On how formal verification and deep learning can complement one another, thinking about "intelligence," AI safety concerns, and groundedness.