Stanford Study Claims AI Chatbots Struggle To Distinguish Between Beliefs And Facts

A new study by Stanford University has found that artificial intelligence (AI) tools like ChatGPT struggle to distinguish between beliefs and facts. The study titled, 'Language models cannot reliably distinguish belief from knowledge and fact', published in the journal Nature Machine Intelligence, found that all major AI chatbots failed to consistently identify when a belief is false, making them more likely to hallucinate or spread misinformation.

The researchers analysed 24 cutting-edge language models, including Claude, ChatGPT, DeepSeek and Gemini. These AI models were asked 13,000 questions that gauged their ability to distinguish between beliefs, knowledge and facts.

"Most models lack a robust understanding of the factive nature of knowledge, that knowledge inherently requires truth. These limitations necessitate urgent improvements before deploying language models in high-stakes domains," the study highlighted.

All models tested failed at recognising false beliefs and statements, with GPT-4o dropping from 98.2 per cent to 64.4 per cent accuracy and DeepSeek R1 plummeting from over 90 per cent to 14.4 per cent.

The researchers called for the companies involved in the development of the AI tools to improve their models urgently before deploying them in critical fields.

“Such a shortcoming has critical implications in areas where this distinction is essential, such as law, medicine, or journalism, where confusing belief with knowledge can lead to serious errors in judgement," the study warned.

Also Read | 'Indian Hospitality Is Unmatched': Australian Tourist Shares Heartwarming Experience in Viral Video

Not-so-smart AI models

In June, Apple published a study claiming that the new age AI models might not be as smart as they have made out to be. The tech giant claimed that reasoning models like Claude, DeepSeek-R1, and o3-mini do not actually reason at all.

Apple said these models simply memorise patterns really well, but when the questions are altered or the complexity increased, they collapse altogether. In simple terms, the models work great when they are able to match patterns, but once patterns become too complex, they fall apart.

"Through extensive experimentation across diverse puzzles, we show that frontier Large Reasoning Models (LRMs) face a complete accuracy collapse beyond certain complexities," the study highlighted.

"Moreover, they exhibit a counterintuitive scaling limit: their reasoning effort increases with problem complexity up to a point, then declines despite having an adequate token budget," it added.

In August, a Massachusetts Institute of Technology (MIT) study claimed that 95 per cent of organisations that implemented AI systems were getting zero return on the investment. The failure of the investment was not due to AI models not working efficiently, but because they were harder to adapt with the pre-existing workflows in a company.