Why ChatGPT lies to your face — confidently #shorts #ai #chatgpt

Añadido: 2026-05-06

[00:00:00]Now the uncomfortable part.

[00:00:02]Ask ChatGPT about a niche fact, a recent event, or an obscure [music] person, and it may give you a detailed, confident, completely wrong answer.

[00:00:12]Fake citations, invented [music] quotes, papers that don't exist. This is called hallucination, and it's not a bug. It's baked into how the model works.

[00:00:22]Remember, the model isn't retrieving facts. [music] It's predicting probabilities over tokens.

[00:00:28]When you ask [music] something well-represented in the training data, the correct token dominates the distribution.

[00:00:35]But when you ask something rare, obscure, or [music] recent, no single token dominates.

[00:00:41]The model picks a plausible-sounding answer.

[00:00:44]It has no in