Mind · Memory
PinkElephantContext#5131
Anything placed in context stays active, even when told to avoid it
Mechanism
A language model conditions on everything in its context window. Naming a thing to avoid, or showing previous outputs to not repeat, keeps that content present and pulls generation toward it. Steering therefore works by controlling what enters the context, not by adding negative instructions.
Invariants
- Content in context influences output regardless of the instruction attached to it
- Removing content is the reliable way to stop its influence
Failure modes
- Telling the model not to repeat earlier examples that are still visible produces near-duplicates
- Long conversations accumulate content that silently biases later turns
Used by
Identity
The name is the hash of the definition. Change a word and it becomes a different pattern.
- Sema ID
sema:PinkElephantContext#mh:SHA-256:5131b6e6652a91b151c3605d0bce9dbc3d763c3c2cc4b803435dc83ec9ff45e4- Reference
PinkElephantContext#5131- Ring
- 1
- Tier
- 2