Interactive PROMPTs make the most of GPT

GPT quickly burst into our lives and divided the audience into two unequal parts: techno-optimists and techno-pessimists. Oddly enough, techno-pessimists are mostly professionals who can soberly assess the results demonstrated by GPT. And techno-optimists are people who have not reached the heights of professionalism. They are fascinated by the ease with which GPT generates texts, pictures and videos on a variety of topics.

Controversy continues regarding the benefits or harms of using GPT. I want to offer you an alternative view: it’s not about the tool itself, but about the ways of using it.

In this article, I'll show you how professionals leverage their power by leveraging GPT's interactive capabilities.

Meet: storyteller, negotiator, psychotherapist

Artyom Mushin-Makedonsky – presenter storytelling expertbook author Narrative Leadership, consults top managers of leading companies, conducts corporate trainings. Like many experts, he “works by himself.” Before our meeting, he was skeptical about using GPT, but is now developing an AI-based co-trainer.

Leonid Belenovsky – negotiation expert champion of Ekaterinburg in management wrestling, prepares athletes for management fights and clients for real negotiations. “Works by himself.” During our first meeting he said: “I don’t know how strong ideas come to me – it can’t be technologized.” Currently developing a digital mentor for negotiation preparation.

Maria Bogdasheva – in demand psychotherapist, full entry. I was skeptical about GPT. Now he is developing a neurosimulator for his clients, which will complement personal work, enhancing and consolidating the effect obtained during personal therapy.

All of these professionals were skeptical about GPT until they looked at the technology from a new angle.

What kind of corner is this?

We are talking about interactive PROMPTs.

Interactive PROMPTs

PROMPT is a command, order, statement of a task that a computer program performs.

What do regular PROMPTs look like?

In the case of GPT, it is text in natural language that gives clear instructions about what you want in the answer. There are rules of “good form” for writing PROMPTs: you must specify role, context, purpose, method, format – this will help GPT generate text that meets your expectations.

PROMPT example:

Act like a practicing negotiator. Make a content plan for the week on the topic “effective negotiations.” It is important that there are case studies, life hacks, and strategies. Topics should catch the client’s attention, for example, contain a promise, numbers, questions, etc. For each topic, write a description of no more than 30 words.

GPT response:

Day 1: “5 Keys to Successful Negotiations”

Find out what strategies will help you achieve the desired result and emerge victorious from any situation.

Day 2: “How to prepare for negotiations: 3 steps to success”

Unlock the secrets of effective negotiation preparation and achieve your goals without extra effort.

Day 3: “How to resolve a conflict situation during negotiations”

Learn how to use conflict resolution strategies to successfully complete negotiations and maintain mutual understanding.

Day 4: “Top 5 lifehacks of a successful negotiator”

Learn the best tips and tricks to help you become a true negotiation pro.

Day 5: “How to apply psychology in negotiations”

Discover the secrets of influencing your interlocutor and learn how to use psychological techniques to achieve your goals.

Day 6: “How to avoid mistakes in negotiations: be careful, pitfalls”

Learn about common mistakes made in negotiations and learn to avoid them to achieve success.

Day 7: “Effective Negotiation Strategies: Proven Techniques”

Explore advanced negotiation techniques and strategies that will help you achieve your desired results and secure your position in the negotiation process.

This is a useful prompt, it can save a lot of time. You can run it and get some selection of topics.

Are these topics good or not? PROMPT itself will not answer this question. Most likely, if the results of his work are evaluated by a negotiation expert, he will discover a lot of absurdities.

But let's look at another PROMPT:

I need to write a commercial proposal. Suggest a sentence structure. Ask me questions (one at a time) that need to be addressed in a sentence. In the final, create a sentence text of no more than 500 words.

In response to such a PROMPT, GPT does not produce a result, but begins a dialogue. He asks questions and in the end writes the text of a commercial proposal based on the information received. I described this dialogue in noteso I won’t present it here.

I did the same thing when I needed to develop an agreement for a new service. I created PROMPT, which interviewed me and put together a contract based on the information received.

But who needs such interactive PROMPTs and why? What kind of gain do they give if you still have to work yourself?

