It's almost as if Epistemology and epistemic chains are relevant to making truth claims! Being unable to make coherent epistemic derivations is true of MOST of

It's almost as if Epistemology and epistemic chains are relevant to making truth claims! Being unable to make coherent epistemic derivations is true of MOST of humanity, and unless rigorous epistemic processes are followed and explicated they will not be embedded into the foundational data. This is actually what the symbolic people get wrong IMO - you can't "code in" epistemic reasoning that humans will accept because it's a social not mathematical function. That is to say, the majority of humans accept as truth claims that have no consistent and complete epistemological grounding (Hi godel!). You can test this yourself. Go find a random group of college educated people and try and have a discussion on gettier problems[1] or the munchausen trilemma[2] to the point where everyone can coherently state the paradox. Don't try to resolve them (impossible), but rather see how incredibly hard it is just to get to the point of hitting the paradoxes even once they are familiar with the problem. I promise, unless people have sat with these problems for a long time in different contexts they won't be able to even conceptualize these epistemological problems. That might be fine if everyone building, testing, funding, using LLM or other AI systems is deeply aware of these epistemic problems but they aren't - not even close. So with modern data-driven learning, that leaves you with attempting to build systems which can derive epistemic chains from the existing corpus of data. But current data systems do not contain enough data from random internet users that would reveal coherent and repeatable epistemic chains for MOST problem classes. My guess is that there are some epistemic chains available for fairly trivial systems. So any process which attempts to build coherent reasoning chains, and their successors, which are "RL loops" or markov control process, will always fail if the foundational data they are sitting on does not have epistemic grounding and contextual framing. [1]https://en.wikipedia.org/wiki/Gettier_problem https://en.wikipedia.org/wiki/Gettier_problem [2]https://en.wikipedia.org/wiki/M%C3%BCnchhausen_trilemma https://en.wikipedia.org/wiki/M%C3%BCnchhausen_trilemma