Back

A rapid primer on crafting more effective GPT-4 prompts

Insight

01 Jun 2023

-

Giorgio Tarraf

It is increasingly important to understand how to effectively communicate with, and draw useful responses from, Large Language Models (LLMs) like GPT-4. This primer covers principles for effective prompt-writing, and for leading conversations in meaningful directions.

By following the simple strategies outlined below, you'll be well-equipped to communicate effectively with GPT-4 and harness its full potential. I will also discuss the roleplaying technique for creating more engaging and immersive conversations with GPT-4.

  1. Identify the purpose and focus of the conversation. Clearly define your objective—whether it's providing information, answering a question, or engaging in casual conversation. This will help you craft a prompt that is specific and relevant.
  2. Use specific language. Ensure that ChatGPT understands your prompt by using clear, concise, and relevant language. Avoid ambiguous terms that may lead to confusion. Avoid company specific jargon unless you define them.
  3. Avoid open-ended or overly broad prompts. While open-ended questions may seem tempting to trigger GPT-4’s creativity, they can result in disjointed or unfocused conversations. Instead, be as specific as possible, defining a clear purpose and focus for the conversation. It is better to directly ask it to be creative rather than imply it in the question’s phrasing.
  4. Provide context and documents. GPT-4 can now use Bing to search the web. To make use of this, enable beta features in your premium account settings. While default search works well, I use the plugins Link Reader and Xpapers, which, together, give ChatGPT access to links, including search; and a database of peer reviewed papers.
  5. Remind it of what it needs to know. GPT-4 can only hold around 3000 words of memory. If you go beyond that, it will try compressing the information. You will lose nuance as you input more information.

Examples of GPT-4 prompts

Good example: "Compose a response to the following email from a client asking for a progress update on the project, including current status, challenges, and next steps." You can then include the email: “email: content of the email.”
This prompt is specific and provides clear context and a goal for the response, making it easier for the AI to generate a relevant and useful answer. When you include the email in your prompt, remember to remove personal information.

Bad example: "Write a response to this client email."

This prompt is too vague, lacking context and information about the purpose of the message or the client's concerns or questions. Even if you include the email’s content, the response is likely to be generic and less useful. You will likely need to use another prompt to improve it.

Good example: “In five bullet points, explain the latest breakthroughs in early lung cancer detection research in simple terms. I want you to search the web and scientific articles. Cite your findings.”

With Link Access and Xpapers enabled, ChatGPT will output clear information with links and citations. Requesting citations seem to reduce the likelihood of hallucination, and links enable rapid fact-checking.

Bad example: “Tell me the latest updates in lung cancer research.”

This prompt can lead to a variety of outputs as it is not directive enough. If search or the proper plugins are not enabled, it will cite outdated breakthroughs or hallucinate recent ones. It is also more likely to generate fake links to sources.

Roleplaying with ChatGPT

The "act as" technique is a valuable method for enhancing ChatGPT prompts by instructing the AI to assume a specific role or persona. This approach can lead to more engaging, immersive conversations and realistic simulations.

To utilise the "act as" technique, include the phrase "act as" followed by the role or persona you want the AI to adopt. For instance: "Please act as a translator and language coach and suggest some improvements to the content below."

Case Study 1: Enhancing brainstorming sessions

You might want to use ChatGPT to brainstorm ideas for a project. Utilising the "act as" technique, you can instruct the AI to assume the role of a subject matter expert or a creative thinker.

  1. Clearly define the brainstorming session's goal to maintain focus.
  2. Use specific prompts to direct the conversation.
  3. Encourage the AI to ask questions and offer feedback to keep the conversation interactive.
  4. Utilise the "act as" technique to ask the AI to assume the role of a subject matter expert or creative thinker.

Sample prompt: "Act as a creative thinker and help me brainstorm innovative ways to improve the way we identify clients."

Case Study 2: Improving decision-making

You can use ChatGPT to weigh the pros and cons of different options and make decisions. Instruct the AI to act as a decision-making expert to analyse the situation and suggest a course of action.

  1. Begin with a clear objective for the decision-making process.
  2. Use targeted prompts to focus on the options and factors to be considered.
  3. Maintain a logical and structured approach throughout the conversation.
  4. Utilise the "act as" technique to direct the AI to assume the role of a decision-making expert.

Sample prompt:  "Act as a decision-making expert and help me evaluate the pros and cons of implementing the following process: ..."

Case Study 3: Developing marketing strategies

ChatGPT can be used to develop marketing strategies by providing insights on target audiences, creating promotional materials, or generating social media content. Instruct the AI to act as a marketing strategist to assist with these tasks.

  1. Clearly define the marketing goal or objective.
  2. Use specific prompts to focus on various aspects of the marketing strategy.
  3. Maintain a consistent tone and style throughout the conversation.
  4. Utilise the "act as" technique to direct the AI to assume the role of a marketing strategist.

Sample prompt: "Act as a marketing strategist and provide me with ideas for a social media campaign to promote our new product."

We hope this primer has provided you with useful, and even playful ways of drawing more compelling results from this latest iteration of GPT. This is the start of a learning curve ... and like search, fluency in the language we use will develop as the technology, and our relationships to it, evolve.

01 Jun 2023

-

Giorgio Tarraf

Illustration courtesy of geralt.

Get the future in your inbox every week.

Your e-mail address is used to send you newsletters and marketing information on L’Atelier. You can unsubscribe at any time by using the unsubscribe link in our emails. You can find more information about the management of your personal data and your rights in our Data Protection Notice.

02/03

Related Insights

A rectangular image of a robot in a warehouse, using a giant magnifying glass to choose the correct box to place on its forklift.
Insight

AI’s biggest leap is hiding in plain sight … and it’s not LLMs

AI-powered visual analysis spots complex medical conditions and guides 30-tonne trucks on public roads, making it among the most transformative technologies today—even if it doesn’t make headlines like large language models.  

    Screenshot of a space-themed shoot-'em-up game, Shoon, created in Midjourney. At left, one small ship fires at several larger ones on the right, with a large explosion in the middle. The battle occurs over a grey cityscape.
    Insight

    The menagerie of ways AI is transforming video game creation

    Stunning AI-generated art is making it easier for developers to create concept art—but it could also endanger the jobs of many artists. AI's advances could have even greater implications down the line.

      Insight

      Everything you need to evaluate open-source (vs. closed-source) LLMs

      A guide to understanding open- and closed-source LLMs, including the ethical and security risks on both sides. Get details on what you should evaluate, and our shortlist of notable open-source LLMs.

        03/03

        L’Atelier is a data intelligence company based in Paris.

        We use advanced machine learning and generative AI to identify emerging technologies and analyse their impact on countries, companies, and capital.

        About us