54. NLP vs ChatGPT: Classical NLP and Generative AI
54.2 Decision Checklist
Why. Choose tools with a simple checklist, not hype.
What. Prefer classical NLP when you have labels, stable categories, need cheap bulk scoring, and want coef_/feature stories. Prefer Gen AI when text is open-ended, you need drafting/summaries, or labels are scarce but quality prompts + human review work.
How.
Decision checklist
- Do you already have hundreds+ of trusted labels? → classical baseline first.
- Do you need new sentences written (reply drafts)? → Gen AI.
- Is data too sensitive for an external API? → classical or private hosting.
- Still start with a TF-IDF baseline even if you later add LLMs.
def pick_lane(n_labels, needs_draft, sensitive):
if needs_draft:
return "gen_ai (+ human edit)"
if sensitive and n_labels >= 200:
return "classical_nlp_local"
if n_labels >= 200:
return "classical_nlp_baseline_then_maybe_gen_ai"
return "gen_ai_or_rules_with_human_review"
print(pick_lane(800, False, True))
print(pick_lane(20, True, False))