How to Write a Text using OpenAI's GPT-3 as a Conversational Partner

Most AI-based text generation tools are designed to continue writing a text after you provide it with a "seed phrase" of about 100-400 characters. Most of the time, the results are generic and such tools are unusable for writing long texts or developing an idea.

We adopted a different approach in InfraNodus. 

Instead of writing a text for you, we use OpenAI GPT-3 natural language generator tool to generate the research questions for you. You can then write your text and develop your discourse by answering the research questions. So instead of having the machine doing your work for you, you are working together.  

Below we offer an example workflow that you can use to write a new text. Here we will be exploring the concepts of the "motion sensors" and our goal is to write an article that could provide an interesting angle on this topic, something that did not exist before. 

OpenAI GPT-3 x InfraNodus Writing: Step-by-Step Instruction

...

Step 1: Open InfraNodus and go to the Apps page. Then choose the Insight Generator App:

Step 2: You will have the editor panel opened on the right and the Insight Recommender System that will be generating the research questions at the bottom.

You can then add your first idea (even a few keywords) or click the "Generate Questions" button to generate the research questions:

Step 3: Once you added your idea, for example, "motion sensors", it will be visualized as a graph and the recommendation system will show a recommendation for what you could possibly think next to develop this discourse further. As you don't have much information added, InfraNodus will always recommend to add some more ideas or think of something relevant beyond the topics you already mentioned:

...

Step 4: To generate a set of research questions, click the "Generate Questions" button. InfraNodus will use OpenAI's GPT-3 natural language generator to produce a set of research questions relevant to the topic of your discourse:

...

Step 5: Try to think of the possible answers to those questions and add your answer into the editor to continue writing your text and developing your discourse further.

You can also click the "Regenerate" button to generate more research questions or the "+Ideas" button to generate relevant ideas, instead of the questions. 

In our case, we decide to elaborate on the question "How can sensors be used to measure motion?" by adding the following statement into InfraNodus:

"We can use motion sensors to measure the human body movement in order to provide feedback about the dynamics of their physical motion and its variability."

The system will then visualize our idea as a graph and detect the two topics that could be connected in an interesting way, proposing an interesting link:

We can then click the "Generate Questions" button to generate some interesting research questions that would bridge this link:

...

Step 6: Let's try to answer the question proposed by GPT-3: "How does the human body's response to physical activity vary throughout the day?"

We could say that, in fact, there is a research that shows that human movement variability changes based on the emotional state. If a person is in a calm state of equilibrium, their involuntary body movements will start to have chaotic fractal properties, while if they experience anxiety, then the fractal property of the body movement will fade and decrease. 

Add this into the graph and generate a new set of research questions:

...

Step 7: Answer those questions. For instance, the first one, asking about the relation between variability and motion:

"It has also been shown that the movement with higher variability is more adaptive and resilient. This is due to the fact that when a movement has a high degree of variability, it can adjust to the incoming impulses on any scale (be it the changes in the environment or surface. This idea is also confirmed by the studies in robotics, which have shown that a centipede robot with a fractal property of its gait was more adaptive and resilient in complex terrains."

Add this into the graph, click "Generate Questions" and we have another question generated that proposes us to elaborate more on that last subject:

...

Step 8: You can reiterate several times, using the GPT-3 tool to come up with interesting ideas for your research. You can always click "Regenerate" to generate a new set of questions or click on "+Ideas" to generate ideas instead of the questions. Those ideas will not be the answers to the questions, but, rather the ideas that will be linking the structural gaps between the topical clusters identified by InfraNodus. For example:

We can reiterate several times and generate multiple questions / ideas this way.

...

Step 9: You can also use the Analytics Panel > Insight Tab to detect a structural gap in the graph and to generate the research questions that would bridge this gap to come up with the new innovative ideas:

...

Step 10: Finally, you can either export your text to finish editing it in an external text editor or use the Interpret Graph feature to write a coherent text on top of the ideas that you generated.

...

Note, that the GPT-3 feature is only available to the Pro and Premium users. InfraNodus Cloud subscribers can use it in the demo mode. Log on to https://infranodus.com to try it out!

...