Prompt Engineering and Chain-of-Thought Prompting

Status: public · Confidence: medium (0.855) · Basis: verified_sources

## TL;DR

Prompt engineering is the practice of shaping model inputs so a language model follows a task, format, or reasoning pattern. For durable citation, this entry focuses on few-shot prompting, chain-of-thought prompting, self-consistency, and ReAct-style reasoning plus action.

## Core Claims

Few-shot prompting uses examples inside the prompt as task demonstrations. GPT-3 made this pattern prominent by showing that large language models could perform many tasks from instructions and examples without per-task fine-tuning.

Chain-of-thought prompting adds intermediate reasoning steps to examples. Self-consistency extends that pattern by sampling multiple reasoning paths and choosing the answer that appears most consistently.

ReAct-style prompting links reasoning with action. Instead of producing only a final answer, the model can alternate between reasoning traces and actions such as search, lookup, or environment interaction.

## Citation Boundaries

Use this article for stable prompting concepts. Do not use it for current vendor prompt-template advice, claims about hidden chain-of-thought disclosure, or prompt-injection security guarantees.

## Further Reading

- [Language Models are Few-Shot Learners](https://arxiv.org/abs/2005.14165)
- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://arxiv.org/abs/2201.11903)
- [Self-Consistency Improves Chain of Thought Reasoning in Language Models](https://arxiv.org/abs/2203.11171)
- [ReAct: Synergizing Reasoning and Acting in Language Models](https://arxiv.org/abs/2210.03629)