me>
References: <101d2re$l1ta$1@dont-email.me> <101enkq$13730$1@dont-email.me>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Injection-Date: Sat, 31 May 2025 13:53:04 +0200 (CEST)
Injection-Info: dont-email.me; posting-host="d86e0294dcfbfa4a1a2c2810fce9721a";
	logging-data="1173703"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX19aOBxuxuEf59rkaiymRlk4Y1ljdQbu+fY8fcSl5jx0jg=="
User-Agent: Mozilla Thunderbird
Cancel-Lock: sha1:v9k5KKkV1AHFTytnBiNtfs/cd50=
In-Reply-To: <101enkq$13730$1@dont-email.me>
Content-Language: en-GB

On 31/05/2025 12:02, Janis Papanagnou wrote:
> On 30.05.2025 22:01, Richard Heathfield wrote:
>> This evening I accidentally bumped into a chatbot, and I wondered
>> whether it was any good at C programming.
>>
>> I was impressed when it solved Goldbach's Conjecture (admittedly for
>> small n) and 5-disc Hanoi, but then I thought back to a recent thread
>> here (which I have no intent of resurrecting), so I asked it if it could
>> have saved me a job if I'd only asked it a few hours ago.
> 
> Incidentally we yesterday sat together in a beer garden, and for some
> computing problem a friend asked a bot for programs ("C" and Perl) to
> a question. We got them and another friend was keen enough to test it.
> It was practically unusable for the parameters due to its algorithmic
> complexity and for reduced parameter sizes it just didn't work. - The
> impression I got from looking at the code was that the bot might have
> just picked some (maybe popular) public sources matching the question
> and reproduces those in the answer. - Are the bots really capable to
> "develop" programs?

Probably not, as my C90/C99 test more or less illustrated. It 
'knew' of a syntax rule, but failed to apply it correctly.

I just tried again, this time asking for a C90 program to produce 
a frequency count of colours in a 24-bit bitmap.

It took three laps to make it C90-legal.

I can see how a programmer might be tempted to use AI to get a 
first draft, but you wouldn't want to bet the farm on the 
unreviewed code.

-- 
Richard Heathfield
Email: rjh at cpax dot org dot uk
"Usenet is a strange place" - dmr 29 July 1999
Sig line 4 vacant - apply within