Why AI hallucinates with total confidence

追加: 2026-05-05

[00:00:00]AI doesn't read words like you and I do.

[00:00:02]It breaks everything down into small chunks called tokens. What you send is called input tokens. What it sends back is called output tokens. This is how AI companies charge you. It's basically a way of measuring usage.

[00:00:19]The context window is what an AI can see at any one time. The bigger the context window, the more the AI can remember.

[00:00:27]The more it can remember, the better its answers.

[00:00:30]Every model has a limit on how many tokens it can handle at once. Bigger context windows have more information, but also cost more money.

[00:00:42]You've probably seen AI say something completely wrong like it's a complete fact. That is what is called a hallucination. It's not lying on purpose. It's just predicting what sounds right based on patterns. But sometimes those patterns don't lead to somewhere that actually exists. But it doesn't know that, and it just says it with complete confidence.

[00:01:04]Here's your reminder just to always fact-check anything important.