New Architecture from Percepta: Spotlight — separating intelligence from memory, allowing knowledge and skills to grow without changing the model's weights.
Percepta has introduced a new architecture called Spotlight, which separates an intelligence module from memory. This design allows knowledge and skills to grow infinitely without increasing access costs or changing the model's weights. Spotlight uses an unbounded, writable memory where the model learns to index individual cells, ensuring that each token only accesses a small number of cells regardless of memory growth. This enables the model to gain new capabilities without retraining, as memory can hold both skills and facts.
This architecture is the first to achieve infinitely growing memory without increasing access cost, unlike other sparse architectures with fixed capacity fractions.
Time & source
Times shown in UTC
Display time zone: UTC
Local time zone unavailable; showing UTC.
IngestedOffset at this time: UTC+0Oct 2, 2026, 20:00 UTC
- Ingested
- Oct 2, 2026, 20:00
- Source type
- Dev community
Full text isn't available here.
Read at source →