Content-based addressing replaces positional stretching for long context

Content-Based Addressing for Long Context (arXiv 2609.07314, Wed 9 Sep digest) replaces RoPE's growing positional counter: the token stream is divided into units, ordinary RoPE positions are retained within each unit, and every completed unit gets an address computed from its content. Adding units applies the same learned map to new content rather than extending a positional range or identifier table. The authors prove the construction preserves local RoPE exactly and leaves attention over non-local content unchanged in form.

Every rescaling/interpolation scheme attacks the positional train-test mismatch by stretching a counter; this removes the counter for macro structure and lets content identity do the addressing. Early and small-scale, but a genuinely different primitive for the 1M-context problem — DeepSeek V4.1-Flash's CSA2 takes the sparse-indexing route to the same end (ev-20260910-01).