Topics — What a work is about. Nucleation holds whatever reading the owner files here, whether or not an Anthology of the SOTA topic could also hold it; instruction for ML practice belongs to the anthology (ADR-048). The words are seeded from the clusters the reading-time triage of 2026-09-25 found, and grow by decision.
Representation learning — how learned systems come to represent their data — self-supervised and contrastive objectives, spectral embeddings, information bottlenecks, knowledge-graph embeddings, and whether representations converge.
2 of 52 ARG documents. Back to the full index.
| # | Title | Summary | Status |
|---|---|---|---|
| ARG-008 v2 | A language model’s response to its context is functionally a parameterized frame, so context-conditioned models give a controlled setting in which to test the theory | The owner’s U17 analogy, which A43 enlarged into “the primary formal example”. Undercut by the assistant in its own outlines: a prompt is not a social situation, model behaviour need not match human inference, and judging with the generating model measures only its preferences. | Active |
| ARG-011 | On the crossed topic-and-act design, judgement factorizations that recover the act where a topic-only factorization fails would show that a factorization recovers only the structure in the relations it is fitted to | The planned evidence for the October thesis, CLAIM-129, is CASE-040, which crosses topic with act. Rejected for that design as filed: the act is fixed by whoever writes the items and the topic-only baseline is built to fail on the critical cells, so “a pass shows that the items were built as intended”. The record conceded the point on 2026-10-10 and repaired the thesis’s defeat condition. | Rejected |