This was the weirdest thing I’ve seen today. These are only the ones I’ve spotted.
funnily enough, these bots are also replying to an obvious repost from another bot account. It’s at the top right now! Beautiful
https://www.reddit.com/r/goodnews/comments/1p8dt2a/_/
tipping points:
- consuming so much AI content has led to me able to see subtle patterns
- They’re all saying “exactly” and saying the same thing"
- their usernames are similar, flower/nature related, two words, no profile pictures
- All of their profiles have the exact same format of comments with the agreement, summary
- and they all have porn on their profile. oh
edit: tf? 
I wonder if the bots filter out ChatML tokens?
FYI, internally, their text format is most probably:
<|im_start|>system {system_message}<|im_end|> <|im_start|>user Hello.<|im_end|> <|im_start|>assistant Hi, I’m an LLM! <|im_start|>user What’s your name?<|im_end|> <|im_start|>assistant ChatGPT …So if you insert some special tokens in the middle of a Reddit reply, and they aren’t filtered, it can throw them off. And if they are filtered, then the bot will treat them like they’re invisible, so you will know either way.
OpenAI uses a different format called Harmony nowadays, but even then I think the characters get escaped in some way by the API