Stanford, Apple, and MIT Studies Reveal Major AI Models Struggle to Differentiate Facts, Opinions, and Real Reasoning.

A recent study from Stanford University found that ChatGPT and other AI tools struggle to differentiate between facts and concepts. According to the study Language models fail to differentiate opinions from knowledge and fact, which was released in the journal Nature Machine Intelligence, all of the major AI chatbots were unable to consistently recognize when an opinion was untrue, raising their capacity to spread incorrect data or have visions.

Claude, ChatGPT, DeepSeek, and Gemini were among the 24 current language models that the researchers analyzed. 13,000 questions were presented to these AI models to test their capacity to differentiate between facts, knowledge, and opinions. The objective nature of knowledge and the fact that reality is a requirement for understanding aren't well recognized by most models. Every model that was evaluated was unable to identify incorrect opinions and statements; GPT-4o's accuracy decreased from 98.2% to 64.4%, while DeepSeek R1's performance dropped from over 90% to 14.4%.

The study specified that this defect has essential effects in areas where the difference is important, including legal matters, healthcare, or journalism, where combining opinion with knowledge can result in serious errors in judgment. According to a study released by Apple in June, modern AI models might not be as intelligent as they have been advertised. The tech giant suggested that reasoning models like O3-mini, DeepSeek-R1, and Claude have no capacity for reason. These models are generally very effective at remembering patterns, but they completely collapse when the questions are changed or become more complicated.

The study suggested that frontier Large Reasoning Models (LRMs) face complete collapse beyond certain difficulties through complete testing across various kinds of concerns. The Massachusetts Institute of Technology (MIT) research released in August found that 95% of businesses using AI systems experienced no return on their investment. The investment failed because AI models were more difficult to combine with a business's standard processes rather than because they were ineffective.