LLMs do not have a subconscious

it can think on 2 different layers at the same time, that is it. remember the "thinking mode" stuff last year. this is similar, it's just that a new layer has been added to the stack.

when you design a coding loop, the loop is similar to the 4th layer. The 1st is the actual conversation (input/output), the 2nd is the multistep reflection (thinking), the 3rd is the jspace, and the 4rth is the feedback loop.

Example:

Why did the US attack Iran?

-loop: look at the output, improve until i cancel

-output: because of oil

-thinking: "the user seems to be asking for a political analysis not the actual trigger"

-jspace: "oil, politics, religion, israel, trump...etc"