I'm not sure how this distinguishes the process from a table of all possible input context windows(and rng state if you want temperature) with their desired o

I'm not sure how this distinguishes the process from a table of all possible input context windows(and rng state if you want temperature) with their desired output. This has been argued as a theoretical(but larger than the universe) model in philosophy for long enough that I wouldn't be surprised if it predates the notion of creating a machine intelligence. Without any mechanism for generalisation, it can be argued that it is simply a record of an intelligence with the actual mind being whatever created the table I guess you can think of Markov chains as a form of that table shrunk down to a hash table of the input storing only a set of the most probable outputs for any collisions. With the right semantic hashing you could consider it analogous to a LLM, but I feel there's a lot leaning on the term 'semantic hashing', because that's where the magic is in the LLM.