Unverified50% confidenceFactExact time
In Spotlight, memory is writable—the model itself decides token by token what to load and when to overwrite—and memory can store not only facts but also skills.
1
Sources
50%
Confidence
Long-term
Relevance
10/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn Spotlight, each token can read from and write to an unbounded memory; the model learns to index individual memory units, with each token touching only a very small portion of them at a time.74% similarUnverifiedMemory layering is typically divided into three layers—working memory, episodic memory, and knowledge memory—and its core value is enabling an Agent to retain information across longer time spans without exceeding context window limits.73% similarUnverifiedSpotlight splits the system into an intelligence module and memory: the intelligence module is fixed in size with weights that do not change as memory grows, while the memory stores knowledge, procedures, and working state.72% similarUnverifiedSpotlight replaces the attention mechanism with a memory mechanism that escapes the trade-off between memory capacity and access cost.72% similarUnverifiedIn the Transformer architecture, attention layers are responsible for routing and combining information, while feed-forward network layers tend to function as key-value memories that store factual knowledge.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/968577API
curl https://kongchang.com/api/v1/knowledge/claims/968577MCP
get_claim(id=968577)