Hey
@primal would very cool to implement language models (can be small and kinda easy to run on-device) to implement a more personalized content discovery
I donβt think scaling is the issue of Nostr (can be fixed) but rather being useful without the need of a centralized honeypot server creating timelines.