NYC Midnight : Creative Writing & Screenwriting Homepage
Forum Home Forum Home > GENERAL DISCUSSION > Creative Writing Corner
  New Posts New Posts RSS Feed - ChatGPT
  FAQ FAQ  Forum Search   Events   Register Register  Login Login

ChatGPT

 Post Reply Post Reply Page  <1 23456 19>
Author
Message
Suave View Drop Down
NYC Midnight Black Belt
NYC Midnight Black Belt
Avatar

Joined: 25 Jan 2015
Location: Thailand
Status: Offline
Points: 25027
Post Options Post Options   Thanks (0) Thanks(0)   Quote Suave Quote  Post ReplyReply Direct Link To This Post Posted: 15 Jan 2023 at 6:53am
Originally posted by BarbaraFL BarbaraFL wrote:

Originally posted by Suave Suave wrote:

Originally posted by BarbaraFL BarbaraFL wrote:

I just asked it to write a 250-word micro story about my dog (I am not in the competition) just out of curiosity...and it killed off my dog LOL

Ask it to do it again and see what you get.

I asked it why it was so mean to kill off my dog in the story and it apologized

Haha, that is hilarious!
Back to Top
Popsicle Crows View Drop Down
NYC Midnight Black Belt
NYC Midnight Black Belt
Avatar

Joined: 30 Mar 2021
Location: Canada
Status: Offline
Points: 2697
Post Options Post Options   Thanks (0) Thanks(0)   Quote Popsicle Crows Quote  Post ReplyReply Direct Link To This Post Posted: 15 Jan 2023 at 9:27am
Originally posted by BarbaraFL BarbaraFL wrote:

Originally posted by Suave Suave wrote:

Originally posted by BarbaraFL BarbaraFL wrote:

I just asked it to write a 250-word micro story about my dog (I am not in the competition) just out of curiosity...and it killed off my dog LOL

Ask it to do it again and see what you get.

I asked it why it was so mean to kill off my dog in the story and it apologized


OMG, it kicked a puppy. The ultimate bad guy move. LOL

Can you ask it if it can self-destruct?
And again, Trey's story '177 Seconds' wins.

Back to Top
copperdogma View Drop Down
NYC Midnight Groupie
NYC Midnight Groupie


Joined: 16 May 2021
Status: Offline
Points: 240
Post Options Post Options   Thanks (0) Thanks(0)   Quote copperdogma Quote  Post ReplyReply Direct Link To This Post Posted: 15 Jan 2023 at 1:51pm
My wife is a university prof and I've stressed her out by telling her about ChatGTP. She's now worried her students will be using it to cheat.

She did some research and found this ChatGTP detector: https://openai-openai-detector.hf.space/

I tried it on a dozen things that I wrote or I had it write, and it detected the generated stories 100% of the time.


Which is still pretty good.

The issue will be that AI and the way these models are trained are something of an arms race. Almost in a literal sense. One popular way of training them is antagonistic training. Basically you have two AIs, one generating content and the other checking that content and rating how close to the goal it is. All an AI company would have to do is use the ChatGPT detector as the antagonist in its training, which would cause it's generating AI to create more and more realistic writing until the detector was consistently rating the generated story as "100% human made."

So the detector will be good for... Oh.. I'd guess a year, max?
Back to Top
Popsicle Crows View Drop Down
NYC Midnight Black Belt
NYC Midnight Black Belt
Avatar

Joined: 30 Mar 2021
Location: Canada
Status: Offline
Points: 2697
Post Options Post Options   Thanks (1) Thanks(1)   Quote Popsicle Crows Quote  Post ReplyReply Direct Link To This Post Posted: 15 Jan 2023 at 1:59pm
Originally posted by copperdogma copperdogma wrote:

My wife is a university prof and I've stressed her out by telling her about ChatGTP. She's now worried her students will be using it to cheat.

She did some research and found this ChatGTP detector: https://openai-openai-detector.hf.space/

I tried it on a dozen things that I wrote or I had it write, and it detected the generated stories 100% of the time.


Which is still pretty good.

The issue will be that AI and the way these models are trained are something of an arms race. Almost in a literal sense. One popular way of training them is antagonistic training. Basically you have two AIs, one generating content and the other checking that content and rating how close to the goal it is. All an AI company would have to do is use the ChatGPT detector as the antagonist in its training, which would cause it's generating AI to create more and more realistic writing until the detector was consistently rating the generated story as "100% human made."

