"Detecting AI-Written Content"... ChatGPT Creator Releases Identification Program
"Complete Trust Not Yet... Still Under Development"
[Asia Economy Reporter Jeong Hyunjin] OpenAI, the creator of the chatbot 'ChatGPT' that sparked a global artificial intelligence (AI) craze, has developed and released a program that can determine whether a piece of writing was created by AI, CNBC and other media reported on the 1st (local time). OpenAI stated that while the program is not completely reliable, it is more stable than the identification program released in 2019.
According to the report, OpenAI announced that it would release a free web-based tool called 'AI Text Classifier' for teachers on the same day. This tool was created to identify cases where students use ChatGPT instead of writing essays themselves, a problem that arose in schools. This measure came two months after the release of ChatGPT in November last year.
To use the program, users must input English text containing more than 1,000 words. OpenAI explained that the performance significantly decreases when other languages are applied. Upon inputting the text, the tool evaluates it on five scales: ▲ Very unlikely (does not seem to be written by AI) ▲ Unlikely (does not seem to be written by AI) ▲ Unclear ▲ Possible (may have been written by AI) ▲ Likely (seems to be written by AI).
OpenAI stated that the program is not perfect and is continuously being developed. According to a blog post by OpenAI, when testing the tool with English texts, it accurately identified 26% of AI-written texts as 'likely written by AI,' but mistakenly judged 9% as human-written. OpenAI added, "The reliability of this tool generally improves as the length of the input text increases."
Hot Picks Today
As Samsung Falters, Chinese DRAM Surges: CXMT Returns to Profit in Just One Year
- "Most Americans Didn't Want This"... Americans Lose 60 Trillion Won to Soaring Fuel Costs
- Man in His 30s Dies After Assaulting Father and Falling from Yongin Apartment
- Samsung Union Member Sparks Controversy With Telegram Post: "Let's Push KOSPI Down to 5,000"
- "Why Make Things Like This?" Foreign Media Highlights Bizarre Phenomenon Spreading in Korea
This is not the first program to determine whether a text was written by AI. OpenAI first released an early-stage detector in 2019. OpenAI evaluated that the current program is much more stable than the tool from four years ago. Additionally, earlier this month, Edward Tian, a computer science student at Princeton University, released 'GPTZero,' which can detect whether a text was written by AI.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
