In this Jam Session, Karin shares an early conference-talk outline for the International Society for the 2026 International Society for Neo-Platonic Studies Annual Meeting , proposing a deliberately “geometric” way of narrating what LLMs are doing: not primarily symbol manipulation, but the continuous transformation of meanings in high-dimensional embedding space. The talk braids that computational story to older cosmological motifs—sacred alphabets, creative letters, and word-as-constituent power—then grounds it with an accessible bridge (RGB color space) before moving into word embeddings, semantic neighborhoods, and why “meaning” becomes navigable once it is spatialized. The centerpiece is the highly debated “Platonic representation hypothesis”: the claim that, as models scale, different modalities and languages seem to converge on structurally alignable internal representations, suggesting a strange modern echo of “forms,” while also raising the suspicion that models may simply be formalizing the imprints of human embodiment and culture (a quite contest proposal).
The second half opens into a wide team discussion: how this might reshape Khora’s visual language (gradients, constellations, motion as meaning), whether embedding spaces open language toward the ineffable or further entrench us in a self-referential net, and what “objectivity” could mean when representations are learned from human corpora. The group also probes limits: datafication vs. mathematics, intuition vs. syntax (with music as a test case), randomness and unpredictability, and the “model collapse” problem as evidence that human language carries a generative depth AI text may imitate but not reproduce. It’s a Jam that functions as both research synthesis and design provocation, ending with practical next steps for refining the talk and aligning it with Khora’s evolving metaphysics of knowledge.
