54. NLP vs ChatGPT: Classical NLP and Generative AI
54.3 Carry-Forward Habits
Why. Students who only know ChatGPT skip metrics and leakage; students who only know TF-IDF fear LLMs. Bridge both.
What. Classical skills (split, metrics, leakage) still matter when you evaluate LLM labels. Gen AI course skills (prompts, RAG, embeddings) sit on top — they do not erase ch31/ch37 habits.
How.
Carry-forward habits
- Hold out a human-labelled test set even if an LLM helps label train data.
- Compare LLM judgements to a classical baseline on the same test slice.
- Log prompts/versions like you log model versions (MLOps lite later).
print("Habit: same test reviews scored by TF-IDF model AND by LLM judge → compare")
Where to go next for LLMs
Live Gen AI course: what-is-an-llm, embeddings, RAG, API first calls. DS stays the classical foundation.