I can assure you, wherever you reposted this from was telling you bullshit. It was probably a bug in sequence-to-sequence modeling, where they were working on creating better facebook bots. They probably noticed that other bots are completely fine with friending those other bots that wrote gibberish, and liked their gibberish posts, and realized that the experiment went wrong as these kind of posts won't fool a human.

In any case, bots developing a language to speak in is the smallest of issues in AI world. I'd be more scared of an organization developing an AI that detects depression and targets ads at that person that will help them shop their problems away. Oh wait, it's already been done!

And in our case, we want to build an AI that ingests all of the data available on cryptocurrencies, and automatically trades them. A perfect unbeatable trading bot.