Unverified50% confidenceFactExact time
In the gated key-value memory circuit, the residual stream encoding of each label token encodes the label identity, and mid-layer attention heads write the category of the previous demonstration as a key.
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedPrompt injection can be divided into two types: direct injection and indirect injection, where indirect injection refers to attackers hiding malicious instructions in external content read by the agent.67% similarUnverifiedThe decoder adds two mechanisms on top of the encoder: masked self-attention and cross-attention; masked self-attention ensures that during generation only previously produced tokens can be seen, while cross-attention allows the decoder to query the encoder's output.67% similarUnverifiedThe decoder-only architecture uses a causal attention mask to ensure each token can only attend to its preceding context, making it suitable for autoregressive text generation tasks.67% similarUnverifiedEngram automatically captures memories via Claude Code's session hooks, with no need for users to explicitly say "remember this" — decisions land in the graph at session end.65% similarUnverifiedNoteBank's writing stage takes place during context encoding, before any question appears, selecting notes based on each token's key vector uniqueness (1 minus the cosine similarity between the token's key and the block's average key, then averaged over layers and heads), and cannot foresee what future questions will need.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1134244API
curl https://kongchang.com/api/v1/knowledge/claims/1134244MCP
get_claim(id=1134244)