Langford explains FBT theory: nextlat's fixed-point convergence and para-RNN parallel training
JohnCLangford · x · 2026-09-04
Langford explains why nextlat was unnecessary for FBT: the nextlat paper's proof implies convergence to a sequence of fixed points, and after reading the para-RNN paper it was clear how to efficiently train a parallel traversal of that set of fixed points. A technical footnote to his Full Bandwidth Transformer / recursion research line.
More from Research
- Fei-Fei Li's World Labs unveils Atlas world model: 3 photos replace 300 for 3D capture — theworldlabs · 2026-09-05
- Late interaction takes the VLDB VecDB stage: TACHIOM results show it scales efficiently — lateinteraction · 2026-09-05
- Elicit tops biomedical search benchmark with 60.3% recall, 20 points ahead of rivals — elicitorg · 2026-09-05
- LLMs fail basic probability coherence: P(rain)=0.7 but P(no rain)=0.2 — Moh1tAgarwal · 2026-09-05
- Cohere Labs tests 690K+ tools for agents — only 2.6% can complete a job task alone — cohere · 2026-09-04
- Tabula Sapiens 2.0 atlas maps 1.1M+ cells across 28 human tissues — StephenQuake · 2026-09-04