AI Models Struggle with Hallucinations, Researchers Propose Evaluation Updates
Summary
Large AI language models continue to hallucinate plausible but false statements, prompting researchers to propose updated evaluations that discourage confident guessing by penalizing confident errors more severely and rewarding expressed uncertainty.
Key Points
- Large language models like GPT-5 and chatbots like ChatGPT still hallucinate plausible but false statements despite improvements
- Current model evaluations encourage hallucinations by rewarding lucky guesses and not penalizing confident errors enough
- Researchers propose updating widely used evaluations to discourage guessing by penalizing confident errors more than uncertainty and giving partial credit for expressing uncertainty