Thought: we need to synchronize all communication (messengers, notes, emails, etc.) as fine-tuning of the model so that the responses become more and more like your thoughts.
The main question I have is how to create a question-answer link, that is, not just loading text snippets. The OpenAI API expects the format
```
{"prompt": "<prompt text>", "completion": "<ideal generated text>"}
```
One idea is to cut off part of the phrase and train on its ending.
Help me with your thoughts, and what do you think of the idea in general?