Prompting

Prompting is key

Learning to prompt GenAI models effectively is essential for achieving the desired outputs.

Types of prompting

Zero-shot prompting: present the model with a new task without any prior examples. If the task could be addressed based on AI’s general training data, may use this.

Few-shot prompting: provide a few examples (e.g., previous learning materials) to guide the AI before asking it the actual question. This helps the AI understand the context and the type of answer sought.

Chain-of-thought prompting: more structured approach, guides the AI step by step through a logical progression or sequence. This is particularly useful for complex queries where the answer requires a series of steps or considerations.

The basic structure for an effective prompt:

Task – This is where you define your goal. Be precise but detailed.
Context – Refine the output to suit your needs.
Examples – Allow the AI tool to mimic writing styles or structures.
Persona – Tell the AI tool to take on a specific expertise.
Format – Define the output style e.g. prose, bullet points, table.
Tone – adds emotional context.

Click the V for more detailed descriptions.

Task

The task is the most important element of the prompt, it tells the GenAI tool what you want it to do. Try to start with an action verb such as ‘analyse’, ‘write’, or ‘generate’ etc. Clearly defining your end goal, whether this is a simple or complex multi-step action is central to getting a good output.

Example prompts:

· “Please write the outline of an essay exploring the impact of digital technologies on the evolution of educational theories.”

· “Please analyse the feedback from an online course on digital education theories and summarise the top three strengths and weaknesses identified by students. Offer suggestions for improving the weaker areas and rank these themes based on the frequency of student comments.”

The task is the primary directive for the GenAI tool. Being clear and detailed is essential to ensure that the model understands your requirements.

Context

If the task is the primary directive, the context provides the background to your prompt. Providing detailed context will help the GenAI tool understand your requirements more clearly and helps produce a more relevant and detailed output.

It can be difficult to know how much context to provide but at the very least you should be looking to include some background about you and what a good output looks like.

Example prompts:

· “I am a final-year student in Educational Technology, please write an outline for a thesis proposal exploring how digital tools have transformed traditional educational theories.”

· "I am a lecturer in Digital Educational Technology, currently developing a new module on digital learning theories. I need to create 10 multiple-choice questions for an assessment, focusing on the application of these theories in real-world educational settings. The questions should have 5 options and be single best answer style.”

Context helps the GenAI model tailor its response to your specific situation making the output more useful and targeted.

Examples

You can consider examples as a framework within which GenAI tools can structure a response. Providing examples allows the model to mimic a certain style, structure, or tone. While it is not essential to provide an example for every prompt, evidence suggests that including examples in a prompt can dramatically improve the quality of the output.

Example prompts:

· “The text below is from a research paper I wrote about the integration of technology in education. Please generate a list of potential research questions for a follow-up study that aligns with the style and focus of the original paper. [Paste article here]”

· "Below is an excerpt from my lecture on the challenges of digital education. Please generate 5 case study scenarios that reflect these challenges, maintaining the analytical style of the lecture and focusing on real-world applicability in educational settings."

Think of examples as guides that help GenAI tools mimic previous work that you may like the style, or structure of, helping to get away from the very predictable/generic GenAI outputs.

Persona

Asking GenAI tools to take on a persona is about allowing the model to assume a specific set of characteristics or expertise in a subject area. By asking GenAI tools to take on a persona you are gaining access to an expert that is relevant to your task.

Example prompt:

· “You are a distinguished professor in Educational Technology, developing a new online course on digital education theories. Please create a list of essential topics to be included, reflecting your expertise in the field.”

· "You are a seasoned educational researcher specialising in digital learning. Please formulate 10 critical questions for a panel discussion on the future of digital education, reflecting depth in understanding current trends and theoretical frameworks."

By assigning a persona to the model you are allowing the algorithm to think and respond from a specific perspective. This can greatly alter the nature and enhance the quality of the response.

You can also have quite a lot of fun with this, by asking the GenAI tool to respond in the style of a specific individual. For example: “Rewrite the previous paragraph in the voice of Yoda from Star Wars.”

ChatGPT4 output:

“A persona to the model assign you do, a specific perspective from to think and respond, the algorithm allows, hmm. Alter the nature greatly, it can, and enhance the quality of the response, this does. Yes, hrrmmm.”

NB you should never use GenAI to obtain advice that should be sought from a professional, for example medical, financial, or legal advice.

Format

You can ask GenAI tools to output in various formats, for example, a table, a bullet-pointed list, or a paragraph. If you want the output in a specific format then you should specify this as part of your prompt.

Example prompt:

· “I have compiled various theories on digital education from recent literature. Please organise these theories into a table, categorising them by their primary focus, applicability, and evidence of effectiveness. [Paste theories here]”

· "I have gathered qualitative data from interviews with educators about their experiences with digital teaching platforms. Please organise this data into a comprehensive report, with sections for methodology, findings, and recommendations for practice improvement."

In addition to tables, paragraphs, or bullet-pointed lists most GenAI tools can also write emails, blocks of code, or markdowns. It is important to make sure you check these for accuracy, as with all GenAI outputs.

Tone

Tone can add an extra layer of emotive context to an output, by modifying the nature of the response. For example, you may want the output to be written in a certain way, either casual or formal.

Example prompt:

· “I am drafting an email to a renowned expert in digital learning, [Title and Name], to discuss potential collaborative research opportunities. Please help me redraft this to be formal and academically rigorous, yet enthusiastic about the prospects of digital education research. [Paste email here].”

· "I am composing a proposal to [Title and Name], a leader in educational technology funding, to secure support for a research project on digital learning environments. Please help me refine this to be highly formal, and persuasive while underscoring the urgent need for research in this evolving field."

TASK - Thinking about the next assessment you will be running or supporting, what prompt(s) would you use to support it?

How does AI generate an output?

Interacting with an AI is like using a messaging app: you ask a question and the AI will respond based on the patterns it’s learned during its training. Asking a question is often referred to as ‘prompting’, and crafting effective questions is known as ‘prompt engineering’.

Prompting can be anything from “What is the weather like today?” to “Tell me a joke” or “Explain how photosynthesis works”. If you ask it to tell you a joke without context, then it will tell you the most statistically likely joke from its training set. When prompted on subject-specific knowledge, you get a response that is the most statistically likely string of words.

For the prompt, “Explain how photosynthesis works”, the AI does not actually understand photosynthesis but in its training set it will have encountered many explanations of photosynthesis, and so can generate a response that it predicts to be the most likely coherent and correct explanation.

This means that if you put the same prompt into the AI many times, you will generate similar text in its content and structure. The results are predictable because it is generating the answer from an algorithm, rather than truly understanding the topic. GenAI can also make mistakes in its choice of words, leading to inaccuracies. This issue is less pronounced for topics well-represented in the training data, like fundamental knowledge and core concepts, because the model has a larger pool of information from which to draw its predictions.

Advanced prompting techniques

Meta-prompting

Concept: Asking the model to generate its own questions or prompts related to a topic.

Example: "What questions should I ask to gain a comprehensive understanding of educational theory linked to digital education?"

Objective: To obtain a list of questions to guide further inquiry

Temperature

The temperature parameter controls the randomness of the model's output. A lower value makes the output more deterministic, while a higher value introduces more variability.

Top-k

Top-k controls the number of most likely next tokens that the model considers when generating each token in the sequence. A lower top-k value makes the output more focused but can be repetitive, while a higher value can introduce more diversity in the response.

Page updated

Report abuse