For many years, Edward Tian has been considering the potential of artificial intelligence.
But the Princeton University computer science student first saw how sophisticated this technology has gotten at a jam-packed lecture last year. His thesis advisor requested the students to identify which passages were authored by humans and which were produced by AI after displaying a batch of text in front of the class.
Many students made incorrect guesses. Then he realized there was a problem that required a fix.
Mr. Tian told, AI is here to stay, and this technology will only become better and better.
But he also thinks we need to approach this future wisely.
This is why the 22-year-old spent his winter break working on a program that can accurately identify if a text was written by a person or a bot.
He developed it in response to ChatGPT, a free online chatbot that can expertly compose practically anything from English essays and news stories to meal plans and computer code, all from a straightforward suggestion.
He developed it in response to ChatGPT, a free online chatbot that can expertly compose practically anything from English essays and news stories to meal plans and computer code, all from a straightforward suggestion.
Since its release, ChatGPT’s popularity has alarmed some people, including certain US colleges, which have disabled it on their servers to stop cheating by students. Others fear the bot will replace writers and other creatives’ professions or that hackers would use it to develop malicious software, among other more darker uses.
Mr. Tian, a Princeton senior, claimed that his app, GPTZero, was the first step towards addressing a variety of issues that might come up when artificial intelligence advances and becomes more widely available.
Perplexity and burstiness are the two factors that the app looks at in a text and provides scores to.
According to what it has seen during training, the app first determines how familiar it is with the material being shown. It is “more likely to have been written by a human” if it is less familiar, which increases the text’s ambiguity, according to Mr. Tian.
By scanning the text and determining how varied it is, it then calculates burstiness. Is there a mix between short and long sentences, for instance? Is the writing more level and consistent, or the opposite?
A human-written essay will vary greatly, Mr. Tian claimed, if you plot accurately across time. It experienced unexpected increases as well as ups and downs.
Although he has made GPTZero available for public usage in beta form, he is currently trying to improve it. He showed how the program can accurately distinguish between a letter written by ChatGPT and an essay that was published in the New Yorker magazine.
He claimed that with a false positive rate of less than 2%, the software accurately predicted the difference between the texts.
Numerous users have downloaded and used Mr. Tian’s app since it was first released. Since then, he claimed, he has had calls from educators and admissions representatives from universities all around the world who are curious about how it operates.
While Mr. Tian acknowledged that GPTZero was developed to tackle academic plagiarism, he believes that apps like his will also be used to tackle other problems that will arise as artificial intelligence gains popularity, such as internet disinformation campaigns.
But he is not anti-artificial intelligence; on the contrary, he expressed excitement over its development and admitted that it has helped him create computer code and handle other issues. But he added that it was crucial to create security measures for every new technology since it lends its application credibility.
But he asserted that the popularity of his app testifies to “a human drive to know the truth” more than anything else.
