Overfitting my meat-brain

I’m scheduled to take the AWS Certified AI Practitioner exam at 1 p.m. tomorrow.

Up until May, I had only used ChatGPT and Copilot sporadically, claiming “AI” was the devil and “real programmers” coded by hand. (I was such a Luddite.)

I didn’t know the difference between linear regression and clasificación. In May, I was lucky enough to get a graduate internship at my current company that exposed me to large language models and model tuning.

My handwritten notes on classification metrics, drawn as a two-by-two grid. True positive: predicted fraud, was fraud. False positive: predicted fraud, was clean. True negative: predicted clean, was clean. False negative: missed fraud, circled and labeled WORST.

My first class in my master’s degree is DTSC 540, Introduction to AI, and the concepts run parallel to the AWS Certified AI Practitioner certification.

I am learning so much about AI, machine learning, deep learning, generative AI, and their historical foundations. I feel I finally found a concept I am interested in enough that it will keep my attention through the three years of a full master’s degree. (This is my fifth time starting a master’s program. The first four times, I opted out with graduate certificates.)

On the other side of that is feeling overwhelmed—so much new terminology and so many new concepts to try to wrap my head around. When I first started my graduate internship, I made the mistake of trying to read AI Engineering by Chip Huyen, which is definitely not a book for beginners.

Tomorrow’s date is an arbitrary deadline, and I could easily push it back and study more, but I want to at least try to pass by overfitting my meat-brain. (Overfitting is when an ML model memorizes its training data instead of learning patterns that generalize.)

My handwritten notes on a Kobo e-reader. Fit model: training data plus training algorithm equals model. Model.fit(X, y). Overfitting equals memorizing the test. Evaluate: run the model on test data. Accuracy is less than precision, because chasing rare data gives false accuracy.

I’m going to take the exam tomorrow, let the chips fall where they may, and adjust fire from there.

My handwritten notes on drift. Data drift equals input changes. Concept drift equals environment changes. Below that, in capital letters: RETRAIN ON FRESH DATA!!!

If I have to keep studying and take it again, so be it. Either way, I know I will continue to be exposed to these concepts as I continue this crazy journey learning about generative AI in its nascent form.

Written on September 29, 2026