Hey MrJorby9410 I am talking about | OpenAI | Page 1

rustic coral Dec 14, 2022, 12:48 AM

#

This is what I was wondering. So far none of what I created has been detected but I have a feeling that soon it will all start to look similar

lost turret Dec 14, 2022, 12:52 AM

#

what are you checking it with

zealous knoll Dec 14, 2022, 12:55 AM

#

rustic coral This is what I was wondering. So far none of what I created has been detected bu...

As in not detectable for plagarism or not detectable by an AI. I am noticing the hugging face gpt-2 detector can also pick up chat gpt content as being AI

clever cedar Dec 14, 2022, 12:56 AM

#

Big D mentioned Originality.ai

I tried plagiarism checkers like grammarly and none of them detected it.

I’m can’t really experiment with originality.ai because it costs, but my thought process was to ask to paraphrase. Maybe even multiple times.

I saw someone in the YT comment section mention that he just asked for lots of nuance.

#

Before I sent this message I asked it to rewrite its answer with grammatical errors (for example) so I think there are definitely possibilities to make it undetectable just gotta find the right requests

#

Technically you could also manually rearrange the sentences, so

@duckyface2.0#8815 I’m not sure there’s really anything at the moment. And they’re gonna release an updated version that’s 10x more powerful I heard

zealous knoll Dec 14, 2022, 1:05 AM

#

@clever cedar I think it has the potential to be undetectable but my thinking is we need to finetune it with non-ai examples of real world human copy... I dont know if it will default back to AI variant or continue using the same tone and predictability of the human copy hence my question. I am taking this project quite serious so happy to take this chat private if anyone comes across this thread

clever cedar Dec 14, 2022, 1:10 AM

#

Yeah, I’ve been trying to make it seem more human like, but it’s really not that good at it. I asked it to interject slang and it just said “yo” at the beginning. And it’s incredibly bad at jokes lol

I would like if there was a way to download ChatGPT and manually give it hundreds of examples of human written essays.

#

If the grammar thing doesn’t work maybe try asking it to rearrange the words. Or paraphrase AND rearrange the words(?)

#

Maybe also try an example. I went to a human written article and copied a portion of it and then asked ChatGPT to rewrite the answer it already gave me in the writing style of the text I copied.

It did change the structure of its answer, but unsure if it’s still detectable.

#

———

I just found that you can ask it also to put a specific amount of grammatical errors in. Which may help you gauge grading just a little bit if a professor grades grammar.

zealous knoll Dec 14, 2022, 1:20 AM

#

@clever cedar Thank you very much for your responses mate, its very interesting indeed

clever cedar Dec 14, 2022, 1:21 AM

#

Of course!

#

Funny enough

#

I feel like in the future someone could write a script that would run ChatGPT several times and then run each answer through a content analyzer. And then grade the answer as good or not based off of if it was detected

#

Only worry is I wonder how bad the grammar would get

#Hey MrJorby9410 I am talking about