Important

AI Safety & Bias

What can go wrong, and why. · 5 min read

Bias in, bias out

Models inherit the biases of their training data. Image generators trained mostly on Western media skew Western. Hiring models trained on past hires repeat past discrimination.

Privacy & data leakage

Anything you paste into a public model may be used to improve it. Don't paste customer PII, proprietary code, or trade secrets into tools you don't control.

Human in the loop

For decisions with real consequences (hiring, lending, medical, legal), AI should assist, not decide. Always have a human review the output before it affects someone's life.