[ad_1]
The conversations generated by ChatGPT’s artificial intelligence are quite sophisticated, allowing you to write persuasive essays. So Edward Tian, a computer science student at Princeton University, created an app called GPTZero that can “quickly and efficiently” label essays as written by an individual or by ChatGPT. Did.
I spent New Year’s Eve creating GPTZero — an app that can quickly and efficiently detect whether an essay is written by ChatGPT or by a human
— Edward Tian (@edward_the6) January 3, 2023
In a recent series of tweets, Tian provided an example of GPTZero in action. The app determined that John McPhee’s New Yorker essay “Frame of Reference” was written by a human, and that the LinkedIn post was created by a bot. On Twitter, he said he created the app while on vacation, motivated by the growing potential for AI plagiarism.
On January 3rd, Tian tweeted that GPTZero is not working. Probably because he had more web traffic than expected. In his Substack newsletter published by Tian today, more than 10,000 people have tested the version of his GPTZero published on his Steamlit. (At the time of writing, both versions of gptzero.me and Streamlit are showing errors, but this could be due to the amount of traffic.) In the newsletter, Tian said his They updated the GPTZero model to “significantly reduce the false positive rate and improve the output results.” ”
GPTZero uses “perplexity” and “burstiness” to determine if a sentence was written by a bot. Perplexity is the randomness of text within a sentence and whether the way the sentence is structured is unusual or surprising to the app. Burstineness compares these sentences to each other to determine their identity. Human writing is more bursty. That is, we tend to write with more sentence variations.
Concerns about plagiarism have increased since OpenAI launched ChatGPT on November 30, 2022. Over 1 million people have used it in the first 5 days. AI-powered software can write basic essays and emulate the style of established authors. I can give instructions. Although there are impediments to execution, the results are recognizable in a suitable style.It is not difficult to have an AI write high school English style essays and find results that are almost indistinguishable from assignments written by students. That said, there are still limits to what you can do. They are easily puzzled by riddles and sometimes make up facts. StackOverflow has also banned coding feedback generated by ChatGPT due to the high frequency of errors.
In December, OpenAI announced it would “watermark” ChatGPT output to combat plagiarism.
In his newsletter, Tian said he is working on further updates to GPTZero, including “improving model functionality and fully scaling out the app.”
[ad_2]
Source link
