chatGPT knows py5! (although not so good)

[Maltiec]

Some days ago my student showed me sketch in py5 made with a help of chatGPT, I've decided to make some experiments too. This one looks nice:

import py5

def setup():
    py5.size(400, 400)

def draw():
    py5.background(255)
    x = py5.mouse_x
    y = py5.mouse_y
    d = py5.dist(py5.width / 2, py5.height / 2, x, y)
    # Calculating circle color
    r = py5.remap(x, 0, py5.width, 0, 255)
    g = py5.remap(y, 0, py5.height, 0, 255)
    b = py5.remap(d, 0, py5.width / 2, 255, 0)
    py5.fill(r, g, b)
    py5.stroke(0)
    py5.ellipse(x, y, 50, 50)

py5.run_sketch()

Results:

1. chatGPT knows py5 (though in above example it made a mistake and used py5.map instead of py5.remap, I've made correction myself)
2. For now if I try to tell it to use objects and py5, it tells that doesn't know how to code with py5 and shows the alternative with pygame.
   I'll continue my experiments. )

[Maltiec]

update - it can use objects if I insist on using class and give a name for it:

import py5

class myCircle:
    def __init__(self, x, y, size=50):
        self.x = x
        self.y = y
        self.size = size
        self.color = (0, 0, 0)
    
    def display(self):
        py5.fill(*self.color)
        py5.circle(self.x, self.y, self.size)
    
    def move(self, dx, dy):
        self.x += dx
        self.y += dy
        self.color = (py5.map(self.x, 0, py5.width, 0, 255), 
                      py5.map(self.y, 0, py5.height, 0, 255), 
                      0)

circle = myCircle(py5.width // 2, py5.height // 2)

def setup():
    py5.size(400, 400)

def draw():
    py5.background(255)
    circle.display()
    circle.move(py5.random_uniform(-5, 5), py5.random_uniform(-5, 5))

py5.run_sketch()

[hx2A]

No, Py5Vector didn't appear until 0.7.0.

https://github.com/py5coding/py5generator/releases/tag/py5-1277-0012-0.7.0a0

Btw, it generated code for the same prompt when I changed python to js and py5 to p5

Hmmm, so perhaps ChatGPT has some ability to extend its understanding of p5 to py5, but it is not complete.

I wonder how ChatGPT will impact the libraries and code people use? I imagine that libraries that ChatGPT can't write good code for will be used less. Will this start a cycle, where they continue to get used less often, and others that it does know get used more? I have concerns this will have a negative impact on the open source world's ability to innovate.

[hx2A]

BTW, there is a py5 release today!

https://github.com/py5coding/py5generator/releases/tag/py5-1292-0019-0.9.0a0

[Maltiec]

No, Py5Vector didn't appear until 0.7.0.

https://github.com/py5coding/py5generator/releases/tag/py5-1277-0012-0.7.0a0

Btw, it generated code for the same prompt when I changed python to js and py5 to p5

Hmmm, so perhaps ChatGPT has some ability to extend its understanding of p5 to py5, but it is not complete.

I wonder how ChatGPT will impact the libraries and code people use? I imagine that libraries that ChatGPT can't write good code for will be used less. Will this start a cycle, where they continue to get used less often, and others that it does know get used more? I have concerns this will have a negative impact on the open source world's ability to innovate.

Well, further experiments show that it is a problem of understanding by network, so after I told her to evaluate her code by using Py5Vector class she (ehh, I am a bit of AI optimist so sometimes I see a personality inside ) ) made refactoring, and used vector class from Processing Python mode. This is chatGPT3.5, I'm going to try v.4 - may be it'll work better. And also there is a fine tuning available - may be some new training data could be injected into network.

[Maltiec]

update: chatGPT4 is much better at py5 programming. Either dataset was updated, or she is really can "think ahead" when lacks examples.

[hx2A]

Excellent. I'm glad ChatGPT4 can write solid py5 code. That can only help encourage new users.