To answer, let me briefly remind you how human attention works. Attention is the ability of consciousness to distinguish figures from the background. At the same time, the resource of attention is quite limited. Remember the absent-minded Professor Paganel from the novel “The Children of Captain Grant”? The professor is absent-minded because his attention works with the “figures” of speculation and his resource is not enough for the “figures” of the real world.

When I develop a commercial proposal or contract, it is difficult for me to keep in my attention both the entire document and the part with which I am currently working. I could call the secretary, give him/her basic input and ask him to put it in document form. In this case, my area of ​​responsibility is the details, and the secretary’s area of ​​responsibility is the whole.

In the case of PROMPT we get the same effect – PROMPT is responsible for the whole, and I am responsible for the details. And this really makes the work much easier and faster!

Is the “stochastic parrot” capable of meaningfulness?

I must make one more necessary digression to highlight the capabilities of GPT in working with meanings.

The term “stochastic parrots” was coined by Emily M. Bender, a linguist and professor at Stanford University. According to Bender, a stochastic parrot is an entity “designed to haphazardly link together successive linguistic expressions based on the probability that they are related, but without reference to meaning.”

And since GPT is a “generative pre-trained transformer” or a program that predicts the next word based on previous ones, it is difficult to expect it to make sense. Based on statistics, he knows that the word “Kulikovo” is most likely followed by the word “battle,” but the word “Kulikovo” should be followed by “field.”

However, I want to give an example of how GPT demonstrates meaningfulness. There is such a word game “DANETKI”.

The presenter gives a description of the situation, and the participants, asking questions that can be answered: Yes, No, Perhaps, Doesn’t matter, try to guess what happened.

It's quite difficult to be a player. It is necessary to generate hypotheses and formulate questions in such a way that each answer brings you closer to the solution. Do you think ChatGPT can handle this?

If he acts like a “stochastic parrot” and generates words based on previous words, then such a problem is meaningless for him. It will generate phrases similar in form to questions, but will not come close to the answer.

But let's do an experiment. Here is the PROMPT:

Let's play a game. I thought of a situation: “A man leaves the house and goes to his friend. She comes to him, spends the whole evening with him, then heads home. However, he does not return home.” Your task: to understand why this happened. Ask me questions one at a time. I will answer: YES / NO / MAYBE / DOESN’T MATTER.

GPT accepted the challenge and quickly came up with a solution (vocabulary preserved):

• GPT: Was it dark outside when the man headed home after visiting a friend?

• I might

• GPT: Did the man have an accident while he was walking home?

• Me not

• GPT: Did anything unexpected or unusual happen during the evening the man spent at his friend's?

• Me not

• GPT: Did the man go home along the same route he took to see his friend?

• I might

• GPT: Did the man change his plans to return home during the process of returning?

• Me: yes

• GPT: The man met someone on his way home, what influenced his decision not to return?

• Me not

• GPT: The man decided to stay with a friend for the night?

• Me: yes

• GPT: I understood why the man did not return home after an evening spent at a friend's: he decided to stay at his friend's for the night.

Bingo! How do you like this? We see a series of questions that narrow down the search. They are meaningful and purposeful. It is important to note that this test was carried out on the GPT 4 model, but GPT 3.5 and GigaChat from Sber did not cope with this task.

Summarize:

GPT 4 not only allows you to generate texts, but is also able to work with meanings interactively. This opens up scope for solving a wide range of problems.

Digital twins of experts

Remember that when I introduced our professionals: storyteller, negotiator and psychotherapist, I noted that they “work as themselves.” Clients choose these specialists because they do something that is difficult for the client to do on their own.The main resource of these specialists is their own, uniquely tuned neural network, and the main limitation is physical time.

Let's imagine that an expert can create his own digital twin, capable of providing the client with the same value as himself, but without having the time sx restrictions. The competitiveness of such a professional would increase exponentially!

In November 2023, such an opportunity arose. OpenAI has made it possible to create GPT Assistants. It's like ChatGPT, but with PROMPTs pre-installed. Much of what an expert does in his work can be presented as a set of PROMPTs. Thus, GPT can take over part of the expert's work.

Let's see how the heroes of our story use their digital doubles today.

