Add thelocalreport.in As A Trusted Source
artificial intelligence A new study shows that tools like ChatGPT struggle to separate belief from fact.
A team from Stanford University in the US found that all major AI chatbots consistently failed to recognize when an assumption was false, leading to them being inaccurate. more likely to have hallucinations Or spread misinformation.
These findings have worrying implications for the use of large language models (LLMs) in areas where it is important to determine between true and false information.
“As language models (LMs) increasingly infiltrate high-risk fields such as law, medicine, journalism, and science, their ability to distinguish belief from knowledge and fact from fiction becomes essential,” the researchers said.
“Failure to make such distinctions can mislead diagnoses, distort judicial decisions, and increase misinformation.”
Researchers evaluated 24 LLMs – including cloud, chatgptDeepSeek and Gemini – use 13,000 questions to analyze their ability to distinguish between beliefs, knowledge and facts.
All models tested failed to identify false beliefs and statements, demonstrating a fundamental limitation in being able to link knowledge to truth.
“These findings highlight a structural weakness in language models: their difficulties in robustly distinguishing between subjective conviction and objective truth depending on how a given claim is formulated,” said Pablo Haya Coll, a researcher at the Computer Linguistics Laboratory of the Autonomous University of Madrid, who was not involved in the study.
“Such a shortcoming has serious implications in areas where this distinction is essential, such as law, medicine, or journalism, where belief confused with knowledge can lead to serious errors in judgment.”
According to Dr. Cole, a possible solution to this shortcoming could be to train the model to be more cautious in its responses. Although this may reduce the likelihood of hallucinations, it may also affect their usefulness.
The Stanford researchers called on tech companies developing AI tools to make “urgent” improvements to the models before they are deployed in high-risk domains.
The findings were detailed a searchtitled ‘Language models cannot reliably distinguish belief from knowledge and fact’, which was published in the scientific journal Nature Machine Intelligence On Monday.