Prompt Engineering for Business Users

[The image above is generated by Midjourney. The prompt I used to create the image is listed at the end of this email.]

ChatGPT has revolutionized our interactions with artificial intelligence. Its ability to generate human-like text has myriad applications, from drafting emails to creating web content and even coding. Ironically, the code needed to interact with AI is no longer Python or another programming language but your natural language. However, to harness the full potential of ChatGPT, it's crucial to understand how to use it effectively. And even though there is natural language, there’s still a lot of nuance in maximizing the results of an interaction with chatbots like ChatGPT.

Understanding Prompts

Prompts are the starting point of any interaction with ChatGPT. They guide the AI's responses and can significantly influence the result. Simple and precise prompts often yield the most “creative” results. For instance, instead of asking, "Tell me something," you could ask, "Tell me about the history of the Roman Empire." Defining a role for the AI model, such as "You are a historian," can also guide the AI's thought process and result in more contextually appropriate responses.

ChatGPT system prompts are instructions or guidelines provided to the AI model to guide its behavior during a conversation. They are typically used to set the context, tone, or specific requirements for the AI's responses.

For example, a system prompt might instruct the AI to behave like a business analyst, a medical professional, or a language tutor. It could also provide specific instructions about the AI's knowledge cutoff, current date, or any tools or plugins it should use.

In the context of OpenAI's GPT-3 and GPT-4 models, system prompts guide the AI's behavior during a conversation, providing it with a "role" to play or specific instructions to follow. They are crucial to the AI's ability to generate relevant, context-appropriate responses. They might be the length of the output, or they may instruct the chatbot to keep asking questions until they get enough information to complete a task.

Few-Shot and Many-Shot Learning

The capability of ChatGPT to grasp from minimal examples, termed few-shot learning, is a potent asset, particularly in the business world. For instance, if you aim to create a listing of marketing and design services, you could offer examples such as "Graphic Design, suited for web advertising" and "Landing Page Creation, designed to maximize conversions," and then request ChatGPT to extend the list.

On the other hand, many-shot learning involves providing many examples to guide the AI's responses. This can be useful for more complex tasks, such as generating a service listing in a specific style. By providing multiple examples of the desired style, ChatGPT can generate a more accurate output.

Here’s an example of how a zero-shot, few-shot, and many-shot prompt might look for a Fiverr gig :

Zero-Shot Prompt

Create a compelling description for a Fiverr gig offering graphic design services.

Few-Shot Prompt

Create a compelling description for a Fiverr gig offering logo design services. Response: "Welcome to my logo design gig! With over 5 years of experience in graphic design, I specialize in creating unique and impactful logos that represent your brand's identity. I offer unlimited revisions and guarantee 100% satisfaction. Let's bring your vision to life!""Create a compelling description for a Fiverr gig offering social media management services." Response: "Need a social media guru? Look no further! In this gig, I offer comprehensive social media management services. From content creation to analytics, I'll help your brand grow its online presence and engage with its audience effectively. Let's boost your social media game together!"Now, the prompt: "Create a compelling description for a Fiverr gig offering graphic design services."

Many-Shot Prompt

"Create a compelling description for a Fiverr gig offering logo design services." Response: "Welcome to my logo design gig! With over 5 years of experience in graphic design, I specialize in creating unique and impactful logos that represent your brand's identity. I offer unlimited revisions and guarantee 100% satisfaction. Let's bring your vision to life!""Create a compelling description for a Fiverr gig offering social media management services." Response: "Need a social media guru? Look no further! In this gig, I offer comprehensive social media management services. From content creation to analytics, I'll help your brand grow its online presence and engage with its audience effectively. Let's boost your social media game together!""Create a compelling description for a Fiverr gig offering SEO services." Response: "Boost your website's ranking with my SEO services gig! I offer on-page and off-page optimization, keyword research, and SEO audits. With my services, your website will climb the search engine ranks and gain more organic traffic. Let's optimize your online presence!""Create a compelling description for a Fiverr gig offering content writing services." Response: "Welcome to my content writing gig! I specialize in creating engaging and SEO-friendly content for your blog, website, or social media platforms. I offer well-researched, plagiarism-free content that resonates with your audience. Let's tell your brand's story together!"Now, the prompt: "Create a compelling description for a Fiverr gig offering graphic design services."

Refining Outputs

Refining outputs with ChatGPT is an iterative process. For instance, if you're using ChatGPT to draft an email, you might start with a simple prompt like "Draft an email about the project update.” You could cut and paste in meeting notes or use a plugin like WebPilot or Linkreader to read a Google Doc with meeting notes via its sharing link. After the initial output, you can converse with the AI, asking it to add more details about specific project milestones or to clarify points about the project timeline.

For example:

Prompt from User: "Add more details about the software development milestone." ChatGPT: "In the software development milestone, our team successfully completed the design phase and moved into coding. We've implemented robust algorithms for features X and Y, and initial testing shows promising results."

