Session 7: Prompt Design — Zero-shot vs Few-shot

Day 3 · Prompt Design & Iteration (3-1)

Warning🚧 Being prepared

This page is being finalized ahead of the course (Aug 3–7, 2026) and may be incomplete or change before your session. The syllabus and readings are ready now.

Learning objectives

By the end of this session, you will be able to:

  • Name the main classes of prompting strategies — instruction structuring · in-context learning · thought generation · decomposition · ensembling · self-criticism — and say which ones we run this week.
  • Locate the components of a prompt — directive · examples · output formatting · style instructions · role · additional information — in a real prompt.
  • Distinguish zero-shot from few-shot prompting and judge when examples help.
  • Explain chain-of-thought prompting and which tasks it improves.
  • Explain the train/test contamination trap and why you tune on a validation set but report on a held-out test set.

The message of the day: a prompt has parts you can design, and small changes to those parts change the score.

Agenda

  1. A map of prompting strategies — the many named techniques group into a few strategy classes, each answering a different question about the model call: instruction structuring · in-context learning · thought generation · ensembling · self-criticism · decomposition.
  2. Instruction structuring — the six components of a prompt (directive · examples · output formatting · style instructions · role · additional information), read off Huang & Mizumoto’s (2025) writing-feedback prompt.
  3. In-context learning and thought generation — zero-shot vs. few-shot, what examples actually teach, and the results table from Kim & Lu (2024), including where prompting alone stops improving and fine-tuning takes over. Then chain-of-thought, and the Example Prompt 1 → Example Prompt 2 refinement.
  4. Three strategy classes to know about, not run — ensembling (self-consistency), self-criticism (and why not to use an LLM to check your own labels), decomposition (prompt chaining, RAG).
  5. Two errors to avoid before you tune — train/test contamination, and tuning on validation while reporting on test.

The prompt components return in Session 8, mapped onto the CEFR classification task, and again in Session 9 as the error-analysis iteration loop.

Reading

Read (see all course readings):

  • Huang, J., & Mizumoto, A. (2025). Prompt engineering: Enhancing AI-driven language learning and feedback. In L. McCallum & D. Tafazoli (Eds.), The Palgrave Encyclopedia of Computer-Assisted Language Learning (pp. 1–8). Springer Nature Switzerland. https://doi.org/10.1007/978-3-031-51447-0_103-1
  • Kim, M., & Lu, X. (2024). Exploring the potential of using ChatGPT for rhetorical move-step analysis: The impact of prompt refinement, few-shot learning, and fine-tuning. Journal of English for Academic Purposes, 71, 101422. https://doi.org/10.1016/j.jeap.2024.101422

Running example (Day 3). We anchor the whole day in Huang & Mizumoto’s worked prompts — Example Prompt 1 (a generic chain-of-thought paragraph-feedback prompt) refined into Example Prompt 2 (the structured Task / Criteria / “My paragraph” prompt). That same structure returns in Session 8 (mapped onto the CEFR classification prompt) and in Session 9 (as the error-analysis iteration loop). A self-study map of the wider technique landscape: the Prompt Engineering Guide.

Slides & Colab