File size: 534 Bytes
5298d15
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
# Gpt-Smart-Vision

## Overview
**Gpt-Smart-Vision** is an 'Impossible AI' architecture that combines a frozen ResNet-18 vision encoder with a tiny recursive folding brain (~64k parameters).

### Mechanism
It uses a **Vision-to-Latent Projection** to map high-dimensional visual features into a recurrent manifold. The model uses an expanded **Mega-Tokenizer** (156k tokens) which includes 5,000 specific visual primitives.

### Developer
Created by **BikoRiko**, a Grade 8 developer. Part of the 'Impossible but Safe' AI initiative.