Gated DeltaNet generalizes ~10× beyond training lengths on StateBench's hardest level
allen_ai · x · 2026-11-04
Closing the StateBench thread: a flexible Gated DeltaNet handled sequences 10× longer than those seen in training on the hardest level. The team also proves that Gated DeltaNet with non-negative eigenvalues can solve some periodic tasks under fixed-precision arithmetic with rounding.
More from Research
- STOC submissions include dueling P=NP and P!=NP proofs; blogger bets both wrong — thegautamkamath · 2026-11-04
- DeepSeek V4.1 Flash reads raw weights: 3 of 4 RL checkpoints identified correctly — poppear · 2026-11-04
- CoworkOS Debuts: Three-Layer Agent Coworker Framework Lifts Qwen3.6 by 9.76 Points — KyeGomezB · 2026-11-04
- STOC 2027 draws 1665 submissions and publishes all of them, with author names, immediately — thegautamkamath · 2026-11-04
- DOLORES: Training-Free Meta-Scaffold Lifts 8B Models Past 32B Baselines — StellaLisy · 2026-11-04
- Formalizing PTX in Lean to formally verify GPU kernels written by AI agents — gharik · 2026-11-04