Control / Research term
Representation engineering
An umbrella term for methods that analyze or intervene on a model's internal representations to study or influence behavior.
The label covers probes, steering vectors, representation reading, and related techniques. Some methods observe activations; others modify them. An internal direction may correlate with a concept without being a complete or causal representation of it, so monitoring and control claims need separate evidence.
Builder example
These methods offer another experimental surface for model developers with activation access. They do not automatically detect or correct problematic reasoning, and most hosted-product builders cannot inspect the required internals.
Common confusion: Prompt engineering and representation engineering sound similar, but they operate at completely different levels. Prompt engineering changes what the model reads as input. Representation engineering changes how the model processes that input internally. They complement each other and apply to different classes of problems.