Limit penalization on blocks / mutes for a cooldown of 180 days. Fix #658 #660
Conversation
| // what they said is controversial in one period of time but it might be | ||
| // not in other (for example, @RWMaloneMD or others) |
|
LGTM! |
|
Great! I wonder, would 180 days still be an excessively long time? I suspect twitter will likely decide on their own duration, in any case. |
|
|
Right. A simple cutoff may not be the best method either. Having all stats decay by some function over time may be a better idea, though it might be expensive in terms of compute time, since the docs say that the scio graph is generated several times a day. When I follow the trail of function calls here, I see that this is being used to generate a graph of various statistics between users that gets fed into the AI in order to try and predict who will interact with who. I'm not seeing a place anywhere that states explicitly that blocks and mutes are "penalized", only that they're offered to the robot overlord as a metric for consideration. It may well be that some users are more likely to pursue interactions with users who have a lot of blocks, while others don't, and the bot is trying to sort out who's who. It's probably worth trying to untangle the knot a bit more. |
So you said or did something and now you don't want to live with the consequences of people not wanting to talk to you. I have a better fix for this: Grow up instead of trying to commit changes so you don't have to learn. |
@izzy I indeed grew from it… the algorithm didn’t 😉 |
Diminishing only the subset of metrics that humans feel negatively about will only cause inflation of the other metrics. If the NN is any good then it will just inflate the reduced metrics to compensate, meaning that recent blocks/mutes would likely have oversized effects compared to the present. |
|
I actively fight against the full-cancellation of tweeters. LGTM |
|
LGTM, simple and to the point and if anything a good start. |
|
It may be a good stop-gap method, but this does not solve the problem. On this comment I layout how you would attack this time-based decay. Just unblock and reblock the account would essentially reset the counter. #1386 (comment) |
|
Twitter doesn't have a cool off period for block/re-block? If not that's a separate issue entirely and doesn't negate this one. |
|
Using blocks should punish the blocker, not audiences. There's no better way to suppress whistleblowers than giving rings the ability to silence dissent. Twitter had a proactive goal of facilitating "healthy public discourse", blocks are the opposite of that (tho fine for personal reasons). My small/moderate account (5.5K followers) went from ~2K visibility/tweet down to 200 after getting blocked by Yud (AI doom guy), I don't block anyone. It is really discouraging and makes using Twitter near pointless if my followers aren't actually seeing my content as they've explicitly opted into - they're not really mine. Next up: I start an email newsletter? :/ Let's get these changes merged. Tho 180 days is also insane, blocks should have 0 effect on others, only what you see. |
|
Hoping to see this merged at some time! |
@Nightcaat What are you doing here |
Currently, is reported that blocks and mutes are excessively penalized (see #658).
I myself experimented the side effects from cancellation efforts from a woke majority and saw how engagement decreased significantly after that happened.
By setting a window limit on blocks and mutes we still take those blocks in consideration, but only during a cooldown period (180 days), so the account is not permanently cancelled.
This PR fixes it by setting a cooldown of 180 days for blocks and mutes