Nah, probably mildly similar to how those links are, but certainly not accurate. It would have to do an enormous amount of additional computation and would need to have a langchain sort of setup to actually get genuine weights mapped from its neural network and would probably take days.

(i saw a visualization in a video of an actual LLM network of vectors and their weights in 3D space and it was insane. I think it was a simplified one and probably would require an absurd amount of RAM just to load and navigate through.)

It's response:

"Visualizing these connections in a 3D space would require immense computational resources and specialized software designed for high-dimensional data visualization. Moreover, the sheer scale and complexity of the model make it challenging to create a meaningful and comprehensible visual representation.

While I can create abstract or artistic representations of a neural network, these would not accurately reflect the specific connections and weights of the GPT-3.5 or DALL-E 3 models. They would be more symbolic than factual."

But still i thought the image was cool and fun to share. 😆 Neat to think about it actually trying to represent that information visually and how those weights would work in a self-reflective way, though.