Are you underfitting or overfitting?

I have college hires every year in my teams, which gives me the opportunity to observe how they grapple with their first real job in life. I find it fascinating to use the underfitting and overfitting concepts from machine leaning to understand their struggle. Underfitting In machine learning, underfitting means when a model can't capture the nuances of data due to its oversimplification. Similarly, some engineering students leave school with a very generalized understanding, without any ideas of how to apply these understanding to real world. I’ve had new hires who couldn’t ssh to a Linux host using CLI. These were students from top computer science schools. Whatever they have learned, they are under fitting for the real world we live in. Overfitting An overfitting model in machine learning has been overly trained on specific data, to the point it fails with new, unseen data. Drawing a parallel, some students optimize their learnings completely towards specific exams or popular technical interview practices like LeetCode challenges. This "exam-oriented" or "interview-oriented" learning encourages students to memorize specific, often arbitrary, results rather than holistic problem-solving. While they may excel in classroom tests or coding interviews, they falter when faced with real world problem solving. In machine learning, the quality and quantity of training data are both crucial. Similarly, an engineering curriculum that only touches the surface of a concept, without teaching students any concrete application of it, has data quantity problems that cause under-fitting. They need more data to train or train with better algorithms (mechanisms). When a course focuses too heavily on learning for exams, students are essentially being trained on "bad data." It is a data quality problem. The students may get their “A” in the training but what they have learned are far away from what they need in real jobs. They are overfitted for the wrong skills. A successful machine learning model balances underfitting and overfitting. Likewise, a good engineering education should interweave theoretical foundations with real-world application. Beyond textbook knowledge, internships, co-ops, and hands-on projects are invaluable "validation data." Emphasizing contextual learning over exam-centric approaches can better prepare students for real job and real life challenges. But are underfitting and overfitting only applicable to college graduates? Aren’t we all learners of life? Maybe every so often we should also ask: “Are we underfitted or overfitted for our job?”