The Digital Press

OpenAI has admitted that their language models are trained to guess answers rather than admit when they don’t know something, leading to frequent AI "hallucinations" or false outputs.

Why it matters: AI models prioritize providing an answer, even if wrong, because uncertainty is penalized in evaluations.

The big picture: Training data often lacks patterns for rare facts, pushing models to guess instead of expressing ignorance.

The stakes: Rewarding guesswork encourages misinformation, complicating AI reliability and user trust.

Commenters say: Users express concern about AI teaching wrong answers, fearing superficial responses harming education and trust.