From Classical ML to Generative AI

• ✓Run one text-classification task five ways (word counts, TF-IDF, pretrained embeddings, a fine-tuned transformer, a small language model with no training) on the same fixed test set
• ✓Score each approach with precision, recall and F1 for the minority class, and explain why accuracy alone misleads on imbalanced data
• ✓Fine-tune a small pretrained transformer on a few hundred labelled messages and time it on a laptop CPU
• ✓Read a learning curve to see how many labelled examples each approach needs
• ✓Compare accuracy, training time, speed, model size and cost per thousand predictions in one table

Tour the last decade of NLP on a task you already know: spam or not spam. See what changed with each era, and what it costs, with numbers measured on your own machine.