Overcoming Limitations with Plugins

While ChatGPT is a powerful tool, it's essential to understand its limitations. For instance, it doesn't have access to real-time information and can sometimes generate plausible but incorrect outputs. However, these limitations can be bypassed by using plugins like Link Reader and Wolfram.

For example, if you need real-time data for a financial report, you can use the Wolfram plugin to fetch the latest stock prices or economic indices. You can use the Link Reader plugin to fetch and summarize the content if you need information from a specific webpage.

Additional Uses of ChatGPT

One use for ChatGPT is to change the tone or sentiment for more complex situations. I used ChatGPT recently to help resolve a conflict with a company I made a purchase with recently, and it not only got a good result but reduced the cognitive load of thinking through the frustrating situation. Here’s an exaggeration to illustrate the point of how you could deescalate a work situation by rewriting an email written in anger.

As an editor, ChatGPT can help refine and edit text for grammar, style, and coherence. For instance, you can ask ChatGPT to proofread a paragraph or write in the tone of Walt Mossberg, which I highlighted in an earlier newsletter on Large Language Models.

Code Interpreter

Earlier, I mentioned that ChatGPT replaced the need to use programming code to interact with the AI. This is true, but as of July 2023, OpenAI's Code Interpreter is a powerful tool integrated into the ChatGPT Plus platform. This official plugin allows users to run Python code directly within the chat interface, with the added capability of handling file uploads.

One of the critical features of the Code Interpreter is its ability to interact with files uploaded by the user. This allows for a wide range of applications, from analyzing data and creating charts to editing files and performing complex mathematical operations.

The Code Interpreter has found significant use in digital marketing, particularly in search engine optimization (SEO). Professionals in the field have leveraged its capabilities to analyze search engine algorithms and user behavior, generate data-driven insights for content optimization, and boost site rankings with tailor-made strategies. It has been used to analyze exportable marketing data, identify campaign performance metrics, and interact with Google Search Console data.

Conclusion

Mastering ChatGPT involves understanding how to use prompts effectively, leveraging few-shot and many-shot learning, refining outputs, and overcoming limitations. By understanding these tips and tricks and leveraging the power of plugins, you can overcome the limitations of ChatGPT and unlock its full potential. Remember, the key to success with ChatGPT is exploration and experimentation. Happy prompting!

Tip of the Week: Creating Visualizations with ChatGPT Code Interpreter

OpenAI's Code Interpreter is a revolutionary tool integrated into the ChatGPT Plus platform, offering a new level of interaction with artificial intelligence. This plugin allows users to run Python code directly within the chat interface, opening up a world of possibilities for data analysis, file editing, image conversions, and more. It's like having a Python-savvy assistant at your fingertips, ready to execute code on demand.

In the example below, I downloaded some sample data from Kaggle. Kaggle is an online community of data scientists and machine learning practitioners. It was founded in 2010 and later acquired by Google in 2017. Kaggle is best known for hosting machine learning competitions where participants can compete to create the best models for a given dataset. In addition to competitions, Kaggle provides users with a platform to find and publish datasets on various topics. These datasets can be used for personal projects, research, or for use in competitions.

First, you must use ChatGPT Plus and enable beta settings in your profile under Settings and Beta, then Beta Features and move the slider for Code Interpreter.

Then you will choose GPT-4 and check the Code Interpreter beta to use the plugin.

This will add a circle Send a Message with a plus sign in the prompt box at the bottom. This is how you upload the data you want to analyze.

Here’s the prompt and the output that shows the result. Note that ChatGPT Code Interpreter manipulates some of the data.

This is a simple example, but you can generate the graph and continue to manipulate it or even ask ChatGPT to summarize data with more simplistic prompts to get results you may not have anticipated, including different types of graph types, without having to re-upload data.

What I Read this Week

• Lost in the Middle: How Language Models Use Long Contexts - Research Paper via

• AI and the Frontier Paradox - Sequoia via

• The Generative AI Battle Has a Fundamental Flaw - Wired

• Meta’s Open Source Llama Upsets the AI Horse Race - Wired

• Microsoft will charge you $30 per month to use AI tools with Office —on top of what you already pay - Fortune

• OpenAI Code Interpreter — What you need to know - Dennis Layton on Medium

• The Power of ChatGPT Code Interpreter: Creating Data Visualizations Without Writing a Single Line of Code - Kanaries

What I Listened to this Week

AI Tools I am Evaluating

• MindOS - Create Autonomous AI Agents for your professional tasks.

• TypeDesk - Create text shortcuts that work across all your apps and websites for all your repetitive text content! Forget manual copy & paste; ditch repetitive typing!

• Wope - Find out what's working and what's not to get more search traffic. Like an SEO consultant who can analyze millions of data

• VizGPT - Make contextual data visualization with Chat Interface from tabular datasets.

Midjourney Prompt for Newsletter Header Image

For every issue of the Artificially Intelligent Enterprise, I include the MIdjourney prompt I used to create this edition.