It seems that in a world where models are becoming more and more similar to each other, unique data for training is becoming key. If something is publicly available (like a Wikipedia article), the value of that data approaches zero; it will be in every model.
On the other hand, data that is not accessible to anyone, especially personalized data, is a different story. For instance, the history of your correspondence in email, logs of your chats in ICQ during childhood and in Telegram now, as well as your notes in a notebook or a second brain system like Roam—this is what represents real value.
Projects that can utilize this data (ideally not just in RAG format) will stand out compared to all others.
What do you think?