Randall Munroe, the xkcd guy, has some fun takes on these Google fact-box mistakes: twitter.com/xkcd/status/133352996707912…

Randall Monroe shared a nice series of these last year: twitter.com/xkcd/status/133352996707912… "What year did Tom Hanks land on the moon?" "1970" "When did the Simpsons learn to control fire?" "Dec 17, 1989" "Which river did George Washington drown in?" "The Allegheny River" It's got to be tough to train an AI that can generalize to (or at least recognize) questions outside its training set. Unfortunately, when 99% of questions containing 'when' and 'Neil Armstrong' have the answer 1969, it will need a pretty sophisticated algo to avoid the trap of learning that it should always answer 1969. I wrote up another example of this last year based on a question with a real answer that Google still gets wrong: tedsanders.com/why-does-google-think-th…... Google (and plenty of other search engines) will all helpfully tell you that the slowest animal is the sloth, which is plainly incorrect if you just look at a video of a sloth and then look at a video of a worm. Despite their reputation, sloths still move visibly.

Try asking for details about things that didn't happen. Google is broken. twitter.com/xkcd/status/133352996707912…

news.ycombinator.com/item?id=28220225 theandrewbailey 5 hours ago [–] Try asking for details about things that didn't happen. Google is broken. twitter.com/xkcd/status/133352996707912…

Doesn't Google use language models heavily? But we still get this wrong behavior: twitter.com/xkcd/status/133352996707912…

Ironically the top tweet in the More Tweets section for me was "Being good at things is not the point of doing them." twitter.com/jodotcom/status/13337287196…