Week 04 — Prompt Engineering

The prompt as a programming surface. The standard patterns, the diagnostic techniques for failing prompts, and the management practices that survive production.

Lecture

In-context learning (Brown et al. 2020) · zero-shot, few-shot, chain-of-thought (Wei et al. 2022) · ReAct (Yao et al. 2022) · tree-of-thoughts · structured output (JSON mode, function calling) · prompt-template management for production.

Read before the lecture

• Wei et al., *Chain-of-Thought Prompting Elicits Reasoning in Large Language Models* (NeurIPS 2022)

Recitation — paper discussion

Yao et al., *ReAct: Synergizing Reasoning and Acting in Language Models* (ICLR 2023) (paper)

Come ready to argue one side of each:

• What does ReAct buy that pure chain-of-thought doesn't?
• When does ReAct fail?

Reference text for this week: chapter 04 of the bilingual notes — EN PDF · FR PDF.