Grading student work is a time-consuming task, especially in STEM fields like physics. What if AI could help? Researchers at the University of Central Florida explored using GPT-3.5 to grade student answers to a physics conceptual question. They didn't retrain the model; instead, they used a clever prompting technique called "scaffolded chain of thought." This approach provides GPT-3.5 with a detailed rubric and guides it to compare student answers to specific criteria step by step. The results were impressive. The AI grader, using scaffolded prompting, achieved 70-80% agreement with human graders – comparable to the level of agreement *between* two human graders. This suggests that with the right prompting, AI could potentially handle grading tasks with human-level accuracy, freeing up instructors' time for other important work. The next step is to expand this technique to other STEM fields, like engineering, where evaluating complex solutions to open-ended problems is a major part of education. While this initial study focused on a single physics problem, it opens the door for exciting possibilities in automated assessment, potentially transforming how educators evaluate and provide feedback to students. Further research will need to examine the reproducibility of these findings and investigate the effectiveness of this approach on a wider range of problems.