Mind · Strategy
NoiseInjection
Breaking local optima
Definition
Mechanism
If the Agent detects it is looping or repeating text, it deliberately injects high-temperature Noise or a random oblique Strategy Card into its Context to force a trajectory change.
Preconditions
- Clean signal/process
- Need for exploration/robustness
Postconditions
- Perturbed output
Invariants
Risks
Failure modes
Derailment into nonsense
the injection is large enough that the Signal does not return.
False positive
legitimate repetition is read as a loop, and a working trajectory is perturbed.
Relationships
Builds on
References
Autonomous Goal-Directed Entity
Infrastructure
Structured capability advertisement: agent_id, endpoint, protocols, capabilities, metadata
Infrastructure
Portable execution environment with inherited constraints
Physics
Irrelevant information
Infrastructure
Raw information emission
Adaptive high-level planning
Nothing in this library builds on this pattern yet.
Reference
Parameters
temperatureRatio, (0.0, 1.0]- Magnitude of the injected noise, as a share of full randomness.
detection_thresholdRatio, (0.0, 1.0)- Similarity between successive outputs that counts as looping.
cooldownDuration, unspecified- Minimum interval between injections, so noise does not become the trajectory.