Don't Point the Finger at AI for Its Gibberish—There's a Fundamental Problem with Evaluation Criteria
2025-10-30 / Read about 0 minute
Author:小编   

When you come across ChatGPT spewing out incoherent nonsense, resist the urge to jump on the bandwagon and ridicule it right away; the root cause may well lie with humans. The latest research has shed light on the fact that AI hallucinations are not merely technical hiccups. Instead, they originate from systemic flaws entrenched in the training process. The prevailing mainstream AI evaluation criteria are riddled with significant problems: a correct guess garners 1 point, whereas both remaining silent and making an incorrect guess result in a score of 0 points.