|
import streamlit as st |
|
|
|
|
|
st.title("CodeBlast Dream Catcher") |
|
|
|
|
|
content = """ |
|
**CodeBlast Dream Catcher** proposes an alternative approach for searching multidimensional space for knowledge based on the following principles: |
|
|
|
1. An "all possible combinations space" exists in a multidimensional space where one doesn't create knowledge but discovers it. |
|
2. This multidimensional space is best searched with LLMs using goals, as goals carry the recipes for accomplishing them. |
|
3. There exists a multidimensional "all possible" Codestral goal space consisting of interconnected goals resembling a graph. |
|
4. This multidimensional goals space can best be searched by remapping it to the 2D Infinite Canvas proposed in the LabLab.ai Build Your Business StartUp Hackathon's "Navigating the Infinite Plane". |
|
5. The infinite canvas can be created using a 50256 base number system derived from the GPT-2 tokenization labels. |
|
6. To avoid the costly computational expense of base number conversion, hidden and unhidden states are created in the 2D infinite plane. |
|
7. These hidden and unhidden states correspond to the conscious and unconscious mind, proposing that the human brain uses a similar mechanism to avoid the heavy cost of base number conversion. |
|
8. Thus, searching for knowledge becomes a simple mapping problem in 2D and 1D space in both hidden and unhidden states. |
|
""" |
|
|
|
st.markdown(content) |
|
|