Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

> Google Docs uses a "Markov Chain" to predict the next word for autocompletion purposes. In this case, of course, since we've already written "And" 5 times, the only logical next word would be "And", as showed in Djikstra's 1989 paper on the subject. Therefore, the Markov Chain never terminates and hence the memory chain overflows with infinite ands.

Does anyone know why this bug doesn’t repro for some words other than And if this is the case?




That statement follows with:

> Obviously, this is partly intentional- Gregory Markov modelled his famous Chain after his younger brother, who would try to finish all of Gregory's sentences for him. The one way Markov could fool him would be to repeat the same word multiple times, and then say "Jinx", also I made all of this up, good luck Google Docs team


Probably because that commenter just made it up.


That comment is a joke. SmartCompose does not use a Markov Chain, it is either still an LSTM or got upgraded to a small transformer based model.


I'm not sure that comment is true based on the second paragraph of it:

> Obviously, this is partly intentional- Gregory Markov modelled his famous Chain after his younger brother, who would try to finish all of Gregory's sentences for him. The one way Markov could fool him would be to repeat the same word multiple times, and then say "Jinx", also I made all of this up, good luck Google Docs team


That was one of the funniest tech comments I've read in awhile.


How is it not code review first comment to limit that lookup to like, 10 steps at most? Baffling




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: