not uploading datasets could result in more diverse LLMs based on Nostr. which i prefer at this point.