,
6 tweets,
6 min read
Read on Twitter
Whilst looking at post-debate hashtags (specifically #LetYangSpeak), we happened upon an interesting bot: @cooperativeness. At first glance the tweets look plausibly human, but it posts 24/7 via a custom app. Also, the profile pic is stolen.
cc: @ZellaQuixote
cc: @ZellaQuixote
All but 18 of @cooperativeness's 12K tweets were tweeted via custom software called "From My Computer". As far as we can tell, this is the only account using this tool. Currently it tweets every 55 minutes, although the rate has varied and it goes dormant periodically.
What does @cooperativeness tweet about, and how does it decide? The wordcloud built from its tweets consists mostly of common words that don't point to a specific subject matter, and hashtags are a mix of things that have trended at varying points in time.
Checking for repeated phrases unearths a pattern: many of @cooperativeness's tweets appear to be generated based on a template with multiple options for certain words. The most common example is this reply (presented with kitten pics), used 547 times.
Not all of @cooperativeness's tweets appear to have been assembled in this fashion. Many, including the political tweets (in which you'll note potential contradictions) look more consistent with the use of an AI text generation tool, such as GPT-2.
(previous thread on GPT-2 text generation)