So the detector will be good for... Oh.. I'd guess a year, max?
It looks like the creators of the program are going to put something similar to a watermark into it so people can't use it for cheating.
Back to Top
laleighda View Drop Down
NYC Midnight Regular
NYC Midnight Regular
Avatar

Joined: 17 Aug 2014
Location: Florida
Status: Offline
Points: 429
Post Options Post Options   Thanks (2) Thanks(2)   Quote laleighda Quote  Post ReplyReply Direct Link To This Post Posted: 15 Jan 2023 at 2:02pm
Thanks for posting that tool! Writing is one of the greatest skills a person can have and I try to promote it in my classes (I teach at a state college, criminal justice). I've gotten good at detecting when someone copies and pastes from a website, but am happy to have a tool that'll help detect AI.

Thanks again!
Back to Top
copperdogma View Drop Down
NYC Midnight Groupie
NYC Midnight Groupie


Joined: 16 May 2021
Status: Offline
Points: 240
Post Options Post Options   Thanks (0) Thanks(0)   Quote copperdogma Quote  Post ReplyReply Direct Link To This Post Posted: 15 Jan 2023 at 2:06pm
Originally posted by Popsicle Crows Popsicle Crows wrote:

Originally posted by copperdogma copperdogma wrote:

My wife is a university prof and I've stressed her out by telling her about ChatGTP. She's now worried her students will be using it to cheat.

She did some research and found this ChatGTP detector: https://openai-openai-detector.hf.space/

I tried it on a dozen things that I wrote or I had it write, and it detected the generated stories 100% of the time.


Which is still pretty good.

The issue will be that AI and the way these models are trained are something of an arms race. Almost in a literal sense. One popular way of training them is antagonistic training. Basically you have two AIs, one generating content and the other checking that content and rating how close to the goal it is. All an AI company would have to do is use the ChatGPT detector as the antagonist in its training, which would cause it's generating AI to create more and more realistic writing until the detector was consistently rating the generated story as "100% human made."

So the detector will be good for... Oh.. I'd guess a year, max?
It looks like the creators of the program are going to put something similar to a watermark into it so people can't use it for cheating.

I heard about that! I'm a programmer by trade and I was having trouble imagining how they'd do that. This article has a good explainer:

https://theamericangenius.com/tech-news/chatgpt-watermark/#:~:text=You%20can't%20currently%20use,stage%20of%20ChatGPT%20feature%20rollouts.

TLDR: Cryptogenic watermarking. "The key to making AI content watermarking virtually undetectable is ensuring the distribution of words still has a unique and random appearance like regular AI-generated text."

That's interesting. If they do it consistently enough within the text, anything you don't rewrite fully by hand would still contain that distribution ("watermark"), so the more you leave unchanged the higher it would rate on something detecting generated text.

That actually sounds workable.
Back to Top
Popsicle Crows View Drop Down
NYC Midnight Black Belt
NYC Midnight Black Belt
Avatar

Joined: 30 Mar 2021
Location: Canada
Status: Offline
Points: 2697
Post Options Post Options   Thanks (0) Thanks(0)   Quote Popsicle Crows Quote  Post ReplyReply Direct Link To This Post Posted: 15 Jan 2023 at 2:12pm
Originally posted by copperdogma copperdogma wrote:


I tried it on a dozen things that I wrote or I had it write, and it detected the generated stories 100% of the tiI heard about that! I'm a programmer by trade and I was having trouble imagining how they'd do that. This article has a good explainer:
https://theamericangenius.com/tech-news/chatgpt-watermark/#:~:text=You%20can't%20currently%20use,stage%20of%20ChatGPT%20feature%20rollouts.

TLDR: Cryptogenic watermarking. "The key to making AI content watermarking virtually undetectable is ensuring the distribution of words still has a unique and random appearance like regular AI-generated text."

That's interesting. If they do it consistently enough within the text, anything you don't rewrite fully by hand would still contain that distribution ("watermark"), so the more you leave unchanged the higher it would rate on something detecting generated text.

That actually sounds workable.
It does! Fingers crossed that they make it happen. Sounds like they want to be successful creatives instead of the bad guys that pisses off people. Better for them to improve their own program and be the heroes.
Back to Top
Deschain View Drop Down
NYC Midnight Addict
NYC Midnight Addict
Avatar

Joined: 18 Nov 2022
Location: Baltimore, MD
Status: Offline
Points: 689
Post Options Post Options   Thanks (2) Thanks(2)   Quote Deschain Quote  Post ReplyReply Direct Link To This Post Posted: 15 Jan 2023 at 2:40pm
Originally posted by copperdogma copperdogma wrote:

From chatting with my other writer friends, we posed this question: If you read a story and loved it, what would change if you were then told it was written by AI?

