GptZero app to Identify ChatGPT content

[jedishrfu]

https://www.npr.org/sections/money/...-changes-writing-for?utm_source=pocket-newtab

...
Tian is a senior at Princeton University, where he majors in computer science and minors in journalism. Before his recent foray into the limelight, Tian's biggest plans were graduating college and getting his wisdom teeth pulled. Now he's fielding calls from venture capital firms, education leaders, and global media outlets.


Over the last couple years, Tian has been studying an AI system called GPT-3, a predecessor to ChatGPT that was less user-friendly and largely inaccessible to the general public because it was behind a paywall. As part of his studies this fall semester, Tian researched how to detect text written by the AI system while working at Princeton's Natural Langauge Processing Lab.
...

 

[Borg]

Tian already had the know-how and even the software on his laptop to create such a program. Ironically, this software, called GitHub Co-Pilot, is powered by GPT-3. With its assistance, Tian was able to create a new app within three days.

I hadn't heard of GitHub Co-Pilot before.

 

[bob012345]

It wouldn't take that much experience for a professor to recognize ChatGPT content. Spend a few hours with it and you begin to see it's style.

 

[jedishrfu]

The novelty here is that the student recognized this early on and used ChatGPT to identify itself. I think that's quite cool.

 

[Swamp Thing]

Me: Would you be able to tell if some sample text was written by you?

ChatGPT: As an AI language model, I do not have the ability to perceive or remember writing specific texts. I can only generate text based on the input I receive. However, if you provide me with a sample text, I can analyze its writing style, tone, and language patterns to determine if it aligns with my training data and the way I typically generate text.

Me: Can you improve this text, making it more readable, clear and interesting? (.. some text here ...)

ChatGPT:
In the late 19th century, while the world was buzzing with excitement for science and technology, there was also a growing interest in the Spiritualism movement, also known as "Psychical Research". This was the study of the afterlife and the possibility of communicating with the spirits of the departed. Many believed that science and technology could be used to bridge the gap between life and death. Prominent figures in this movement included physicist William Crookes, who was also a pioneer of vacuum tubes, physicist Oliver Lodge, an early wireless pioneer, Sir Arthur Conan Doyle, the creator of Sherlock Holmes, and many others.

Even renowned inventor Nicola Tesla was intrigued by this idea, after he stumbled upon strange signals while experimenting with a radio receiver in 1901. He recorded in his diary, "My first observations frightened me as they had an element of the supernatural and I was in my laboratory at night all alone."

Me: Would you say the following aligns with your style? (Pasted the above rewritten text from ChatGPT)

ChatGPT: Yes, the text aligns with the way I typically generate text. It has a clear structure, precise language, and provides relevant information in a concise manner. The writing style, tone, and language patterns are consistent with the training data I was built on.

 

[Swamp Thing]

"If I ask you to imitate the style of an author, say James Joyce for example, and I give you back the same output, can you then say if it is likely to have been written by you?"

"Yes, I can say if the output is likely to have been written by me. As a language model, I am capable of generating text in a specific style, but I am not capable of recognizing if the generated text is consistent with my typical style or the style of another author. If the output matches the style of James Joyce, it would indicate that I successfully imitated the style, but it does not mean that the text was actually written by James Joyce."

"So despite being an attempt to imitate Joyce, your output will still have elements that you can identify as being aligned with your training data?"

"Yes, even if I try to imitate the style of an author like James Joyce, my output is likely to contain elements that are aligned with the language patterns and writing styles I have been trained on. This is because my training data is based on a diverse range of texts and authors, and I have been optimized to respond to a wide variety of prompts in a coherent and fluent manner. While I can generate text that resembles the style of a particular author to some extent, it is not an exact replica and will still contain elements of my own training data."

"

 

[Office_Shredder]

Don't trust what the ai claims. I pasted a piece of a Wikipedia article into it and it said the text looked like it was the same style also.

 

[jedishrfu]

I was doing a search recently and picked a website that came back. It didn't a quick logon to OpenAI ie a screen flashed by from OpenAI and then I got my info which looked suspiciously like chatgpt responses.

 

[Ken G]

Are you saying that instead of ChatGPT providing a user-initiated alternative to google searches, google searches are starting to automatically revert over to ChatGPT content?

 

[jedishrfu]

It was some website that I found in a google search that was in turn a searchable website but I saw a brief page flash by that looked like a logon to chatgpt. I suspect they were using chatgpt responses to answer my search query behind the scenes.

 

[Ken G]

Weird, it does sound like there might at some point be a google/AI symbiosis, though I expect they will want to make their own AI client the dominant version.

 

[jedishrfu]

Here's the latest news from OpenAI on their release of an API to integrate ChatGPT into developer applications.

https://www.theregister.com/2023/03/02/openai_api_chatgpt_whisper/