Results for "intent mismatch"

Ensuring AI systems pursue intended human goals.

A mismatch between training and deployment data distributions that can degrade model performance.

A high-priority instruction layer setting overarching behavior constraints for a chat model.

Differences between training and inference conditions.

Train/test environment mismatch.

Control shared between human and agent.

Inferring human goals from behavior.

Controlling robots via language.