Hi,

On Wed, Jan 7, 2026 at 1:05 AM Charles R Harris via NumPy-Discussion
<[email protected]> wrote:
>
> Hi All,
>
> Some of you might find Matthew's post on using AI for coding useful. It also 
> contains a link
> to his rewrite of NumPy in Rust as an example.

Thanks for posting this - it's an thoughtful and enjoyable read.

To summarize, Matthew's big picture argument is that AI makes it much
easier to generate code, which, in turn, makes it easier for him to
think about more fundamental issues like design.  He finds AI helps
him to "climb the abstraction hierarchy".

He notes that "LLMs generate a lot of junk" but compares this to "...
concerns that I suspect came around when we developed compilers and
people stopped writing assembly by hand, instead trusting programs
like gcc to pump out instruction after instruction of shitty machine
code."   Then he describes the various practical ways he uses to
control and detect LLM code junk.

Matthew's piece is compelling, but I wonder what the circumstances
are, in which we'd accept a compiler that generates a lot of junk - in
the sense of incorrect or badly-written code.  I don't think I'd trust
myself with such a compiler, but I'm sure there are some people, such
as Matthew, who do have the levels of understanding and discipline to
make that work.

He's using the Claude command line agent.   I tried the Pro plan,
briefly, at $20 billed monthly, but very quickly exhausted the usage
limits.   I wonder if Matthew is using the $100 pcm Max plan, which I
can't afford.   I can see some routes to an end to the age of the
"hobbyist" (https://en.wikipedia.org/wiki/An_Open_Letter_to_Hobbyists).

Of course, as I keep saying in various places, here's still the
problem of voiding copyright.

But it's good to see the discussion becoming much more substantial.

Cheers,

Matthew

-- 
This email is fully human-source.    Unless I'm quoting AI, I did not
use AI for any text in this email.
_______________________________________________
NumPy-Discussion mailing list -- [email protected]
To unsubscribe send an email to [email protected]
https://mail.python.org/mailman3//lists/numpy-discussion.python.org
Member address: [email protected]

Reply via email to