Polylog
Sign inSubscribe
← The Polylog AI Intelligence Brief

Morning Edition · Wednesday, July 15, 2026Published at 1:32 AM EDT · New York

A Linear-Attention Method Claims Exact State Tracking Without a Growing KV Cache

Semidirect Fourier Delta Attention adds phase-controlled memory and chunked kernels to the recurrent-state approach that trades long-context recall for fixed memory.

Save
A Linear-Attention Method Claims Exact State Tracking Without a Growing KV Cache

A new paper introduces Semidirect Fourier Delta Attention (SFDA), a linear-attention design aimed at the central weakness of the approach. Linear attention replaces softmax attention's key-value cache, which grows with sequence length, with…

Continue the AI Intelligence Brief

Track frontier labs, chips, export controls, model releases, regulation, and AI infrastructure.

  • 5 AI intelligence signals a day
  • Frontier labs, compute, and chips
  • Model releases and AI infrastructure
  • Source-grounded analysis with confidence labels

The Global Intelligence Brief stays free.

Subscribe for $19/moAlready a member? Sign in

Part of a tracked trend

Context Length Becomes the Near-Term Model Frontier

Frontier labs increasingly treat context length, not raw learning ability, as the near-term axis of progress for in-context learning, with inference cost the binding constraint on realizing very long context; expect recurring pushes toward dramatically longer usable context windows.

More from this edition