3 ms·
They don't know and can't really tell. ML fraud/flagging systems are opaque and non-interpretable. They can also change with each re-training that might happen
by bitL 7y ago
They don't know and can't really tell. ML fraud/flagging systems are opaque and non-interpretable. They can also change with each re-training that might happen daily.
- wwweston 7y agoWhich is why ML flagging fundamentally shouldn't be used for some unreviewed or non-reversible judgments.
- bitL 7y agoI can understand that the choice is more like: 1) auto-review with false positives/negatives 2) no review at all Frankly, I don't have a solution outside an unrealistic one hiring external experts reviewing each submission which would probably deplete the pool of developers all around the world.