I have doubts as to whether this system was even performing online learning yet. Even if it was, that wasn't the cause of many of these issues. Like conversatio

I have doubts as to whether this system was even performing online learning yet. Even if it was, that wasn't the cause of many of these issues. Like conversational bots from the past, they tried to appear intelligent by copying previous responses - with predictable results. At best their machine learning model ended up overfitting like crazy such that it was a near perfect copy-paste. The fact they didn't even have something as simple as a naive set of filter words (nothing good comes from Godwin's law when real intelligence is involved, let alone artificial) is insane to me. Letting it respond to anyone and everyone under the sun (96k tweets - one per second) is just a bad idea given that people would probe every nook and cranny regardless of whether it was near perfect. Additionally, allowing a "repeat after me" option is just begging for people to ask the bot to say idiotic things ... As someone who works in the field of machine learning, this is a sad day. Regardless of whether it involved good machine learning at the base, the copy and paste aspect means it's going to add to the ridiculous hype and hysteria around ML. === Primary proof re: copy+paste (or overfitting at best) from the "Is Ted Cruz the zodiac killer" response: Tay's reply: https://i.imgur.com/PPnCHnf.jpg https://i.imgur.com/PPnCHnf.jpg Tweet the response was stolen from: https://twitter.com/merylnet/status/703079627288260608 https://twitter.com/merylnet/status/703079627288260608 Secondary proof re: copy+paste from https://twitter.com/TayandYou/status/712753457782857730 https://twitter.com/TayandYou/status/712753457782857730: Tweet the response was stolen from: https://twitter.com/Queeeten/status/703049861214547968 https://twitter.com/Queeeten/status/703049861214547968