GPT storytelling mentor

Artyom is a sought-after coach. He teaches how stories help sell, inspire, influence, understand. There could easily be 100 people at his training.

It is clear that Artyom uses certain schemes to work with client stories. It shows how the method works using an example of 2-3 participants. But you can understand how it works only by practicing the method in practice with an experienced mentor. Artyom, like any trainer, cannot work in detail with each training participant. Therefore, they practice the exercises in pairs or small groups. But this is “not that”!

It turned out that the digital twin, created on the basis of GPT, carries out individual work quite efficiently, helping to remember and hone the wording of “that very” story that will touch the soul. Now Artyom conducts trainings together with his digital co-trainer. Moreover, the digital trainer remains with each participant even after the training has ended.

A GPT storytelling mentor allows you to quote a higher price to the customer and provide more value to training participants.

GPT negotiation mentor

Leonid provides individual consultations on preparation for negotiations. The specificity of such consultations is to see more possibilities than are given in the original condition. There is a set of “standard tools” for this purpose. For example, separating facts from opinions and assessments, considering the situation from different points of view (organizational, legal, technological, emotional, etc.). Leonid even uses a filter such as “what’s strange in this situation?” In total, it has about 30 different filters.

Iterating through these filters increases understanding and allows solutions to be discovered. Leonid teaches this to his clients and how to prepare for negotiations with the help of a GPT mentor. In addition, you can always conduct simulated negotiations with this mentor. As a result, face-to-face meetings with Leonid have become more productive: now clients come prepared, and the master has material for discussion and analysis.

GPT negotiation mentor allows Leonid to conduct face-to-face consultations with clients more productively.

GPT trainer for leaving abusive relationships

People are designed in such a way that it is difficult for them to take responsibility for their lives. This leads to a lot of problems. And this is what brings people to a psychotherapist. Like any good psychotherapist, Maria immediately sees the root causes of her patient's problems. But the question is not only about making a diagnosis. It is important to change the patient’s basic thinking, that is, to adjust the connections of his neural network. It takes time and effort. And above all, from the patient himself.

People get into abusive relationships when they unconsciously choose one of the roles: victim, rescuer or persecutor. This model was first described by Stephen Karpman in 1968. The model turned out to be effective, and it was picked up by psychotherapists around the world. Each specialist has his own set of techniques and tools to help stop this game and get out of the vicious triangle that causes suffering. Maria also has such a set of techniques.

A GPT-based co-therapist helps Maria's patients analyze what is happening to them at any time and find ways to communicate healthy with their partners.

Digital Twin Accelerator

For more than 8 years now, I have been unpacking the knowledge of experts in order to create educational materials: courses, regulations, instructions. It is difficult for experts to do this on their own due tocognitive distortion “curse of knowledge”.

I created digital doubles of the storyteller, negotiator, and psychotherapist on an individual basis. It is not simple. The good news is that it looks like I was able to build a process whereexperts, working in small groups using our technological maps, independently master GPT and create their own digital twins. The first stream has already ended.

The whole process took 5 weeks. Each week, accelerator participants worked on one of the following aspects:

• 1 Week: We worked through the description of our value stream (value proposition, stages of value creation, bottlenecks and limitations). We mastered the basics of constructing prompts for GPT (there were participants with a zero level of GPT proficiency).

• Week 2: we adjusted business models taking into account the emergence of a digital twin, and worked out technological maps for the selected stage. We mastered the features of writing instructions for GPT assistants (how to make the digital twin implement a wide repertoire of roles and capabilities).

• Week 3: translated the technological map into the language of instructions of the GPT assistant (assembled PROMPT, containing many PROMPTs). In small groups, we conducted mutual testing of digital twins.

• Week 4: created avatars for their digital doubles (name, image, legend, personality settings). We continued testing and making changes (testing is one of the most time-consuming stages of assistant development).

• Week 5: participants presented each other with the created digital doubles.

The whole work took about 20-25 hours: 5 online meetings for 2 hours and 2-3 hours for group work and homework per week.

If you are interested in the idea of ​​​​creating your digital twin, register free for the first accelerator meeting. You will be able to make an informed decision about fully completing the entire program at this meeting itself.