Realtime input GPT transformers

Due to classical computation, most methods biased today take chunks of input of information, and act on it to produce output information. They do not take continuous realtime input data to produce realtime output data, as intelligent beings, such as humans do, as we continually process data. Even though if computers work on clocks, and have discretization in computing for precise numerical computation, current models rely on input chunk of data, and not continually in realtime are capable of realtime data, such as audio directly from microphone fed to a transformer producing realtime outputs, or realtime video stream plugged into transformer for realtime output of any sort. Imagine it will immediately respond the the input change, and any sort of realtime scenario which is well tackled by humans will be tackled. For that in transformer, when its computing, it should continually update the observation that adds as a context for further computation of state that it continually acts on. Transformers act as learnable state machine representations that given the state and the observation hardwire the state output (i.e. current state evolved into something that depends also on observation), that the next state then goes through that as input. Well, model is trained for such a set of data describing all possibilities, then it learns the way in which patterns emerge, or are evolved.

Soon LLM’s will be hardcoded into hardware (directly etched), highly configurable updatable LLM dedicated hardware, and would act as a storage of knowledgebase, that is undisputed, these are deterministic finite state machines, but just with enormous complexity compared to human level of thinking, but not all things might have been captured in training data yet, neither all possibilities of interpretation of a thing, so human mind also still exist as the thinking tools that is propagated by nature.