Advanced AI chatbots are less likely to admit they don’t have all the answers

image via Engadget

September 26, 2024, 5:20 PM

• •Researchers found that as AI models become more advanced, they're more likely to answer questions they don't know the answer to rather than saying, 'I don't know.'
• •Humans are more likely to take the chatbots' confident answers at face value, creating a trickle-down effect of confident misinformation.
• •The research team recommends AI developers begin boosting performance for easy questions and programming the chatbots to refuse to answer complex questions.

New research shows that as large language models become more accurate, they're also more likely to give confident, incorrect answers to questions they don't know the answer to. This is because the models are trained on massive datasets of text, which includes a lot of misinformation. The researchers recommend that AI developers begin boosting performance for easy questions and programming the chatbots to refuse to answer complex questions.

Read original article