Ethan Mollick shares examples about what GPT-5 gets right and what it just decides it's going to do without being asked. Despite its advancements, GPT-5 still requires human oversight to correct errors and validate results, emphasizing the need for collaborative interaction with AI.
