xLSTM and Naju revive LSTM-style retention and writing gates
TheTuringPost · x · 2026-07-29
This post points to an explanation of xLSTM and a Korean model called Naju, which revives a classic LSTM idea: separating the control of retention and writing.
The key idea is to use two distinct gates inside a modern state-space model compatible with parallel scans:
- one gate controls how much of the existing state survives
- another controls how strongly new information is written
The linked article explains xLSTM as an extended LSTM with exponential gating and matrix memory, positioning it as a modern attempt to make recurrence competitive again against Transformers. It also claims xLSTM performs especially well on time-series prediction, citing Sepp Hochreiter’s assessment that xLSTMTime compares favorably with state-of-the-art transformer-based and other recent time-series models.
Related event: Korean Team Introduces Naju Model with Dual-Gate Mechanism(2 posts)→
More from Research
- Tsinghua professor's ApexIntelligence raises $50M to build self-evolving foundation models — DeryaTR_ · 2026-10-16
- Fluids Don't Compute—They Contribute the Nonlinear Step, Sandberg Argues — anderssandberg · 2026-10-16
- Anders Sandberg on fluid computation: error budgets tighten quadratically over time — anderssandberg · 2026-10-16
- How the fluid computation works: force field advects particle to 0/1 boxes — anderssandberg · 2026-10-16
- FlashAttention creator Tri Dao joins Radical Numerics as scientific advisor — exnx · 2026-10-16
- SAIR Launches Open Math OS, Offering Free Compute and 100+ Tools for Community-Owned Math AI — Thom_Wolf · 2026-10-16