spam filtering with ML seems risky! and moves casts toward the mysterious algo-ranking that content creators have come to loathe on Twitter/IG (ie: no links or no to certain patterns). then again, there are few options and it will probably mostly cut down on spam/noise replies. tough problem to solve…

would be nice to have some place where users could opt in or out of strictness, models, community supported accounts, etc. some of the weird but fun parts of Twitter look a bit like spam to an AI…

we should also be mindful that a lot of users may not have the reach needed to get “un-flagged” in case of wrongful filtering—it basically means their account is nuked and shadow banned due to the algo like what happens on Twitter, see here:

we should also be mindful that a lot of users may not have the reach needed to get “un-flagged” in case of wrongful filtering—it basically means their account is nuked and shadow banned due to the algo like what happens on Twitter, see here:

https://www.matuzo.at/blog/2022/your-account-is-permanently-suspended/

If the model and training data are open, then it seems like no trust is lost.

But I fear it will be "sufficiently" open, which, in reality, means fully closed.