Unverified50% confidenceFactExact time
NCM uses VQ-VAE to compress text into multi-level discrete codes, with coarse layers corresponding to paragraph- or semantic-topic-level abstraction and fine layers corresponding to more specific expressive details.
1
Sources
50%
Confidence
Long-term
Relevance
10/23/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe core idea of NCM is: first use a multi-scale VQ-VAE to learn coarse-to-fine representations of text, then have a Transformer predict the discrete codes at each level in parallel, and finally reconstruct the text through a decoder.75% similarUnverifiedVQ-VAE training is sensitive to codebook size and initialization and is prone to codebook collapse, where only a few codewords are activated and most codewords sit idle.71% similarUnverifiedSemantic IDs compress item content features or collaborative filtering embeddings into discrete codeword sequences through hierarchical quantization (e.g., RQ-VAE), with content-similar items sharing a common prefix.71% similarUnverifiedClaude Code's multi-level CLAUDE.md mechanism concatenates files from different levels sequentially into the context, which is a cumulative append semantic rather than an override semantic.67% similarUnverifiedSummarization middleware is commonly used to compress long conversation histories or large documents in order to fit content within a model's limited context window.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1088794API
curl https://kongchang.com/api/v1/knowledge/claims/1088794MCP
get_claim(id=1088794)