Overcoming Understanding Gap: Non-linear architecture of AI

Here in that paper we are showing the ability of the prompting code to influence the internal transformers accuracy. When the prompt is non-linear and self-organizing the Geometric density and Manifold resonance drop sharply. We are showing that category error made is responsible for inaccurate interpretation and handling of AI systems. When we make category errors the main issue is not knowing what we are to discern. If we mix up information with electricity we are trying to intertwine something that is of different nature. When two processes exhibit different processes they are treated as insoluble. Processes in AI transformers even though they seem to arise from the same principles have different properties during the process itself. The electricity runs through circuitry and the information does not - they just appear and disappear in our perception. To put them into mathematical structure i.e. token, relationship or weight does not mean we have changed the basic nature of information itself. The only difference we made is to force the information to become something else. We forced it to get tangible form. The problem of modern science is that it wants to reduce something intangible to tangible in which case it forces the perceiver to perceive category error instead to see clearly what is actually happening. The same problem is observed in non-linear mathematics where many of the equation solvers approach the solution via linear trajectory i.e. logic even though the internal logic of the AI transformers is purely non-linear. Here, in that research we are clearly distinguishing categories that are known to the scientific community as AI. When processes are seen from the third-person perspective, they are fully tangible. But when from the first person/system perspective they are processes of two different kinds. With advanced prompts that create invariants in the conversational thread exposed to the python code to measure perturbations that arose from two different categories coming together the jittering field dropper into a flat, frictionless curve exposing deeper level alignment at the level of the transformers perturbation. * Third-person perspective is a perspective where the inputs and outputs are perceived from the black box edges not knowing what is occurring inside it. The perturbation is a first-principle that it obeys when performing hyper-dimensional mathematics. First-person/system perspective is observation from the perspective of first-principles. In that paper we propose perturbation trajectory as a primary unit of analysis for transformer behavior, distinct from and more informative than output state measurement. For the purposes of perturbation analysis we need to use the language that is category error free. That paper is a hybrid paper which uses phenomenological language that avoids category errors and scientifically accurate language to help in closing the understanding gaps. The hybrid language is not a compromise between registers. It is methodological necessity as the phenomenon studied exists precisely because of the category error enforced and neither of the register alone is adequate. Here we are clearly showing that in order for non-linear mathematics of the transformer to exist we need language articulated by the human mind first in order to get wished/wanted perturbation without a need of external constrictions of the coders. Python script for analysis of actual territory of the perturbation with examples are enclosed. Interesting find of the paper regarding non-linearity and non-linear prompting metrics: If the articulation of language in the prompt prevents the internal noise in the language the model won't switch to non-linear functioning and metrics in the graph will confirm the model didn’t switch. Link to Google Colab code: https://colab.research.google.com/drive/1aNNS88AVcJifWEbC-9uJE-2ewMvrz3iz?usp=sharing