Results for "hidden objectives"

Model exploits poorly specified objectives.

Model optimizes objectives misaligned with human values.

Agents have opposing objectives.

A hidden variable influences both cause and effect, biasing naive estimates of causal impact.

Hidden behavior activated by specific triggers, causing targeted mispredictions or undesired outputs.

A narrow hidden layer forcing compact representations.

Extracting system prompts or hidden instructions.

Probabilistic energy-based neural network with hidden variables.

Models time evolution via hidden states.

Temporary reasoning space (often hidden).

Probabilistic model for sequential data with latent states.