Re: GPT-4 knows Guile! :)

[Mikael Djurfeldt <mikael@djurfeldt.com>]

[-- Attachment #1: Type: text/plain, Size: 1460 bytes --]

On Sat, Mar 18, 2023 at 9:36 AM <tomas@tuxteam.de> wrote:

> On Sat, Mar 18, 2023 at 09:22:43AM +0100, Mikael Djurfeldt wrote:
> > BTW, in the bouncing ball example, I find it amazing that I could get an
> > improvement of the code by complaining:
> >
> > But all those SDL_ calls look like C bindings. Please use guile-sdl2
> > bindings.
> >
> > (It was also quite entertaining that I had to ask it to write the code
> > "according to the guile-sdl2 manual".)
>
> Perhaps you didn't know, but you are training the model :-)
>

Unfortunately not. I'm prompting it within its 32000 token (GPT-4)
attention span. Next conversation the model is back to exactly the same
state again. Then, of course, it is possible that OpenAI chooses to filter
out something from the dialogs it has had.

So, a trick you can do is to start out every session with a standard set of
prompts (like "keep it short" or whatever) which will then act as a kind of
configuration.

>
> This is called gamification, latest known in early 2000s. Luis von
> Ahn [1] did quite a bit of pioneering work in that (he called it
> "human computation", his PhD was "Games With a Purpose", Google
> licensed a game from him to make people "out there" tag images for
> them.
>
> So I'd say this is established "technology".
>

I think you're right that this will be implemented.


>
> Cheers
>
> [1] https://en.wikipedia.org/wiki/Luis_von_Ahn
> --
> t
>

[-- Attachment #2: Type: text/html, Size: 2292 bytes --]