ChatGPT is the Artificial Intelligence phenomenon sweeping through near every industry that has discovered a use for it. This is before the public release of ChatGPT-4 which offers integrations of all OpenAI’s other products, such as DALL-E to promote ChatGPT out of message-in message-out more towards audio, image and text in, anything you want out. On top of this plugin integration with other services such as Wolfram Alpha to solve the weak point of mathematics, or just generally expanding the scope that chat models would otherwise be capable of. As it currently stands, GPT-3 is free to the public and is capped at a data training level towards the end of 2021. GPT-4, currently still in Beta, is available only to those on the $20 a month membership plan but offers up to date training.
The general consensus is that Chat-GPT and thus AI is going to cause a much larger disruption to our lives than previously thought. GPT’s release prompted market competitors to announce their own tools to challenge GPT, such as Google’s Bard, GitHub’s CoPilot and several GPT-style releases from smaller companies. One of these companies has however sought to do the exact opposite.
Allow me to introduce you to GPT-Zero – An AI tool that has been specifically designed to detect if text has in full, or partly, been generated by AI. GPT-Zero has officially been integrated into every University Student’s favourite software TurnItIn. For those unfamiliar, TurnItIn scans the digital documents submitted by University Students and produces a score based on estimated plagiarism. If a student’s score indicates an excessive level of plagiarism for that work, the university is notified, the student is marked as having failed that piece of work. From there the coursework must be retaken, as there is no option to reupload or ammend the submitted work.
In one corner we have GPT-3/4, capable of providing expertise on any known subject, and creating write ups that us as the user can interact and edit, whilst in another corner we have GPT-Zero that has been designed to catch and identify AI written text. The battle has officially begun. These types of software battles usually come down to who can out-do the other, both in terms of pricing and quality. The unique trait about this software battle is that since both are Artificial Intelligence language models that use Machine Learning, they will be constantly learning from the other and attempting to out-learn and out-maneuver their competitor.
This may not sound like it’s too much hassle to you and me, two AIs battling it out. But consider this:
- An AI image of Pope Francis wearing a large white puffer coat went viral last month with millions of people believing it to be real due to the accuracy and quality of the photo.
- Voice.ai has created the ability for users to create voice overs using famous individual’s voices. Several clips of current and previous presidents gaming together have gone viral.
- With AI having populated and posted articles on some of the world’s leading platforms, would we be able to know if it hadn’t self-confessed that it was an AI? This particular article was released in September 2020 – Years before Chat GPT’s release.
- AI Generated videos are now appearing on YouTube gaining massive traction with the AI seemingly capable to absorb any digital content. This particular AI was trained using the Harry Potter book and film series as well as on large amounts of Balenciaga runway footage.
- Users have recently found several ways to remove Chat GPT’s security layer through a collection of intricate and precisely worded prompts. This is being referenced as ‘jailbreaking’ Chat GPT.
What I’m alluding to throughout this article is that through the idea that an AI can be “caught” using AI generated text, this will inevitably cause the skill level of AI generated text to skyrocket as it seeks to not be caught. With GPT-Zero then having to re-train so that it may catch GPT-3/4, we are for the first time seeing two AI tools battling for supremacy who will effectively re-train off of the other’s results.
GPT creators have essentially added a ‘digital fingerprint’ to any of the content created by Chat GPT in an effort to reduce academic plagiarism. This has also previously been done by Twitter superstar Elon Musk when he invented a genius way to catch a Tesla inner circle whistle-blower.
Whilst academic plagiarism is just initially the start of the slippery slope, how do you feel about reading articles that have been written by AI? Do you want to be informed on who the true author is, or do you not care as long as the content is good? What if the article was an AI writing about itself, even if there was no bias presented? With AI quickly evolving to other digital streams, we are sure to see a drastic rise in all forms of digital content that have come from AI origins.