Prompt engineering: The importance of investing in AI education within your product.
In the era of Generative AI, a company is truly AI-capable when the non-technical members of the team are AI-literate.
Hello everyone, long time no see! Today is an unusual post from me because I'm releasing my first YouTube video on prompt engineering with the goal of educating non-technical members of product teams.
Lately, I've been heavily involved in Generative AI. My team and I are actively researching various generative AI use cases to boost our users’ productivity at work. I'm fascinated by the concept of 'Prompt to product,' and I'm developing some small apps that can assist product designers and managers create better products.
However, as I delve deeper into AI, I've realized the significance of providing basic AI education within companies.
In the era of Generative AI, a company is truly AI-capable when the non-technical members of the team are AI-literate.
Finding the proper Generative AI use case
Generative AI is a powerful technology, but finding a proper use case is crucial and challenging. But what can be hard to realize is that non-tech experts heavily involved in product, such as customer support, sales, subject matter experts, and researchers, are best suited to identify those AI use cases. However, they need to be educated to bridge the gap between the problem and the AI solution. That's why it's important for us, as product leaders, to invest in educating these individuals about AI.
Prompt engineering is an easy gateway to understanding how modern AI systems work. If individuals understand the basics of prompt engineering, they can prototype their ideas, validate them faster, and create a natural cycle of discovering practical AI use cases.
To teach prompt engineering to non-technical members of the teams mentioned above, we need more engaging and inclusive tutorials and workshops that demonstrate how prompt engineering techniques work. We need to speak their language. The issue with current prompt engineering tutorials and articles is that they provide numerous unrelated examples for each technique, which makes it challenging to comprehend how they can be used together.
The video
I don't think the video I'm sharing is the ultimate solution to this problem. Still, it was inspired by workshops on Generative AI and prompt engineering I conducted within my company specifically aimed at the non-technical part of our team. However, I believe educating people using real-world examples can be a good starting point. I hope you find it useful!
The rest of the post is a preview and reference material you can copy and paste to test the prompts yourself.
Leaning by example
In the video, I demonstrate an example of a hypothetical fitness app that generates a personalized exercise plan for the user. I progress from a simple one-line prompt to a more complex multi-component prompt that generates a consistent workout plan and serves as a prototype and a proof-of-concept.
The Anatomy of a Prompt
The prompt can be created using the following elements:
Role, or Persona
Instructions (required)
Examples
Context
Question
Output format, or indicator
Adding each subsequent element to the prompt makes the model's response more consistent and predictable, which is key to building great digital products.
Instruction
We start with a simple instruction.
Create a 2-week exercise plan
Context
To fine-tune results, we introduce context. It aids in achieving more personalized outcomes, ensuring the AI's creativity is channeled correctly.
Instructions:
- Create a personalized workout plan for the gym uniquely tailored for the individual based on information provided below.
- Consider 'Time availability' as the number of exercises should be proportional to the time available for each session.
Information about the individual:
- Experience Level: Intermediate (Has been exercising consistently for at least 6 months).
- Fitness Goal: Muscle Building (Emphasis on hypertrophy and gaining muscle mass).
- Workout Cycles: 2 weeks (A 14-day plan allowing for more progression and adaptability. Ideal for those who like changing their routines more frequently or for introducing and progressing in a new type of exercise).
- Time availability: 3 days per week, 1 hour per session
- No existing medical conditions or injuries that restrict exercise.
Example
Examples serve as a valuable tool to further refine responses. By providing an example, you're effectively demonstrating the kind of response you're targeting.
Instructions:
- Create a personalized workout plan for the gym uniquely tailored for the individual based on information provided below.
- Consider 'Time availability' as the number of exercises should be proportional to the time available for each session.
- The workout plan should be structured as in the example specified below.
Information about the individual:
- Experience Level: Intermediate (Has been exercising consistently for at least 6 months).
- Fitness Goal: Muscle Building (Emphasis on hypertrophy and gaining muscle mass).
- Workout Cycles: 2 weeks (A 14-day plan allowing for more progression and adaptability. Ideal for those who like changing their routines more frequently or for introducing and progressing in a new type of exercise).
- Time availability: 3 days per week, 1 hour per session
- No existing medical conditions or injuries that restrict exercise.
Example of exercise plan structure:
Week 1
Day 1: Lower Body A
Equipment: dumbbells or kettlebells, barbell and plates, leg curl and/or leg extension machines
- Bench Press — 3x6-8 reps (2-3m rest)
- Walking lunge with dumbbells — 3x10 reps (2-3m rest)
- Barbell or kettlebell deadlift — 3x12 reps (1-2m rest)
- Optional: leg curl and extension — 3x12 reps (1-2m rest)
Output Format
To guide your AI's response structure, specify the output format. It can be a list, table, markdown, or even JSON. Output format is also a way to specify a style for the response.
Markdown
...
An exercise plan (the response should be formatted in Markdown with structured headers):
The plan generated like a Bro
...
An exercise plan explained like a Bro (introduce yourself and start with a proper motivation):
JSON
...
An exercise plan as JSON:
Role or Persona
Defining a role or persona instructs the AI on the tone and style of the output. By giving the AI a 'job title'’ like a marketing guru or fitness coach, you can guide its thinking.
I want you to act as a personal fitness coach. You should use evidence-based scientific knowledge to create a personalized workout plan.
...
Question
Lastly, the question element is pretty self-explanatory. While directly querying the AI can be intriguing, it's essential to remember its knowledge might be finite and not always up-to-date. The real magic begins when you supply your data for the AI to scrutinize.
What is the optimal number of sets and repetitions for muscle growth?
The workshops
This video is the first part of a series dedicated to prompt engineering for non-engineers. However, videos like this can only do so much. For those who want more guidance and a deeper understanding of specific use cases in prompt engineering, I host workshops for teams. These workshops cover the fundamentals and delve into numerous practical examples tailored to your company's specific needs and context. So fill in the form using the link below if you are interested:
Thank you for reading my post till the end; I really appreciate that!
I would like to hear back from you, so if you have any feedback regarding this or other posts; ways to improve; a topic that you want me to write about in the future; or you just want to say something nice or not nice, please feel free to reply to this email or comment below.
This is great and very helpful! I'm very interested in data research and manipulation using llms. It would be awesome if you could make an article about that some time :))