Emma

Summary:

  • Current general-purpose AI systems can have both beneficial and harmful capabilities, posing extreme risks.
  • Dangerous capability evaluations and alignment evaluations are essential for responsible AI development, deployment, and security.

Tags:

Research