twitter xp
the primary challenge on the road to epistemically-rigorous generative AI is an internal ontological scheme capable of encoding the provenance of training data & representing it in a single internally-consistent knowledge graph, which is fundamentally a data normalization problem
---
RT @pee_zombie
my life work is to build an information architecture system capable of representing, with maximal fidelity, the entire causality stack invol…
https://twitter.com/pee_zombie/status/1468803399995084804
twitter xp
it's entirely possible that rather than needing to explicitly design and build this into the model architecture, it may emerge organically during training within the model weights as a metabolic optimization, aka the "grokking" phenomenon
https://bounded-regret.ghost.io/future-ml-systems-will-be-qualitatively-different/
twitter xp
the most adaptive epistemics are those straddling the inflection point between the competing requirements of metabolic load & executive utility
---
RT @pee_zombie
much of human behavior is made legible when examined through the lens of materialist computationalism, namely the fact that information processing has a real physical cost in terms of work and metabolic energy required, & your biology resists any efforts it considered extrane…
https://twitter.com/pee_zombie/status/1360720737594732548
twitter xp
in fact, I'd suggest that any realworld agent naturally develops a fucntionally-equivalent internal structure in the process of world-modeling and prediction, as a metabolic optimization, given that such a structure is useful for optimizing actions given limited information