Interaction and Risk Terms

The second half of the examinable glossary: the terms that describe how you interact with an AI system, and the term that describes its most consequential failure.

Interaction

Chatbot. An AI system designed to simulate human conversation through text or speech. Modern chatbots based on LLMs can understand context, maintain dialogue history, and generate coherent, context-aware responses. In RE, chatbots can assist by transcribing stakeholder interviews, retrieving information from documents, or supporting documentation tasks through interactive dialogue.

Prompt. The input provided to an LLM to guide its output. It may be a question, an instruction, or partial text. In RE practice, prompts determine the relevance and accuracy of generated outputs — for example when asking an LLM to propose alternative formulations of a requirement, or to extract functional requirements from stakeholder interviews. The quality of the prompt has a direct impact on the usefulness of the response.

System prompt / instruction. A special type of prompt that sets the overarching behaviour or role of the LLM — for example, "You are an assistant Requirements Engineer…". This ensures generated outputs follow organisational style guides or RE best practices.

The distinction between a prompt and a system prompt is examinable and easy to blur. A prompt is a request. A system prompt is a standing instruction that shapes how all subsequent requests are handled. In module 3's terms, the system prompt is one of the sources of the system context.

Prompt engineering. The practice of designing and refining prompts to reliably achieve desired outputs from an LLM. The syllabus offers a comparison a Requirements Engineer will find natural: this is akin to crafting precise interview questions or template structures in RE, where clarity and structure improve the quality of the information obtained.

That analogy is worth taking seriously rather than treating as decoration. A Requirements Engineer already knows that a badly framed interview question produces a useless answer, and that the fault lies with the question. Prompt engineering is the same skill applied to a different respondent.

Techniques

Few-shot / zero-shot learning. Techniques in which the model performs a task with few examples, or with none, provided in the prompt. In RE these can be used to classify requirements, generate acceptance criteria, or rephrase specifications without extensive task-specific training data.

Chain-of-thought reasoning. An approach in which the LLM is prompted to explain its reasoning step by step before providing the final answer. In RE this can support transparency and traceability — for example when analysing conflicting requirements or evaluating design alternatives.

Keep module 2's caveat attached to that definition. Chain-of-thought improves results on tasks that benefit from structure because the model has seen many step-by-step explanations in its training data. The steps are generated text, not a record of an internal reasoning process.

Failure

Hallucination. A hallucination occurs when an LLM produces statistically likely text that is factually incorrect or unsupported by source data. It is a direct consequence of the probabilistic mechanism, not a malfunction.

In RE, hallucinations are particularly problematic if fabricated details are mistaken for stakeholder statements or domain facts. Requirements Engineers using LLMs must therefore validate AI-generated content against authoritative sources, in line with CPRE principle 6: non-validated requirements are useless.

The syllabus's closing instruction on the topic is the sentence to carry out of this entire course: all AI-generated output must be treated as an unverified draft.