4 ms·

You’re implying their ML/AI interefered with Mark’/ dislike button? Bit of a stretch.

• mschuster91 9y ago
  > You’re implying their ML/AI interefered with Mark’/ dislike button? I don't imply it, but given how many spam filters these days are realized as more-or-less-black boxes, it is a plausible theory. In addition, at least for Twitter and FB, there has been suspicion for long time that reports/blocks have faster consequences when many people report/block, as automated filters take over before a human can look at it. I can imagine that there is a filter that gets triggered when a user gets N reports/blocks in M minutes, with e.g. N=100 and M=30, to automatically flag users as potential spammers, unsolicited nude senders or whatever. This had hit Zuckerbergs account, the spam filter blocked his account... and then someone was tasked with preventing this, by disabling the block function based upon certain criteria. Also, the bug seems to have affected only users blocked by thousands of people. Not exactly your usual testcase, and I don't know if I'd have thought of this scenario when writing the test specifications.
  • kerng 9y ago
    You are assuming that there are test specifications - in many organizations thorough testing before shipping is a thing of the past. A/B and testing in production is the new norm unfortunately. That's probably part of the problem. Since the users are the product being tested it makes even sense.