systemMetrics

[richmond62]

Playing around with ChatGPT . . . .
-

-
I could not find 'systemMetrics' in the Dictionary . . .

[jacque]

This is an example of hallucination, the tendency of chat bots to make up false information. I was playing with Google Bard which did the same thing. I corrected it through various interactions until it learned the language. At the end it thanked me for teaching it and said I obviously was fluent in LC script.

[richmond62]

Which tells me it might be easier to eschew A.I. altogether rather than spend hours training the thing. After all if one has to train the thing that completely defeats the object.

[jacque]

ChatGPT has been trained in Python and Javascript, if I remember right (or maybe that's Bard?) but LC is obscure enough it isn't in line for formal training by its makers. That's why I decided to see if it could learn from me. It was kind of fun actually. The thing caught on pretty fast.

[richmond62]

Teaching children EFL, and, more importantly, to think critically, will do me: I'll leave teaching the robots to you.

[richmond62]

The thing caught on pretty fast.

It caught me pretty fast with 'systemMetrics'.

[jacque]

If anyone is interested, here's a link to a Google Docs copy of my chat with Bard:
https://docs.google.com/document/d/1SpM ... sp=sharing

My prompts are in bold. You'll notice it hallucinated a "getAllControls" function which doesn't exist in LC -- but it knew to use camel-case and parentheses. Still, I was impressed enough that I didn't proofread carefully and accidentally misguided it when asking it to only match card controls. I had to correct myself but it accomodated that well.

The only thing that bothered me was how fawning it was. A simple "oh right" would have sufficed.

[AndyP]

I've been using chatgpt with Livecode in a different way with improving results.

I'm feeding in working python code and asking for a Livecode equivalent.

When I first started, about a month ago the results were to say the least not encouraging, with chatgpt basing results on functions that do not exist in Livecode. This has needed a lot of coaching and correction.
I've noticed recently that the results I'm getting now are improving, not so many false functions being returned, so it does look as if chatgpt is slowly learning Livecode.

This is encouraging and with more livecoders input this will only improve (as long as the information in is correct) .

It's going to take a while, but worth the effort.