Simple because it matters.
Digitalisation & Technology, 24 October 2024
The art of prompting
Effective communication with Chat GPT & Co. for the best results
AI language models such as ChatGPT have become part of everyday life for many people, both privately and professionally. To utilize the full potential of these language models, it helps to understand how to interact with them correctly. This process is called prompting. In the following, our Prompt Engineer Sebastian Holzborn gives practical tips on how to use effective prompting to get the most out of ChatGPT or other language models. He also gives an outlook on how interaction with language models is likely to develop. One thing can already be revealed here: It will become simpler, more natural and more intuitive.
But first a few basics:
What is GPT?
GPT stands for "Generative Pre-trained Transformer". This refers to a language model that is based on machine learning. GPT uses large amounts of text data to understand and generate human-like language. Among other things, this technology makes it possible to formulate texts, explain complex facts and conduct context-related interactions.
What is prompting - and why is it important?
Prompting means formulating an input that is entered into the language model. In other words, a prompt is the request that is addressed to the model. The quality of the prompt determines the quality of the results. The clearer, more specific and better thought out the prompt is, the more appropriate and helpful the response from the language model will be.
Especially for complex issues, specific topics or precise expectations, targeted interaction with a language model such as ChatGPT can offer great advantages. This leads to the question: How can we ensure that ChatGPT provides us with exactly the information we need? And: How reliable is this information? Here are five important tips:
Make it a real conversation
A common mistake when using ChatGPT is to think of the interaction as a one-way street - one question, one answer. But ChatGPT works best when we make it a dialog and give ChatGPT a role. Rather than making an isolated request, it's better to invite ChatGPT to ask questions itself to ensure it fully understands the requirements and delivers the best result.
Practical example:
ChatGPT should create a template for a customer approach. Instead of simply writing: "Create a customer approach for a life insurance company", we can better write: "Act as an expert in customer communication. Create a sales pitch for a life insurance company. If you need more details, feel free to ask me questions." This gives ChatGPT a role and also the opportunity to clear up any ambiguities in advance in order to provide a well-founded answer.
Be precise and concrete
The more specific we make the prompt, the more accurate ChatGPT's response will be. We make sure to set clear framework conditions and provide all relevant information, the so-called context. If we want a response to have a certain structure, such as output as a table or use a particular tone, we specify this directly.
Practical example:
Instead of saying, "Write an email to a customer," we say, "Write a short, upbeat email to a customer to explain new benefits of their car insurance with our company. Use confident language. The customer likes to fish in their spare time."
Encourage ChatGPT to self-assess and recognize hallucinations
A useful, often underestimated trick is to ask ChatGPT to evaluate your own result and make adjustments. For example, we can ask the model to analyze its own answer, rate it on a scale and then create an improved version based on this self-assessment.
Practical example:
"Please create an explanatory letter for a customer who has difficulty understanding the terms and conditions of their occupational disability insurance. Then rate the result you have produced on a scale of 1 to 10 and explain how you arrived at this rating."
A follow-up prompt could be: "Now adjust the result so that you would rate it as an 11." The 11 is outside the scale mentioned and will still lead us to better results.
Self-assessment also helps to recognize hallucinations. Hallucinations are false statements made by the model that sound correct. As a rule, the ChatGPT score is 8 to 9. If the score is 7 or lower, the result should be checked. The explanation also provides information on how good the result is. The more convincing these explanations are, the more valid the result can be classified.
Demand empathy and allow it to adapt
Empathy and understanding play an important role in our lives. What is practical: ChatGPT acts very empathically and solution-oriented by default. We can also explicitly ask ChatGPT to assess the level of empathy of the responses and adjust them if necessary. This ensures that even complex or sensitive topics are communicated in a way that builds trust and promotes understanding. In order to avoid the typical formulations of the language model, it is sufficient to add that the result is a human-sounding text. Capitalization further supports this point.
Practical example:
"Write a letter to a customer whose claim has been rejected. Rate the tone in terms of empathy on a scale of 1 to 10 and strengthen it if you find it too factual or cool. Please write the text as a HUMAN-SOUNDING text." As a rule, a rating of 8 is also given here as a result. Now we can also have the text adapted so that ChatGPT writes the text more objectively: "Please adapt the text you have created so that you would rate it with a 5.".
Context brings better results
The more context we provide, the better ChatGPT can understand what we want. For example, if we are preparing an answer to a complex question, it is helpful if we provide the specific concern, the background to the question and the desired tone.
Practical example:
"Please create a factual and understandable explanation of how companies like insurance companies can use ChatGPT to improve their customer communication. The tone should be professional and informative as I would like to use the text for an internal presentation. Please focus on the key uses such as customer service and marketing without going into too much technical detail."
So much for the tips. But what about their half-life?
The future of prompting: moving towards natural conversations
An understanding of how ChatGPT works and the ability to formulate clear prompts are crucial today in order to achieve the best results quickly. But what will the future look like? AI language models like ChatGPT are expected to evolve towards even more natural and intuitive conversations. Complicated and detailed prompts - such as those presented in the article - could soon be a thing of the past, as AI will increasingly be able to understand intentions from natural dialog and ask for the information it needs in a targeted manner.
This means that interaction with the AI will be even easier for everyone. Users will then not need to have a deep understanding of how the prompts work, but can rely on ChatGPT to understand them through natural conversations and provide targeted support. This would significantly increase efficiency and at the same time enable barrier-free use for everyone.
Conclusion: The optimal use of ChatGPT
ChatGPT is a powerful tool that can make our work and lives much easier - provided we know how to communicate effectively with ChatGPT. With the right prompts, we quickly get the results we need.
The future of prompting will lie in natural, intuitive conversations. Soon, the need for highly specific prompts could diminish as AI automatically understands and asks for more and more context. What this means for all of us is that the use of ChatGPT and other AI tools will become even easier, more effective and more targeted.
Our tip: Just try it out for yourself and test the tips above while you still need them.
Text: Sebastian Holzborn
Your opinion
If you would like to share your opinion on this topic with us, please send us a message to: next@ergo.de