What-would-I-say is a new website that generates
For example, it thinks I might say this on Facebook:
what-would-i-say
Which is pretty accurate!
Here's how they say it works:
Technically speaking, it trains a Markov Bot based on mixture model of bigram and unigram probabilities derived from your past post history.
I think that means the bot strings together words (unigrams) and pairs of words (bigrams) that pop up a lot in your Facebook posts.
It's pretty clever, but it's not perfect.
For example, this proposed status update contains a bunch of terms I might use, but overall it's gibberish:
what-would-i-say
While this one is coherent but gets my political views wrong:
what-would-i-say
Sometimes the bot makes make me sound like a pedantic douchebag:
what-would-i-say
what-would-i-say
what-would-i-say
what-would-i-say
And sometimes it thinks I'm Slate's Matt Yglesias:
what-would-i-say
But when the bot has its moments, it's more insightful than I am.
what-would-i-say
what-would-i-say
Indeed, Markov Bot. Indeed.