"ingo" <nomail@nomail> wrote:

> I tried a few things in the past but most of the time it was clueless. Claude
> seems to work better than OpenAI. Recently I had a chance to play for a short
> while with a payed version of Claude 3.7. It's a lot better than earlier ones.
>
> When one can keep the discussion on an aspect small and focussed, Claude 3.5 is
> mostly fine.

I was able to play with M$ Copilot for a bit yesterday, and it's possible to get
a lot more out of it if you don't ask it to do it all at once.
(I was actually surprised how much you were able to get with your simple initial
prompt!)

Ask it to do a small bit, then the next bit, then the next bit, then ask it "can
these all be done together in one scene / macro?

So I got it to spit out some Householder transformation code, and then some QR
decomposition code, and Hopefully I can use that as a guide to get the SVD
working in SDL.

I'd also like to get the red-black data structure hammered out, and then get
Fortune's algorithm working for a Voronoi diagram.

Maybe we can't get everything we want implemented in source (yet), but we can
get a bunch of features implemented via macros - however slow they might be.

I'm wondering if a concerted effort to leverage all of the AI's out there might
let us start to unravel all of the parser stuff and get that all sorted out so
that POV-Ray 4.0 can start to be worked on.

- BW

Post a reply to this message