As we all know, ChatGPT likes to hallucinate. For us, it's a feature, not a bug. We find the most interesting use of generating LLMs is in aiding imagination and creative thinking. In this tutorial, we demonstrate how you can use InfraNodus' knowledge graph to steer ChatGPT hallucinations in an interesting direction. It is based on identifying the blind spots in a conversations using text graph structure that represents the topics inside and then generating research questions that bridge those gaps with new interesting ideas.
In this way, your prompt is a request to generate a question, which you can yourself answer or which you can feed to AI to generate an interesting response. This creative ChatGPT prompting is based on the human-in-the-loop approach where you are guiding the ideation process, forcing ChatGPT to hallucinate new ideas which are pertinent to the text's original structure.
This video explains how this can be done using an example of a voice note with some ideas that I created in Otter AI.
Step 1: Apply text network analysis
0:47 InfraNodus: under the hood — how text network analysis works
Step 2: Find high-level topics that define distinct text clusters
1:35 Use ChatGPT to reveal high-level topics in your text
Step 3: Reveal and bridge the blind spots between them
Step 4: Save your insights to notes (for further development)
4:28 Save your insights into Project Notes
Step 5: Think of the answers or use AI to generate them
4:43 Answering to research questions (by yourself)
5:37 Use the AI to hallucinate interesting questions
5:48 Asking AI to elaborate on this question
6:07 Human-in-the-loop approach: guiding the agents in an AI conversation
7:01 ChatGPT hallucinations can be super interesting for imagining new things
8:07 Reiterating through more blind gaps, generating more questions
Step 6: Slice off the top layers of meaning, reveal the nuance
10:07 The stranger, the better. Rephrasing ill-formed responses to stimulate your imagination.
11:28 Bonus: removing the surface parts, getting to the deeper ideas
Step 7: Know when to stop: optimal structural diversity of thinking
12:00 Semantic variability and optimal diversity — aiming to find nuanced topics
12:55 Find the new blind spots: bridge them with new ideas