Hacker News Reader

Beyond the question of aggregating someone's likes is the fact that the actual tweets flagged are ludicrous: the system has a third-grader's notion of "bad language" and a temperance activist's view of even _references_ to alcohol (not drinking excessively, or drinking at all, but literally seeing a bottle of vodka somewhere)

View on HN

Even if the language understanding were better, you would still have the issue of the "toxicity" classifier reflecting the personal opinions of the people who built the model and not the range of acceptable behavior for the public at large. These labels aren't objectively measurable statistics: they're personal judgements. The tyranny of algorithmically reified personal moral judgement is far scarier to me than bad keyword matching.

Imperfect people creating imperfect systems to judge others that change the course of life with no means of appeals is specific type of hell.

I guess my point is, that's why you should be wary about giving them that information; because it will be used, in that manner, probably against you, forever.

Reply on news.ycombinator.com