I wonder if that's the next iteration of Wikipedia. Right now, the model is to summarize secondary sources. Once summarization becomes trivial via LLMs, the most valuable thing to do would be to assemble ever-expanding datasets of secondary sources for the LLM to pull from.