Thread by @conspirator0

Allow us to humbly introduce @dgfxdfg001, a bot that's more of a cinephile than a propagandist or spammer. We don't believe this account to be malicious in nature (in fact, it's sort of fun); we're posting about it as a possible example of AI text generation.

cc: @ZellaQuixote

First off, how does the automation work? @dgfxdfg001 tweets at five minute intervals and is currently 24/7 although has taken breaks of varying lengths in the past (or deleted tweets.)

What do @dgfxdfg001's tweets actually contain, and where is the tweet content coming from? The tweets are mostly unique (65043 distinct tweets, 68239 total), and what repeated tweets exist are short phrases like "thank you."

Names seem to be a major theme. . .

Things become clearer when one looks at the most frequently mentioned proper names in @dgfxdfg001's - they're all from #StanleyKubrick movies.

We're uncertain as to the exact technique used (account predates GPT-2 by more than two years, so probably not that), but other ML-based text generation methods or Markov chains are possibilities, with a set of Kubrick scripts as input.