this post was submitted on 13 Dec 2023
214 points (89.3% liked)
Fediverse
28528 readers
220 users here now
A community to talk about the Fediverse and all it's related services using ActivityPub (Mastodon, Lemmy, KBin, etc).
If you wanted to get help with moderating your own community then head over to !moderators@lemmy.world!
Rules
- Posts must be on topic.
- Be respectful of others.
- Cite the sources used for graphs and other statistics.
- Follow the general Lemmy.world rules.
Learn more at these websites: Join The Fediverse Wiki, Fediverse.info, Wikipedia Page, The Federation Info (Stats), FediDB (Stats), Sub Rehab (Reddit Migration), Search Lemmy
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Wouldn’t be surprised if it turns out 99% of them were written by ChatGPT or whatever FBs equivalent is.
There does seem to be some real voices and understanding with some of the comments, so probably a mix of paid and chabot. If chatbot is that good, we are really in some serious trouble.
For 3 to 4 sentences LLMs are indistinguishable from humans.
You don’t start noticing the idiosyncrasies until it gets a bit more repetitive and loses coherence during longer texts.
That's so not good, I hope there are people working on ways to tell.
Hopefully, researchers are working on ways to detect these subtle differences.
It's always an arms race, and I fear it's near impossible to detect LLMs from just a few sentences. Longer texts, sure, but how often are the same few words written in a short social media comment?
I got Chat-GPT to write that comment as well, just to further prove your point.