Percepta unveils Spotlight architecture: intelligence and memory separated, memory grows with no access cost
Recoil42 · reddit · 2026-10-03
Percepta announced Spotlight, a new architecture that decouples intelligence from memory so knowledge and skills can grow without changing model weights.
- Claimed first architecture with unbounded memory growth at constant access cost: every token reads/writes an unbounded memory but touches only a few indexed cells; unlike MoE's fixed sparsity, Spotlight is arbitrarily sparse
- A fixed-size intelligence module computes, while a writable memory holds facts, procedures, and working state — the model decides token-by-token what to load and overwrite
- Because memory can store skills as well as facts, the model can gain capabilities without retraining
- Detailed in two company blog posts
Related event: Percepta Unveils Spotlight Architecture Decoupling Intelligence from Memory(2 posts)→
More from Infra
- Flamegraph CPU profiling called the most efficient way to debug Flink performance issues — DanielLockyer · 2026-10-03
- New handbook explains speculative decoding: why acceptance rate α=0.8 still yields only 1.87× speedup — techNmak · 2026-10-03
- CUDA 13.4 Explained: New Tile Features, PTX Instructions and Rubin sm_107 Preview — NVIDIA Developer · 2026-10-03
- Running 262K context on a 5090+4070 rig: three Strata patches hit 130 tok/s decode — Fz1zz · 2026-10-03
- Self-hosting AI doesn't save money, and this developer does it anyway — basnijholt · 2026-10-03
- azapi_data_plane_resource Closes Terraform's Last Manual Gap for Foundry Hosted Agents — WirelessLife · 2026-10-03