Glossary

What happens when a language model stops answering and starts doing.

The open problem of making sure an AI pursues the goal you actually meant, not a technically correct version of it.

The question nobody has settled, and everyone has an opinion on.

The (imperfect) tool trying to spot what AI wrote.

Not yet law anywhere. Not purely science fiction either. The question our institutions are least prepared for.

The difference between a tool that replaces you and one that makes you formidable.

The hypothesis that AI could hand us a century of medical progress in a single decade.

The difference between an AI that answers your questions and one that can walk into a room

The heartbeat of how models learn — small steps, always toward fewer mistakes.

The scientific effort to open the black box.

Why efficiency rarely saves as much as it promises — and creates far more than expected.

What most people mean when they say ‘AI’ these days.

The economic error that makes every new technology look like a job-killer.

The decision loop at the heart of how agents — and the best human decision-makers — stay ahead.

The horizon where human management becomes optional.

When a model memorizes its training data instead of learning from it.

A map of all possible outcomes, weighted by how likely each one is.

The discipline built into training to stop a model from trying too hard.

The fear that technology destroys jobs permanently — predicted confidently before every major wave.

The research finding that haunted education for four decades.

The moment AI gets so capable that the future stops being readable from the present.

The basic unit of text that language models read, think in, and count.

The friction that determines when hiring someone beats doing it yourself.

Whether a model’s confidence is something you can actually trust.