If every project is funded, is every non-funded thing a non-project?. ChatGPT answers "Not necessarily" and hallucinates an explanation. Here, ChatGPT fails to use a modus tollens inference rule and to answer “yes” to the question. This is a tip of the iceberg of shortcomings the state-of-the-art AI systems, such as ChatGPT and Gemini, exhibit when reasoning with natural language. This is largely because systems are not trained on how to carry out correct reasoning but on how to guess correct answers. The project will change this paradigm by introducing an explicit reasoning process in the training and evaluation.