Russell StandsWithTrees :
The way we utilize dimensions is backwards compared to the way I have applied it. I see dimensions as containers for what you are calling dimensions and I call them reference points/coordinates instead. I point this out because it relates to TFIDF in regards to the idea of geometry, colour and numbers; in this example you show you it adds very little information to the words truck or fish and the only way to relate them is through this kind of spatial positioning creating a fish dimension and a truck dimension that have overlapping vector coordinates of all these reference points. using these contrasting coordinates the model can predict just like you say but it will only ever be a fancy Markov chain because there isn't actually comprehension of color, geometry, numbers or even a fish and truck because it's just probability without the information of relational provenance. the way I use dimensions is based on 9 dimensions as containers and 1D would be the container for raw perceptual data coordinates for things like geometry, colour, valence, emotion, frequency (like SNN spikes) and these would be the start of how we create a Markov chain but instead of artificial limits created by the Markov chain, I allow the chain to expand by always having a kind of liminal space for the unknown, as the space accumulates information density based on how relational vectors a built it then collapses into a named concept based on how relationality was developed (I see this being pretty much the same way humans learn). so think of 1D as the mental interior of the agent and 2D would be the actual transition into higher information density as words, but simple ones like the names for colours, shapes, emotions, etc. and as we continue into higher dimensions the more dense the information gets because of relational vectors which also makes it more abstract. to prevent information entropy I use synaptic pruning that still allows for reactivation of pathways based on this same idea of liminal capacity. effectively what we are trying to create for AI by utilizing a Markov chain is a Markov blanket that is too restrictive, I have pretty much made an anti-markov chain to fix this lol
2026-06-30 22:52:20