The emergence of ChatGPT and competing software like Bard and Copilot is revolutionizing education today, but their functioning and capabilities are largely unknown. University level problems in physics and engineering can be solved, but the nature and limitations of this capability of ChatGPT are worthy of investigation, as attempted in this study. The author used his three-decades long experience in universities to analyze implications of ChatGPT in education. Differentiation, integration, matrices, Fourier and Laplace transforms could be easily solved, but questions are still misinterpreted, giving wrong answers, as seen in this study. The challenge to students was to better formulate the query as text, understandable to ChatGPT. Entering phasor angles, such as for AC circuits, was problematic. Figures are not recognized, meaning it is difficult to input a series parallel or multi-loop circuit. The freely used version of ChatGPT was investigated, and undergraduate engineering students were surveyed. The findings of this study can be used by developers to improve ChatGPT. Teachers can adapt to the software, and homework may now be assigned in forms more difficult to enter into ChatGPT. ChatGPT should not be interpreted as a problem for teachers assigning homework, nor should we fear the atrophy of the intellect of students. Rather, just like the calculator, simulations and Google Search changed assignments over the last few decades, ChatGPT should be taken as a tool for making education and homework more sophisticated than ever. Education can move towards assigning open-ended and complex-engineering problems, which form the basis of today’s Outcome-Based education. A new generation of students will approach problems far more effectively than their predecessors. This study provided some insight and appreciation for what lies in the next few years for education in the light of such AI problem-solving softwares.