Indeed i have seen the concept on youtube back in the days. Also what you saying I agree too, about big LLMs that are trained on others people work (i.e. math). it started as single prompt so the model did its own work with a hight probability of using others people work, after that i worked a lot on this. In the future I am sure that i will use my work too.

Nevertheless, I have gathered many ideas from the comments. if you are interested we can cooperate on the project. its a project out of pure interest exporing on more ways to ingest/spread knowledge.