LÖVE

So I saw a video about LLMs playing Minecraft which lead me to viewtopic.php?t=85089

Took that code but it didnt work so I put it into Llima 3.1 that didnt work so i put it into GPT4o and that kinda worked. SO i put it into Gemini 1.5 pro and that helped. Finally I put it in Claude Opus Free and now we got:

Simple Neural Network With Multiple Hidden Layers

Ive got 1% of the knowledge in the origonal post, I believe this works but honestly im throwing stones in the dark and guessing if i got it right based off the sound

Anyways I'll attach the multiple layers version here. I have a single hidden layers version here: https://github.com/NoodlesReturn/Lua-LO ... etwork/wik ,which should help explain anything.

Its been a good 72 hours of prompting pasting and crying )

My next steps are to impliment this into a game so I can test how good it is and improve accordingly. Since im a noob this might take awhile so if there are any AI wizards out there make urself known!

Let me know if you do any tests and what you find. Let me know if the whole thing is trash lol. Good luck!

PS appreciate this runs on CPU not GPU which will effect things. I will get round to CUDA if i get this part going.
Currently limited by LLMs Output being roughly 8000 characters. There are ways around this but I think I need to take a break from this and refocus.
I definetly feel my next steps will be to modulise this code into multiple files. If anyone can help me with that.....

Not sure why you are trying to reinvent a wheel, additionally doing that through a black box of LLMs?

There are plenty of Neural networks already implemented in Lua, i.e. just search Github. LOVE is Lua, so anything can be easily integrated.

I couldnt understand NN so i figured making one from scratch would be a good way to learn. I also want to improve my love2d skills so 2 birds with 1 stone. Unfortunatly no amount of youtube or articles seem to be helping me understand. I've tried python modules but, I dont really understand what's going on. It doesnt feel like deep full understanding or learning. I find this brute force method of failing and trying to understand why to fix the best way to learn. Equally I dont want to be in the complete dark so my LLM black box function as somewhat dysfunctional teachers. I really wouldnt want to be posting all the questions i've had all over the internet haha there have been too many, and many more dead ends...

And like many, I am really curious on the limits of LLMs as well as which is best for what. improving my prompting skills has been another benefit. i like the term LLM black box though xD

Guess im trying to reinvent the wheel because, i feel like it, no one else has, its fun, perhaps itll help someone whose googling along the way, i have 11 hours till my deadline and im still procrastinating with this >-<, why not? lol i guess i dont really have a reason but without getting too philosophical, why do anything?

On that I've updated it further so it can handle different activations and residual connections (need to go over this part tho). Might be one or two other things i've forgotten but I will do a full feature update when I reach completion. Next I want to look into better ways of displaying the error information. I also really need to split this into seperate files at some point soon. I have collaborated a roadmap that will keep this thing from becoming a DQN or a RNN but still a complex 'simple' NN. I now have a rough goal of where I want to take this thing. Hopefully then, i will feel more comfortable using something like pytorch.

If it was for fun and learning, sure. Still easier and much better to read and learn by checking proper code of somebody's else library then compiled crap code generated by LLMs.

Added some more features. Theres a few issues here and there but its getting there