There is an application that detects ChatGPT generated content that impersonates humans

July 29, 2023 14:00

Edward Tian built the prototype for GPTZero, a ChatGPT detection tool, in his dorm room at Princeton University (USA), when the world seemed to be fascinated with ChatGPT and Tian was worried about the difficulty of detecting fake content written by humans but actually created by AI.


Edward Tian (right), founder of GPTZero

The 22-year-old says ChatGPT’s arrival felt like opening Pandora’s box, coming out of nowhere and with a lot of potential for abuse. GPTZero has attracted 1.2 million users in just five months. The company has also raised $3.5 million in funding.

GPTZero scans text using its own large language model, a collection of other open-source AI models. GPTZero is trained on both human- and AI-generated text, including news articles and questions and answers.

By learning from existing generative AI models, the tool calculates and predicts the probability of words in an AI-generated sentence. It also analyzes text patterns using syntax and sentence length to identify machine-generated text, and does a good job at it, though Tian said there is room for improvement.

Since the chatbot’s output depends on the data it’s fed, it can sometimes mistake human-generated content for AI-generated content and vice versa. That’s why GPTZero uses both machine learning and human labor to recreate content from different AI programs, ensuring that it will detect AI-generated text with increasing accuracy.

GPTZero can detect text from a variety of AI models, including Google's LaMDa, Facebook's LLaMa, and OpenAI's GPT-3 and GPT-4. GPTZero has partnered with educational organizations Canvas and K16 Solutions to collect training data for its AI detection engine. Educators have found the app to be particularly useful in detecting student essays using ChatGPT.

And there are many applications beyond education. The company plans to market its service to enterprise customers in publishing, social media, and companies working on trust and security.

Tian said he also sees governments as potential customers, which could use AI detection tools to shape policy. The company already has 3,000 paying subscribers for its $9.99/month plan.

The startup will soon launch a browser called Origin that will evaluate the accuracy and provenance of facts in a given piece of published text. With Origin, Tian hopes to help media literacy in the age of artificial intelligence by fact-checking and verifying citations for AI-generated content.

According to Tuoi Tre

(0) Comments
Latest News
There is an application that detects ChatGPT generated content that impersonates humans