,

Next Level AI Prompts 

next level prompts

There is a lot of advice on how to make the best prompts, especially from the many self-proclaimed "prompt experts. Prompting is the way you ask questions or give commands to AI systems such as ChatGPT. If you are interested in getting better outcomes from AI, the report "The Prompt Report: A Systematic Survey of Prompting Techniques," is a must-read. It is a thorough, valuable and academic survey of prompting techniques for large language models, quite technical though. It was conducted by a large team of researchers from the University of Maryland, Stanford, Microsoft and OpenAI, among others. 

What advice do they give in the AI prompting survey?

Be clear

It is important to be clear and specific when drafting a prompt. A clear prompt helps the AI model understand exactly what you expect. Try to avoid jargon and ambiguity that may prevent the AI from understanding what you are talking about or what you mean.

Give examples

By giving examples, AI can better understand what you mean. With a few well-crafted examples, among other things, the model can better understand the structure or tone of voice and respond accordingly. This is especially useful when you want a certain output, such as when creating content. It may be better to provide an example, rather than a detailed explanation.

Ask for clarification

Ask the AI model to explain its thought process or ask it to write out step-by-step how it will do your task. By doing this, you can get a good sense of the thought process and see where the AI might make mistakes. You can also give immediate feedback that will improve the output. This is especially helpful with difficult tasks.

Encourage reflection

Even without examples, you can ask the model to think thoroughly by asking it to look at your question or its elaboration step by step. Prompts such as "Let's look at this step by step" can encourage the model to think more deeply and give more accurate answers.

Give a clear role

By assigning the model a specific role, such as "expert in a particular field," you can specify the context in which the answer is given. This allows you to better determine the tone, style and amount of detail of your output, which is particularly useful when you have specific needs and questions.

Vary your use of language

Try different wording in your prompts because small variations can make big differences in the model's response. Experimenting with this will help you figure out the most effective way to get the right response.

Focus on the essentials

To remain brief and concise, it is helpful to use techniques that allow the model to get to the heart of the answer. This can be done, for example, by asking for a summary or giving only the main points. So focus on the essentials.

Ask multiple times

By repeating the question(s) you have several times, you can check if the AI is also consistent in its outcomes. This is because sometimes the AI then gives different responses. By comparing the responses, you can choose the most common or legitimate answer.

Add additional information

Whenever possible, give the model access to relevant external information or knowledge. This can improve the accuracy and quality of answers, especially in subject areas or certain topics where specific knowledge is required.

Improve continuously

Prompting is a process of testing and adjusting. By testing prompts, analyzing feedback and making adjustments, you can continuously improve the effectiveness of your communication with the model. Not only does this lead to better results, but by doing so, you also increasingly understand how the model works.

Screenshot 2024-07-25 at 22.56.32

Interesting insights for prompting in languages other than English

Translate First Prompting

This is a simple strategy where non-English input is first translated into English. This can improve performance because many models are trained primarily on English data.

Language of the Prompt Template

Using English prompt templates is often more effective than templates in the target language, probably because of the overlap with the training data.However, for some language-specific tasks, native-language prompts may work better.

Cross-Lingual Techniques

Several techniques are mentioned that can improve performance in multilingual settings, such as XLT (Cross-Lingual Thought) Prompting and Cross-Lingual Self Consistent Prompting (CLSP). These techniques are especially useful in situations where accuracy in multilingual tasks is crucial, such as complex translation or multilingual question-answering systems.

As you can see, good prompting goes a lot further than asking some random questions in chat. The important thing is to experiment a lot, because what works well can vary by task and AI model. And create a prompt library, where you capture all the prompts that work well for you.

Take a leap forward in your marketing AI transformation every week

Every Friday, we bring you the latest insights, news and real-world examples on the impact of AI in the marketing world. Whether you want to improve your marketing efficiency, increase customer engagement, sharpen your marketing strategy or digitally transform your business,'Marketing AI Friday' is your weekly guide.

Sign up for Marketing AI Friday for free.

Marketing AI Friday