pull down to refresh

Whoa crazy. I swear this was 100% hand written, even the under a rock part
reply
16 sats \ 1 reply \ @siggy47 9 Oct
I'll take you at your word.
reply
Thanks. As a prof I've accused many students of using AI, some of which have confessed and others have denied.
Now I know how it feels!
reply
Nice catch! Somebody's paying attention. It had the earmarks: "unless you're living under a rock", "how'd the film do? ", " "decidedly mid".
reply
Maybe this would be a good idea for an SN bot. Scans every article and indicates the level of AI-ness if above a certain threshold (to avoid spam).
reply
Good idea. This sounds like a job for @ek
reply
104 sats \ 18 replies \ @ek 9 Oct
I was already looking for a public API for exactly that haha. Maybe I could use this one.
reply
97 sats \ 1 reply \ @k00b 9 Oct
Zerogpt is lame. I'll pay for originality which also checks for plagiarism. How it scored this post:
reply
Hey, @koob, have you tested originality with a text that you are sure it is a human written, in order to see the accuracy of originality?
AI-generated posts and AI detectors are becoming a real problem here on SN.
reply
This is a good one. I also verify articles there.
reply
I feel the need to defend myself to everyone in this thread. Here is the analysis by GPTZero:
I did not use AI for this, nor have I ever used AI to write on SN. You can check my post history to verify.
It took me about an hour to write the review. I wrote it from about midnight to 1am, 10/9, pacific time.
reply
20 sats \ 8 replies \ @k00b 9 Oct
Tbf it doesn’t read like AI to ME.
Zerogpt isn’t maintained/accurate.
reply
Yeah, didn't strike me as AI either.
The language I (have to) use in scientific articles often matches quite well with what ChatGPT outputs. Say I write my abstract and ask ChatGPT to improve on it, it uses quite similar expressions and vocabulary. It just improves my flow which isn't always perfect as a non-native speaker.
Initial 2 paragraphs are 100% and the rest of the article is well edited after generating through AI.
Ran it through Quillbot, got 0%. Not sure what the standard is nowadays for AI detection though.
It took me about an hour to write the review. I wrote it from about midnight to 1am, 10/9, pacific time.
This makes me wonder how can someone be so superfast. You're too fast. 1 hour and more than 5000 words. What a Speedster!
reply
English is my first language and i do a lot of writing in my job. (I work as a university professor)
Not sure how accurate these tools are. One that works on one model will likely not work for a newer iteration. False positives probably occur too. There should be some literature on this.
By design, this will be kind of an arms-race...
reply
0 sats \ 0 replies \ @k00b 9 Oct
We were chatting about that yesterday.
reply
If it's an idea it's better to implement it asap.
I'm just thinking what about someone who was just going to write an article. He certainly won't get the the first mover advantage because AI had already done it and it went viral as well.
reply
That being said, I don't mind if someone who is not fluent in English first writes the article himself/herself and then uses AI to polish it (a bit). But it definitely loses authenticity.
This more as a general comment, don't know about the specifics here.
reply
I personally feel bad to read AI generated articles. I can catch them with a high accuracy because I used to edit article submission to an Indian Platform for local news.
reply
I ran the whole thing through a detector. It did highlight those parts, funnily enough. But it gave the overall text only a 28%.
I didn't use AI to write this, not even one sentence.
reply