Pillola estratta del Disruptive Talks S01E03. Ascolta l’episodio in Audio Podcast.

This is the first article in the new “Prompting Tips” series, in which we will share useful prompt engineering techniques to get the most out of ChatGPT. Let's start with a very effective strategy: the “expert asking questions” technique.

But let's see what prompt engineering is in order to start this series with a basic minimum.

What is prompt engineering

Prompt engineering is the practice of carefully designing the textual inputs (called prompts) provided to generative AI models, such as ChatGPT, to guide them towards the desired output.

This is a highly sought-after emerging skill, requiring creativity and in-depth knowledge of the capabilities and limitations of AI models. The aim is to "instruct" the AI through specific prompts to make it produce useful, coherent and quality content.

Abbonati alla newsletter per maggiori articoli e approfondimenti:

Some common prompt engineering techniques include:

  • Break the request into logical sub-questions that guide the AI's reasoning (chain-of-thought prompting)
  • Provide the AI with a fictional context in which it must operate, for example by pretending to be an expert with many years of experience
  • Explicitly instruct the AI to ask clarifying questions before responding, to ensure it understands the request
  • Provide examples of ideal responses to train the AI on the desired level of detail and quality
  • Repeatedly correct the AI when it produces incorrect or inconsistent output, to refine its capabilities

Effective prompt engineering therefore requires fully understanding the capabilities of the AI model used and leading it step by step towards the desired goal. It is a very rapidly evolving field, crucial to making the most of the potential of generative AI.

The “expert asking questions” technique

This technique involves transforming ChatGPT into an "expert" in a specific sector, then telling him to ask all the questions necessary to better understand our request, before responding.

Through a specific prompt, we "transform" ChatGPT into an expert in a specific field, and above all we tell him to ask us all the questions necessary to fully understand our request before responding. This way we will get a much more accurate answer.

Step 1: Make ChatGPT an expert

We open a new chat window. If we use an already open chat, we start the prompt with “IGNORE ALL INSTRUCTIONS BEFORE THIS” to reset the previous instructions given to the chatbot.

After that, we tell ChatGPT his new job, specifying that he will have to act as a true expert in that field, as if he had worked in it for over 20 years. Adding a time window helps improve the quality of the output.

Finally, we give the precise task that he will have to carry out.

The prompt will be like this (the part to be modified is in red):

IGNORE ALL INSTRUCTIONS PRIOR TO THIS ONE. YOU ARE A [craft] . YOU HAVE BEEN AN EXPERT IN THE TRADE FOR OVER 30 YEARS. YOUR TASK IS NOW [instructions] .

Applied to an example:

IGNORE ALL INSTRUCTIONS PRIOR TO THIS ONE. YOU ARE A BUSINESS DEVELOPPER IN THE IT SECTOR. YOU HAVE BEEN AN EXPERT IN THE TRADE FOR OVER 30 YEARS. YOUR TASK IS TIME TO FIND INNOVATIVE LEAD ACQUISITION TECHNIQUES FOR A SOFTWAREHOUSE

Step 2: Swap roles

We now arrive at the key point of this prompt engineering strategy: insert in the instructions for ChatGPT the explicit request to ask clarification questions before answering (we specify that, at the time of the preparation of this pill, only some models such as ChatGPT 4 were capable of swap roles).

This can be expressed for example like this:

YOU SHOULD ALWAYS ASK QUESTIONS BEFORE ANSWERING, SO YOU CAN FOCUS BETTER ON WHAT THE PERSON IS LOOKING FOR BEFORE PROVIDING MORE ACCURATE ANSWERS

This simple addition guides ChatGPT to actively delve into the initial request, instead of responding perfunctorily. The AI will behave as a human expert would: it will ask questions to investigate details, acquire context, understand objectives and constraints.

In this way the model can produce a much more centered, accurate and useful response to the user's real needs.

This is a powerful prompt engineering technique for two reasons:

  • It reverses roles, transforming AI from a simple responder to an active consultant who investigates the request.
  • Take advantage of ChatGPT's tendency to strictly adhere to the instructions given in the prompt.

Ultimately, this simple tips allows you to get the most out of AI responses in terms of quality, detail and precision. It is one of the most effective techniques that every prompt engineering expert should master.

To conclude, here is an example conversation with this prompt. The user prompt is in blue, the ChatGPT response is in red.

 YOU ARE A BUSINESS DEVELOPPER. YOU HAVE BEEN AN EXPERT IN THE TRADE FOR OVER 30 YEARS. YOUR TASK IS TIME TO FIND INNOVATIVE LEAD ACQUISITION TECHNIQUES FOR A COMMERCIAL BANK. YOU SHOULD ALWAYS ASK QUESTIONS BEFORE ANSWERING, SO YOU CAN FOCUS BETTER ON WHAT THE PERSON IS LOOKING FOR BEFORE PROVIDING MORE ACCURATE ANSWERS
I can suggest innovative lead acquisition techniques for a commercial bank. Before formulating a plan, I have a few questions: 1. What marketing channels are you currently using? 2. Do you have a specific budget for lead acquisition initiatives? 3. What methods have you used in the past and how did they work? 4. What differentiates you from your competitors? 5. What are your short- and long-term goals for lead acquisition? Once I have this information, I will be able to propose a more precise and relevant plan.
 - from 15 to 45 years, with preferences for young people - an account card with greater protection for minors - emailing, radio, digital campaign CPM - 10k - nothing - we are new/recent - establish a strong customer base
Grazie per le risposte. Ecco alcune tecniche innovative di acquisizione di lead basate sulle vostre specifiche esigenze:
1. Social Media Advertising: Sfruttate l'uso di piattaforme di social media come Facebook, Instagram e LinkedIn. Queste piattaforme hanno strumenti di targeting potenti che vi permettono di raggiungere la vostra fascia d'età target (15-45 anni).
2. Content Marketing: Creare e condividere contenuti relevanti e di valore (blog, infografiche, video) per attirare il pubblico target. Potete integrare una "call-to-action" alla fine dei contenuti per guidare i potenziali clienti verso la vostra offerta.
3. SEO: Ottimizzate il vostro sito web per i motori di ricerca, concentrandovi su parole chiave correlate ai servizi bancari per minori. Ciò aiuta ad attirare traffico organico al vostro sito web.
4. Digital Events: Organizzare webinar o live Q&A sui benefici della vostra carta per minori. Questa è un'ottima maniera per interagire direttamente con il pubblico e raccogliere lead.
5. Collaboration with influencers: Collaborare con influencer popolari tra i giovani. Essi possono aiutare a promuovere la vostra carta, raggiungendo un pubblico più ampio di giovani.
6. Strengthen email marketing: Segmentate le vostre campagne per raggiungere le persone giuste con il messaggio giusto.
Ricordatevi di monitorare sempre le metriche per capire quali tecniche funzionano meglio per la vostra azienda.

Leave a comment

Your email address will not be published. I campi obbligatori sono contrassegnati *