Training the AI to write better LiveCode

Geoff Canyon gcanyon at gmail.com
Tue Jan 24 02:26:57 EST 2023


On Mon, Jan 23, 2023 at 4:09 PM Bob Sneidar via use-livecode <
use-livecode at lists.runrev.com> wrote:

> Meh. Don't like using Chess as a measurement of AI competence. Chess is in
> AI's wheelhouse. Once a computer can store all the possible permutations of
> chess,
>

I agree on the sentiment, but a small point: the state-space of Chess is
about 10^44 <https://en.wikipedia.org/wiki/Game_complexity>. No computer
has, or foreseeably will, store all the possible permutations. It's the
nature of machine learning that it's pretty much impossible for us to
understand what something like AlphaZero
<https://en.wikipedia.org/wiki/AlphaZero> *does* store, but it's not
anything like "in this position, play this move". Further, it appears
AlphaZero used 16GB of memory. It looks like Stockfish 10 (which I think is
considered stronger than AlphaZero) can use more, but will operate in 16GB
of memory as well.

gc


More information about the use-livecode mailing list