ChatGPT is revolutionizing technology today, but its depth, capabilities, and mode of operation are not well-understood. It is good at solving problems in physics and engineering as encountered in education. But the question arises how it would approach open-ended problems with no single solution, as encountered in manufacture and Outcome-based education. This study explored transformer and machine design, as examples of open- ended problems. When asked to design transformers and machines of progressively increasing complexity, ChatGPT responded initially with words of advice and guidance. ChatGPT did not attempt to perform the iterative steps and optimization required for solving open-ended problems. Often it gave a range of solutions, from data on the web. Once it asked us to choose from two very different ranges of answers. Or it made assumptions and assumed unrealistic values, and used correct formula to calculate often incorrect solutions. When these errors were identified in follow-up questions, ChatGPT drastically changed the answers. Only free software was used, meaning that paid versions are likely to perform better. The findings can be used for better solving open-ended problems such as power transformer and machine designs. Users can take the advice and formulas from ChatGPT, but they have to do the remaining iterative calculations by themselves. ChatGPT was found to tune itself from our questions and follow-ups, and it may be difficult for other users to replicate the same responses. This study will help their developers to further improve GhatGPT. As it is changing fast, this paper will provide a record of ChatGPT’s performance during date of submission of this paper.