seems like there's a small problem loading tilemaps...

🌼🌼🌼🌼🌼🌼

🌼

that's more what I had in mind

for context, I manually implemented the camera and raw triangles for this cube, and got it drawing correctly. it was fun!

I'm investigating whether I should make a custom engine for my minecraft-like game

and then I doodled some robots on the facing page

my friend asked me about the shadows. it's just Godot's default lighting. I know minecraft has some kind of cool custom lighted based on the voxels, but I'm not sure I want to replicate that exactly...

again, like with the problems playing NetHack, LLMs seem to "understand" many things that they are subsequently unable to act upon

this is interesting though: "Agents pinpoint the source
of an issue remarkably quickly... often far faster than a human would. However, they often exhibit a limited understanding of how the issue spans multiple components or files... leading to solutions that are incorrect" (section 3.7)

ok reading the rest of the paper: while making LLMs play NetHack is very funny, it's not quite a fair test of their capabilities. they're not designed to solve problems through tiny discrete actions ("move left") or navigate spaces, and it's not how people are generally interested in using them

so even though the AI knows about NetHack, and knows that eating rotten food in NetHack will kill you, it still chooses to eat rotten food? what's going on? why can't it access this knowledge when it's actually making decisions in the game?

I find it fascinating