THE world is anxiously waiting to find out what the future holds as artificial intelligence continues to integrate itself into everyday life.
As it becomes difficult to differentiate between what is human and what isn’t, an app has emerged that claims it can do the work for you.
1
What is GPT Zero?
GPT Zero is a groundbreaking tool which claims it can detect AI-generated text and tell the difference between something written by a human.
It is almost the antidote to its sister system ChatGPT – as it can sniff out something penned by artificial intelligence a mile off.
The technology is an encyclopedia of knowledge of linguistics, writing styles, books and novels produced by humans.
Amid fears AI software will completely undermine the education system as pupils take advantage, one Princeton student decided to provide a solution.
Journalism major Edward Tian, 22, was terrified of the prospect of computers taking over the writing world so created his own app.
The Canadian whizz came up with his own software – ironically powered by GPT-3 – that could bust those abusing the system.
He told NPR: “Humans deserve to know when something is written by a human or written by a machine.
“We’re losing that individuality if we stop teaching writing at schools.
Most read in Tech
“Human writing can be so beautiful, and there are aspects of it that computers should never co-opt. And it feels like that might be at risk if everybody is using ChatGPT to write.”
How does GPT Zero work?
Users simply copy and paste the text they want to examine into GPT Zero, click the “detect text” button and watch the magic happen.
The free tool employs a series of complex algorithms and statistical characteristics to assess the contents with a fine tooth comb.
It is armed with a wealth of data on both human-written work including news reports, as well as the semantics typically used in AI-generated pieces.
By learning from existing AI models, GPT Zero can then calculate and predict the probability of words in an AI-generated sentence.
It can detect jargon from a variety of popular systems such as Chat GPT, GPT-3, GPT-4, Google’s LaMDa and Bard.
The tool analyses patterns in writing using syntax and sentence length to identify machine-generated text – proving a handy tool for professors.
GPT Zero’s ability to differentiate between text produced by humans and computers has been hailed as a game changer by academics.
Is GPT Zero the same as Turnitin?
Similarity detection service Turnitin is not the same as GPT Zero – but the latter could give it a good run for its money.
Although Edward’s creation can pick up on plagiarism, it is not its primary purpose and may be less accurate.
Turnitin still remains the industry titan when it comes to content authenticity analysis and its main focus is exposing cheats.
However, the tech firm is moving with the times and has now introduced its own AI-detecting tool that they claim is 98 per cent accurate.
According to Turnitin, schools and colleges can breathe a sigh of relief as they say their system can successfully identify content created by writing aids such as ChatGPT.
But GPTZero claims to outperform its rival when it comes to eliminating false positives with its advanced algorithms.
How accurate is GPT Zero?
GPT Zero produces a highly accurate analysis of the text and can efficiently flag content created by computer systems.
But there are some drawbacks, as some users have faced issues when attempting to analyse short texts produced by ChatGPT.
As the AI system creates conversational and context-specific responses, it classes with GPT Zero’s statistic-focused system.
Therefore it is important to be aware that the tool may not be as effective when trying to differentiate short texts.
Text from incompatible AI models also may not be understood by GPT Zero.
As well as this, GPT Zero users have been experiencing delays and slower response times due to the number of people using the app.
But developers are said to be hard at work to fix the overcrowding issue to ensure it can efficiently complete its probe at a fast pace.
Edward shared a tweet on January 3, 2023, just after launching his app explaining that it still had some shortcomings.
He told his followers: “We’re still studying implicit bias in language model generated text right now.
“So hopefully will be adding a few more tests and factors to improve the model.”
