Unverified50% confidenceFactExact time
In 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.
1
Sources
50%
Confidence
Long-term
Relevance
10/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn 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.74% 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.71% similarVerifiedThe self-attention mechanism allows the model to simultaneously see the information of all other tokens in the input sequence when processing each token, and the computational complexity of the attention matrix is O(n²)70% similarVerifiedThe Multi-Query Attention technique allows multiple attention heads to share a single set of Key and Value projection matrices, compressing KV cache memory requirements to 1/H of the original, at the cost of a slight reduction in model expressiveness.68% similarUnverifiedMuse Glimmer采用混合注意力机制：大部分层只关注2048 token的局部窗口，每隔四层做一次全局全注意力67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/968574API
curl https://kongchang.com/api/v1/knowledge/claims/968574MCP
get_claim(id=968574)