stiff dome
note: i’m just a 14 yo newbie in this stuff (i watched no courses). i just tried to observe my behaviors and how people react, then implement it into code with no hard coding (just raw). so don’t expect much — i just need feedback.
📝 details:
built from scratch — no ML libraries or frameworks
just pure logic + 3 built-in libraries
it tries to mimic how a brain works: raw, flexible, always learning
it’s so optimized it can run 10+ layers on a phone
it currently predicts the next character from an input string.
i paused the looping/thinking part until i confirm it’s actually learning properly .
it’s super modular — you can tweak the number of neurons/layers in real time, and it adapts on the fly (like a real brain).
it even handles corrupted or missing save files like a champ.
ASCII input only for now, but i’m working on vision + hearing modules next.
it learns live while running, so yeah… it might fry your potato laptop.
you can even mutate two brains by merging their save files (yes, brain children exist 💀)
this is the second version, btw.
🔗 links:
VirusTotal scan (clean): https://www.virustotal.com/gui/file/4f3a7204855cc0a9812ae45b55c1ffbf214b8887044cdabe83b6b96983f2d5df/summary
Download the .exe from Google Drive: https://drive.google.com/file/d/1YesKhmloboCwGVPYTlfraLm2peAe-elR/view?usp=sharing
watch a video where i train it on a phone(yes.... a phone): https://drive.google.com/file/d/1GAG6gyBuZ8DTH6OetRIpb-K9GKizjPgO/view?usp=sharing
🔎 if you want proof it works:
check the video link above where i enter “hello” twice on a phone 📱
you’ll clearly see the total error drop massively in just one loop (like 634 to 62, or 500 to 2)
⌨ commands:
/options — enter setup process
/save — saves current weights
/exit — saves + exits
⚠️ note:
use a lower learning rate if the total error keeps going up (might be exploding gradients — i think it can recover from it though)