Recursion is more powerful than attention, RLMs express compaction easily
lateinteraction · x · 2026-08-16
The author discusses the distinction between recursion and attention, noting that recursion (RLMs) is actually more powerful, though the analogy remains useful. RLMs can easily express compaction in a way that is difficult for recurrence mechanisms. The reverse simulation (recursion by compaction) requires heavy assumptions.
More from Research
- Advanced Matrix Factorization Algorithms: From Sparse Coding to Phase Transitions — IgorCarron · 2026-08-16
- Researcher laments: paper progress narrowing, reading enthusiasm wanes — sarahookr · 2026-08-16
- Cordis Framework Deep Dive: Achieving High Cache Hit Rates for Long-Horizon Tasks — EstablishmentOdd785 · 2026-08-16
- NVIDIA Releases SWE Instruction-Tuning Dataset for Coding Agents — burny_tech · 2026-08-16
- Algebra+Category Theory Solve Transformer State-Tracking Puzzle — burny_tech · 2026-08-16
- SpeechMap adds page to visualize model vocabulary frequency over time — cephaloform · 2026-08-16