For the purposes of just generating activity in a subreddit and building karma history for bot accounts, replaying proven content verbatim is more reliable and cost-effective.
LLM paraphrasing is likely to either drift away from "what worked" towards unknown territory or introduce tics that don't really cohere quite right. We can confidently assume it's being used as part of other strategies, but it's not actually optimal here.
The real issue is that Reddit (and Facebook, Youtube, Amazon, Tinder, etc) have very little incentive to aggressively police against this until and unless examples like this make big news and start to harm their general reputation. In the meantime, it just makes their sites look more alive and popular. It's good for them right up until they it becomes a defining association with their brand. This lazy approach works for the bots and the sites, so there's no reason to overcomplicate or take on bigger risks.
LLM paraphrasing is likely to either drift away from "what worked" towards unknown territory or introduce tics that don't really cohere quite right. We can confidently assume it's being used as part of other strategies, but it's not actually optimal here.
The real issue is that Reddit (and Facebook, Youtube, Amazon, Tinder, etc) have very little incentive to aggressively police against this until and unless examples like this make big news and start to harm their general reputation. In the meantime, it just makes their sites look more alive and popular. It's good for them right up until they it becomes a defining association with their brand. This lazy approach works for the bots and the sites, so there's no reason to overcomplicate or take on bigger risks.