Whilst I like Nostr exactly as it is, I'm willing to build a more accurate "undesired content" filter if people want better filtering options. BoW filter + a six-layer CNN would understand quite a bit of context. The sticking point is building/getting a consistent, quality dataset of labeled examples and counterexamples.

NIPS-56 reports are dumpster-fire quality.

And after reading this depressing thread (not your fault ppl!) I'm wondering how many people on any network even sincerely want content filtering rather than an excuse to exercise dominance over other users...