5 Prompt engineering
Prompt engineering is crucial in AI, particularly with advanced language models like GPT-4o in Microsoft Copilot. It entails crafting precise prompts to guide models toward producing the desired results. By refining prompts, you can improve the quality, relevance, and accuracy of AI responses.
In the following video, you explore the concept of prompt engineering. You discover the importance of clear instructions and relevant context to effectively communicate with AI and achieve optimal results.
Download this video clip.Video player: Video 5
Transcript: Video 5
Voice-over:
The way we interact with AI has transformed.
We have moved beyond simple commands and queries to a more sophisticated method of communication known as prompt engineering.
Prompt engineering is the art of crafting prompts that guides an LLM-based generative AI to produce a desired response.
A prompt is the input you give to a generative AI experience, like Microsoft Copilot, telling it what you want it to do and how.
It consists of two main components: Instruction and context. The instruction is the written part of the prompt that states the task and the objective.
It should be clear and specific, so Copilot knows exactly what you expect from it.
The context provides information for the response such as the intended audience and the desired tone.
It should be relevant and appropriate, so it can tailor its response to your needs.
Once a prompt is submitted, Copilot reads text in chunks called tokens, which can be as short as a single character or a word.
This is important because every model has a limit on the amount of text it can process at once.
This is why it is important to keep our prompts concise and to the point.
Let’s look at this using Microsoft Copilot.
Step 1: Understanding the Prompt. A prompt is your direct line of communication with Copilot. It is how you instruct it to assist you in your tasks. A well-constructed prompt consists of two essential elements: the instruction and the context. Instruction: This is what you’re asking Copilot to do.It should be concise and explicit.
For instance, Draft an email to a client,’ or ‘Generate a project management plan.'’Context: This gives Copilot the necessary background to customise its response to your needs.
It includes details like the target audience, desired tone, level of detail and any specific guidelines or limitations.
Step 2: Crafting Your Prompt. When you are prepared to engage with Copilot, launch the webpage and select the appropriate style for your task.
‘Please create a comprehensive, professional guide for adult learners on how to effectively network in their industry, with a focus on digital platforms.’ Breaking down the example. Instruction: ‘Create a comprehensive, professional guide’. Context: ‘for adult learners on how to effectively network in their industry, with a focus on digital platforms’ Step 3: Submitting the Prompt.
After crafting your prompt, submit it.
Copilot will then dissect the instruction and context into tokens and produce a response grounded in its interpretation. Tokens are the segments of text that the model uses to comprehend and formulate responses. They can be complete words or fragments of words.
For example, ‘network’ is one token, while ‘networking’ may be divided into ‘network’ and ‘ing’ based on the model’s tokenisation method.
Step 4: Reviewing the response or output.
Once Copilot delivers its output, examine it to verify it aligns with your objectives. If it does not quite hit the mark, you can refine your prompt and resubmit it to achieve a more precise outcome. This process helps you collaborate more effectively over time.
As we advance in LLM-based generative AI, the art of prompt engineering becomes increasingly important. It is not just about asking, but how we ask it.
By understanding and effectively using prompts, we can harness the full potential of generative AI experiences like Microsoft Copilot.
The key to a successful interaction lies in the clarity of your instruction and the relevance of your context.
Video 5
Interactive feature not available in single page view (see it in standard view).
Activity 1 Improve your prompt creation
Timing: Allow approximately 5 minutes
The following video teaches you how to improve your prompt creation abilities to maximise your effectiveness with Microsoft Copilot. It includes best practices such as being specific, selecting the appropriate style, prioritising accuracy, and tailoring your experience to make each interaction more productive and enjoyable.
What is one simple tip for writing a better question to an AI?
Download this video clip.Video player: Video 6
Transcript: Video 6
Voice-over:
Refining your prompt crafting skills can enhance your effectiveness with Microsoft Copilot.
It all starts with a clear objective. Then followed by providing the necessary context, setting expectations for the desired response and following best practices.
So, you might be wondering, what are the best practices when writing a prompt? First, be specific.
The more details you offer, the better Copilot can tailor its service to your needs.
For instance, if you are requesting an image of a dragon, don’t just type ‘Create a dragon’ but describe its colour, size and the environment it inhabits.
Is it perched atop a mountain at sunset? Each detail contributes to a more accurate and vivid creation.
Choose your style. Copilot offers three distinct styles to cater to your needs: ‘more creative’, ‘more precise’, and ‘more balanced’.
If you are writing a science fiction story, opt for ‘more creative’. For factual information, ‘more precise’ will serve you well. And for a mix of both, ‘more balanced’ offers the perfect harmony.
Prioritise accuracy.
When precision is important make sure to guide Copilot with clear and concise prompts.
Whether it is summarising a specific article or explaining a complex concept, the right prompt ensures an accurate response.
Customise your experience.
Copilot is versatile, catering to both novices and experts alike.
Frame your questions to suit your level of understanding, from simple explanations to tackling advanced topics.
Switching topics.
If you are changing the subject, simply select ‘New Topic’ in Copilot. This feature gives Copilot a new start, keeping your conversations focused and free from confusion. 25Control the length.
You have the power to set the length of Copilot’s responses. If you require more information, just prompt Copilot to elaborate.
Embrace variety.
Copilot is not limited to text.
It can present information in tables, outlines, and even diagrams, enriching your experience with a visual dimension that goes beyond mere words.
Let’s see how this works. Let’s say you want to send out an email to your team about a project update.
You are running short on time and could really use some assistance with crafting a draft email.
So, you open Microsoft Copilot to help you with this. You decide to select the ‘precise’ style because you want Copilot to be factual and precise. You also want the email to be shorter with simpler sentences that avoid unnecessary details. You then craft a clear and specific prompt.
‘Compose a concise, professional and appreciative email to my team. The email should celebrate our completion of the design phase, announce the transition to the development phase of our mobile app project and encourage continued hard work. Please keep it to the point.’ In just a few seconds, you can have a draft email ready to send to your team.
This not only saves you time during a busy day, but also allows you to focus on other important tasks.
Remember, effective prompting is rooted in the clarity and specificity of your questions as well as the thoughtful construction of your requests.
With these best practices, you will harness the full potential of Microsoft Copilot, making every interaction more productive and enjoyable.
Video 6
Interactive feature not available in single page view (see it in standard view).
Interactive feature not available in single page view (see it in standard view).
Discussion
The more details you give, the better Copilot can tailor its response to your needs. Rather than making vague requests (prompts), you should give clear and detailed instructions about what you are looking for. For example, instead of simply asking it to: ‘create a dragon’, you ought to describe its colour, size, environment, and other particular characteristics to get a more accurate and helpful response. The main point is being specific and detailed leads to better AI responses.
Note
Microsoft Copilot no longer includes the conversation style feature.
Prompt engineering is essential for working with reasoning engines. Here’s how to make your experience more effective:
- Be specific: Clearly state your needs.
- Use examples: Guide the model by providing examples.
- Set the context: Offer enough background and instructions for clarity.
- Iterate and refine: Adjust prompts based on responses for improvement.
- Limit the scope: Break down tasks to enhance output quality.
- Test approaches: Experiment with different phrasings to find what works best.
With these best practices for prompting in your toolkit, the next section dives into the importance of critical thinking in this process.