Results for "modality alignment"

Maintaining alignment under new conditions.

Ensuring model behavior matches human goals, norms, and constraints, including reducing harmful or deceptive outputs.

Aligns transcripts with audio timestamps.

Ensuring AI systems pursue intended human goals.

Correctly specifying goals.

Ensuring learned behavior matches intended objective.

Model behaves well during training but not deployment.

Tradeoff between safety and performance.

Research ensuring AI remains safe.