Latent Lattice Language Models: adding a horizontal computation axis to the residual stream
HanGuo97 · x · 2026-11-20
Shivam Duggal proposes Latent Lattice Language Models (LLLMs): instead of processing each token only through a single vertical residual stream (reading/writing memory and applying MLPs layer by layer), the approach augments it with a horizontal axis of computation—a lattice of latents that interact with each other.
Framed as a handshake between RNNs and Transformers, the idea has drawn interest from researchers who note that how latents should flow and interact remains largely unexplored—an architecture-level direction worth watching.
Related event: MIT and Adobe Propose Latent Lattice Language Models(3 posts)→
More from Research
- Open-source cancer model from JHU and Stanford hits every tumor across 34 public CTs — MaziyarPanahi · 2026-11-20
- The core dilemma of agentic benchmarks: web access gives realism but kills replicability — davidmanheim · 2026-11-20
- New 79-minute review maps deep learning for gene regulation across cellular contexts, from maxATAC to Bolero — anshulkundaje · 2026-11-20
- Agent infers optics geometry from 48 probes, then tunes a few parameters instead of 100k+ pixels — GordonWetzstein · 2026-11-20
- Same harness trains optical encoders; paradigm generalizes to any controllable physical system — GordonWetzstein · 2026-11-20
- Gordon Wetzstein: coding agent in the loop boosts optical NN accuracy 35% to 96.8% — GordonWetzstein · 2026-11-20