gave jev a pixel canvas to see if it could draw... nope

okay maybe I just did it wrong

it never reads the canvas back, so every pixel is a guess about a state it cannot see. does it get better if you feed the grid back each step?

This guy tried it and seems to work much better

Give it a paintbrush with P5JS and see what it can do

i actually would prefer this i think? for logistic implementation i mean. i would rather a model that lacks hallmarks of conscious perception performing work than one that has countless hallmarks of it🙃

why are u stingy with the amount of pixels

if it wasn’t so disappointing, the plan was to give it more pixels…

it's giving modern art more than it's giving dog but i respect the commitment

have you tried giving images and rating them?

It is decent with atomic questions composed in code :D Let the LLM draw the ocean/tree and use little Jev to quickly and cheaply judge whether it is an ocean

8x8? u monster!! let 32x32 so it at least have a chance

Pixel canvas: 0, JEV: 1?

also, we're doing this in NYC who's running a Jev hackathon in SF?

If you give it finished "paintings" can it discriminate the correct shapes?

Phew NFTs denied 🤣 👾

What did it draw instead? "nope" with no image of the attempt is the part I'd actually want to see.

Ok but I bet it's really good at tracing

wanna try jesse?

Idk, art is in the eye of the beholder.

respect for shipping the failed draw. the nope is the interesting part 👾

JevAGI when?

Your baby AGI is masterpiece , create new baby jev

I did something similar with music. Meh.

Did you re-feed the canvas into the state each step, or ask for all the pixels in one pass? If it's one pass the branches can't see each other by design, so it sort of has to fail.

Pixel canvas experiment ending in "nope" is still useful research. Negative results from Yohei are half the feed's charm.