Even Best AI Models Fooled By Nonsense Sentences

A new study reveals that even advanced AI chatbots, powered by large language models, can still struggle to distinguish between natural language and nonsense sentences. Researchers at Columbia University conducted experiments where people rated the naturalness of sentence pairs, and while more sophisticated AI models performed better than simpler ones, all models occasionally rated nonsensical sentences as natural.

This suggests that AI chatbots may not fully grasp how humans process language. The study’s findings raise questions about relying on AI systems for important decisions and also offer potential insights into human brain functioning by comparing AI language processing to human cognition. The research was published in Nature Machine Intelligence on September 14, 2023.