Artificial General Intelligence (AGI) Blueprint

Visual Thought: A Tiny Demo of How Minds Actually Think

Does an LLM actually see anything when it talks?

When you read "a misty forest," you picture it. When you imagine a blue butterfly in a garden, you see it — instantly, vividly, without trying.

LLMs don't do that. They generate words about scenes without ever forming one.

This tiny demo shows the alternative: a system that imagines a real scene, stores it using a phonetic mnemonic peg system  and recalls it back — the same way human memory actually works. Then it asks an LLM to do the same thing... and watches it fail.

Click through:

  • Imagine — picture a garden with a blue butterfly
  • Store — watch it compress into a symbolic memory code
  • Recall — bring the image back from the code alone
  • LLM Imagine — see what happens when you ask an LLM to do the same

This is a small proof-of-concept piece from a much larger cognitive architecture project (NMCA) exploring visual-symbolic reasoning for AGI. Full research paper linked below for anyone who wants to go deeper.

-- 

📄 Full 471-page AGI research blueprint: Zenodo link

--

AGI Mnemonics encoding and decoding 3D scenes 


(Mental imagery - explicit version with Unity 3D) 

* Quick refresher on the phonetic Major System *

 0 = s, z, soft c 

1 = t, d 

 2 = n 

3 = m 

4 = r 

5 = l 

6 = j, sh, soft ch, soft g

 7 = k, hard c, hard g, q -

8 = f, v 

9 = p, b 

Vowels, w, h, y are ignored — only consonant sounds count

 "Garden": G-R-D-N → 7-4-1-2 → 7412 

"Butterfly": B-T-R-F-L → 9-1-4-8-5 → 91485 

--- 

(Encoded) "butterfly → garden" → 91485 → 7412

 "blue butterfly → garden : 95-91485-7412" 

-- 

(Decoded)  95-91485-7412 -- "blue butterfly → garden"

More information