Everything You Need to Know About GPTZero

Is this a machine or a human-written text? Every day we are moving closer and closer to a reality where this question is more than just a surreal threat. With advancements in AI, specifically language generation models such as GPT-3, capable of producing text that is nearly indistinguishable from human-written content, this threat is causing growing concerns in the education and content writing industries alike.

These advancements enable the possibility of using them for malicious and unethical practices. They have made impersonation and plagiarism a child’s play for anyone. But it is not all over yet; GPT-Zero, the AI-Killer, is here to save the day! This cutting-edge algorithm flips the technology used to build AI like ChatGPT on its head.

GPT Zero uses the GPT model to detect and fingerprint AI-generated text, making it the ultimate tool for content moderation and plagiarism detection. In this blog post, we’ll take a deep dive into the inner workings of GPT Zero, explore its potential uses and benefits, and discover what the future holds for AI-generated text detection.

What is GPT Zero?

November 30, 2022, shocked the world with the release of a prototype of the viral chatbot, ChatGPT. It opened up a world of possibilities for people from all walks of life. However, with this advancement came the concern about its potentially unethical applications.

Since the emergence of ChatGPT in late November, there have been rising concerns about potential academic dishonesty, with reports of students using the language model to submit AI-generated assignments as their own.

Edward Tian, a 22-year-old computer science major and journalism minor at Princeton University, was determined to address this growing concern of AI plagiarism in academia. Determined to find a solution, he devoted his winter break to creating GPT Zero.

GPT Zero is an AI algorithm that sniffs out whether a piece of text has been written by a human or by AI language models like chatGPT. The tool became an instant hit among many educators who had been the most troubled by this new surge in plagiarism.

The app instantly became such a hit that the free servers it was hosted on could not handle the traffic and crashed. Later, Edward took to Twitter to commend the amazing response, expanding on his motivation in a tweet.

“The motivation here is increasing AI plagiarism,” Edward said. “Think, are high school teachers going to want students using ChatGPT to write their history essays? likely not.”

Edward has since grown a fan following of his own. His Twitter handle has become a community of GPT Zero users where Edward himself responds to any queries and provides regular updates about any upcoming features and developments. Follow “@edward_the6” to stay tuned for any updates on the AI sniffer.

The Inside Story

Edward Tian had been pondering the power of AI for years. But how did he come up with this brilliant idea of using the GPT algorithm, which renders human-like text, against it? In an interview with BBC News, he revealed the fascinating story.

It was a packed lecture last year where the computer science major realized the extent of the power of artificial intelligence. His thesis advisor posed a challenge to the class, asking them to pick out which text had been written by a human and which had been generated by AI. To his surprise, many students could not correctly distinguish between the two.

He told Nadine Yousif, his interviewer, that it was then that he had a moment of realization. He realized then that there was a problem that needed a solution. “This technology is only going to get better and better; AI is here to stay,” Edward said. But he added that despite the hype and excitement surrounding the future of AI, there is a dire need to enter this future responsibly.

Determined to find a solution, Edward spent his winter break at a coffee shop in his hometown, Toronto, working on an application that can determine with high precision if a text was written by a human or a bot.

He developed GPT Zero in reaction to ChatGPT, a free online bot that can expertly compose practically anything, from English essays and news stories to diet plans and computer software, from a simple request.

How GPT Zero Works?

Edward believes human writing is beautiful and has aspects that computers can never mimic. This aspect makes human-written excerpts stand out, and it is precisely what his software, GPT Zero, uses to sniff out AI-generated content.

Like the famous detective Sherlock, GPT Zero has its own methods, but fortunately for us, they aren’t too complex to comprehend. We just need to understand “perplexity” and “burstiness” to unravel these methods.

What is Perplexity?

Perplexity, as the name suggests, is just measuring how complex an excerpt is for an AI to understand. A lower perplexity value suggests that the model predicts the text accurately and understands the underlying language better.

For instance, if a model is given “The cat sat on the couch and ____” and predicts “slept” with high confidence, it has a low perplexity score. But if it predicts “danced” with low confidence, it has a high perplexity score, indicating weak understanding of the written text.

What is burstiness?

Burstiness is the measure of the repetition or frequency of words in a given text. A higher burstiness score indicates that the text is more likely to have been written by a human, as it contains a higher number of unique and distinct words.

For example, if 50 out of 100 words in a text are unique, the burstiness score would be 0.5. On the other hand, if only 20 out of 100 words were unique, the burstiness score would be 0.2. The lower burstiness suggests the text was likely written by an AI.

How GPT zero uses them?

Now we have everything we need to decode our Sherlock, GPT Zero’s, methods. GPTZero employs two metrics to evaluate whether an excerpt was authored by an AI: “perplexity” and “burstiness.”

Perplexity, which is a measure of text complexity, indicates if GPTZero is bewildered by the text. If the text has a high perplexity score, it is more likely to have been written by a human. However, if the text is more familiar to GPT Zero—because it has been taught such material—it will be simpler and hence more likely to be generated by AI.

Simultaneously, the burstiness scores indicate variations in sentences. Humans have a tendency to write in bursts, with some longer or more complicated sentences mixed in with shorter ones, and with a variety of word choices of varying degrees of complexity. AI sentences are more consistent, and this can be used to spot them.

GPT Zero analyses the text for these metrics. It gives the input data set scores of perplexity and burstiness. By comparing these values to the expected results for a human-written text, the algorithm can specify whether a text is human or AI written.

In a recent demonstration video of the GPT Zero, the developer Tian shows the app’s analysis in action. When contrasting an article from the New Yorker to a LinkedIn post published using ChatGPT, it successfully discriminated between human and AI writing.