Alignment

Friendly Description: Alignment is the work of making sure AI does what we actually want, in ways that are safe, fair, and helpful. Think of it like training a puppy: a smart puppy can do all sorts of things, but it needs gentle guidance so it understands what's okay and what isn't. Alignment is the same idea for AI, making sure that as it gets more capable, it stays helpful and respectful of people.

Example: Suppose someone asks an AI to write a persuasive message to convince a friend to lend them money. A well-aligned AI might help draft a kind, honest note explaining why the loan is needed, but it would push back if asked to write something deceptive or manipulative. Alignment is what helps the AI understand the difference.