It's upsetting whenever you read something that you feel emotionally invested in and then learn you were deceived by the author. I imagine this would feel the same as when you read something labeled as a memoir or autobiography and then learn the story was fabricated.

However, if I know ahead of time that a story was generated by AI, and I enjoy it, then I guess I enjoy it.
Back to Top
copperdogma View Drop Down
NYC Midnight Groupie
NYC Midnight Groupie


Joined: 16 May 2021
Status: Offline
Points: 240
Post Options Post Options   Thanks (0) Thanks(0)   Quote copperdogma Quote  Post ReplyReply Direct Link To This Post Posted: 15 Jan 2023 at 2:43pm
Originally posted by Deschain Deschain wrote:

Originally posted by copperdogma copperdogma wrote:

From chatting with my other writer friends, we posed this question: If you read a story and loved it, what would change if you were then told it was written by AI?

It's upsetting whenever you read something that you feel emotionally invested in and then learn you were deceived by the author. I imagine this would feel the same as when you read something labeled as a memoir or autobiography and then learn the story was fabricated.

However, if I know ahead of time that a story was generated by AI, and I enjoy it, then I guess I enjoy it.

I know what you mean. I've been applying that same question to other things. What if I knew a video game I loved was mostly created by AI? What about a washing machine? An instruction manual? A piece of furniture? The answer changes in strange and subtle ways with each thing.

I feel like we're going to spend a LOT of time in the (near) future grappling with these questions on an ever-increasing cadence.
Back to Top
taaaylor View Drop Down
NYC Midnight Black Belt
NYC Midnight Black Belt
Avatar

Joined: 14 Jul 2018
Location: Idaho
Status: Offline
Points: 6946
Post Options Post Options   Thanks (3) Thanks(3)   Quote taaaylor Quote  Post ReplyReply Direct Link To This Post Posted: 15 Jan 2023 at 3:32pm
Originally posted by copperdogma copperdogma wrote:

I hear you. And your original posts. And I'm not sure I can disagree with you anywhere.

From chatting with my other writer friends, we posed this question: If you read a story and loved it, what would change if you were then told it was written by AI?

I cannot agree with the fundamental premise that I would ever like something written by an AI. Particularly creative writing. Unlike art, there are NO objective ratios or formulae by which to measure if something is "right" in creative writing. And even then, the art AIs are still trash compared to a competent artist. It's highly telling that ChatGPT can only semi-succesfully reproduce nonfiction text or code. Nonfiction is consistent, if you write like a dry piece of cardboard. It has a formula that can be derived from millions of similar content articles online.

Creative writing is fundamentally incomparable.

Originally posted by copperdogma copperdogma wrote:

Currently ChatGPT, from my experiments, writes terrible stories;) But this stuff if advancing rapidly, so not being worried yet would be like standing in the rain before an approaching hurricane and saying "it's no big deal -- just a drizzle."

This will be the future. I wanted to ask the question about ChatGPT to get people in a community I love thinking about it and the ramifications for the future.


1) Of course I'm worried. Like many visual artists found after Midjourney AI went big, we are relying on the honor system of the people who built the bot and the people who trained it by uploading content. There are artists who never agreed to have their art included in the learning algorithm who have now been included. (I would not have used Midjourney if I knew that; I naively trusted their claim that all images it trained on were in the public domain.) I have 500k+ words available for free on my Reddit writing account. I have no doubt that my work will be stolen for this endeavor, to train robots to try to replace writers like me.

However, I am not worried about being outwritten by a robot. I am worried about how to protect my IP in this era without damaging my readers or my own success. It will be extremely hard to share writing for free on indexable searches in the next 10 years. I've already had my writing stolen over and over by TikTok or Youtube narrators. This is just the next iteration of lazy art thieves. f**k them all. No respect.

2) As long as quantum computing is only a pipedream, ChatGPT will never be as good as a human creative writer who's been writing for more than about five minutes. Full stop. The simple technical limits of the algorithm + the infinite capacity of the human mind make it no-contest.


Edited by taaaylor - 15 Jan 2023 at 3:32pm
SSC R1
250 Micro R2

SSC 2021 winner
Back to Top
 Post Reply Post Reply Page  <1 23456 19>

Forum Jump Forum Permissions View Drop Down

Forum Software by Web Wiz Forums® version 12.05
Copyright ©2001-2022 Web Wiz Ltd.

This page was generated in 0.156 seconds.