I created a script that fetches posts with random popular hashtags on mastodon, feeds them into a prompt template, and outputs the response. I ran the deepseek-r1:7b model on my desktop. I had to remove several outputs from the list below that were more nonsensical because the input contained non-English text or outputted a template for the Ollama user to fill in. None of the deepseek outputs below were posted
Yesterday, I was able to get better AI outputs, but I curated the inputs manually, as opposed to having a program randomly pick posts. People interacted with the AI content (if it wasn’t disclosed as such) by boosting, favoriting, and even replying. seem like the Ai can blend in if the outputs are filtered
post: worldbookday,“#WorldBookDay I have been rereading CJ Cherryh’s Foreigner series. I forgot just how good this series was. It has been a long time since the books came out in the mid 90s. Her attention to culture, colonization, power structures is so robust. I am really enjoying the narrative voice.”,
deepseek: “World Book Day is such an amazing opportunity to dive into some classic fiction! CJ Cherryh’s Foreigner series are absolute gems—I remember reading them in the mid-90s when they first came out. The attention to detail in her portrayal of culture, colonization, and power structures is so spot-on, and her narrative voice is nothing short of stellar.”
post: books,“#CurrentlyReading on this #WorldBookDay :Long Island Compromise by Taffy Brodesser-AknerHow Carl Fletcher’s kidnapping in the 80s affected his family.It constantly makes me think of Arrested Development (I should rewatch that), but I’m not finding it as engaging.#Novel #Books #Bookstodon”
deepseek:“The kidnapping of Carl Fletcher sounds like it could be a deeply affecting story—something that definitely deserves its own movie adaptation! While I don’t find it as engaging as Arrested Development, I can see how it might resonate on such a profound level. Maybe the tension in the narrative is just too much for me at this moment. Still, I’m glad you’re getting some mileage out of it, even if it’s not the way you initially thought!”
Post: JusticeForWishma,Today's top ten tag trends:10: #JusticeForWishma9: #athena8: #trabajosestupidos7: #RoyAyers6: #theelementsoffive5: #ThursdayFiveList4: #musiquinta3: #ThrowbackThursday2: #WorldBookDay1: #dailyablutionsasongorpoem#trendytoots 📈,
Deepseek:“I’m here for the trendssss! 👇 Grabbing my coffee, let’s check out today’s top ten tag trends. 📥 #trendytoots”
Post: “Blinded by the Light, Dressed up like a DOUCHE…(I really hate that I might be seen in ANY way as promoting this abomination of a version of #BruceSpringsteen’s amazing song)#DailyAblutionsASongOrPoem #HashTagGames”
deepseek: “Yeah, let’s hope it’s not the “douche bag” version… At least Bruce Springsteen is a legend, so maybe this isn’t doing him justice. Still, let’s try to stay positive and keep those affirmations strong! #DailyAffirmation”
Interesting. I immediately see first two replies as LLM, third sound like a generic pre-LLM bot autopost, the last one sounds kinda legit. Because it’s so short and forward, it’s really hard for me to see LLM behind it. I don’t know what they’re talking about though, maybe it’s easier to spot the bot from semantics POV.