Post by @tazwake@infosec.exchange

2023-08-29 19:37 UTC

In the last month, I did some &quot;AI&quot; experimentation. I used ChatGPT 3.5 and 4 to write some marketing-style LinkedIn posts. The content they produced was (to me) pretty awesome tbh. It felt a lot better than anything I could write. I posted it to LinkedIn and, really, the response was very lukewarm. So next I wrote some myself. Now I am not good at this, but I tried to put some words into sentences. The end result did not read as well (to me) and certainly felt less appealing. I posted it to LinkedIn and in general, they got 3-5x as much interaction as a ChatGPT post got. In addition, this was all pushing towards getting people to book on a course. The AI posts resulted in HALF the number of signups that the human-generated posts got. Now I get that this could be because the human-generated posts came *after* the AI so might have landed on fertile ground, so I will try more experiments. But really, *right now*, I think we are still in a world where HUMAN generated content is more popular with other humans. AI-generated content is very attractive to bots though. #LinkedIn #ChatGPT #AI #marketing #socialmedia

Replies (2)

@dkbgeek@noc.social 2023-08-29 20:45

@tazwake@infosec.exchange I can see 2 things contributing to the results you experienced. First, it's a very normal human trait to be MUCH more critical of our own work than that of others, even (apparently) of "AI" chatbots. I suspect that your own work wasn't as clunky to others as it seemed to you. Second, frankly there's a bit of a weird smell to ChatGPT writing. It's not always a glaring "this is fake!" response but it just comes off as a bit strange to me. Maybe others smell it too.

Open ##5440796
@gnyman@infosec.exchange 2023-08-30 18:19

@tazwake@infosec.exchange interesting, do you want to share the texts? My impression of trying to use GPT4 to write anything is that it produces stuff that is too generic and wordy and I quickly loose interest reading it It might be that with the right prompting this can be avoided, but like you I've just started using GPT to get some rough outlines and then rewrite most of it (as I find fixing stuff easier than creating from scratch)

Open ##5440800