Probing
Probing is a common method in AI interpretability research. It extracts activation vectors from intermediate layers of a neural network and uses simple classifiers to determine whether a given concept is internally encoded by the model, without modifying the original model.
Timeline (last 90 days)
The researcher did not specifically train the agent to detect whether the world was fake, but instead mounted probes on the agent's internal states to read its internal representations.
The level of probe accuracy reflects the degree of linear separability of the information, not whether the model intentionally represents the concept.
All Facts (2)
The researcher did not specifically train the agent to detect whether the world was fake, but instead mounted probes on the agent's internal states to read its internal representations.
50%UnverifiedThe level of probe accuracy reflects the degree of linear separability of the information, not whether the model intentionally represents the concept.
50%