AI-generated Content is the new tech trend of our age. AI content is easier to create than ever before. There are free tools such as ChatGPT3 to do that for you. But how does one detect AI written Content?
Today, because of AI generated Content, many people including students & researchers have started using such tools for their Content needs. It thus becomes difficult for teachers, clients, Content assessors, moderators, lecturers & examiners to determine whether the Content has been written by a human or a machine. But fret not because various tools now exist to detect AI written Content. Here are the top 5:
Open AI, the company behind the popular ChatGPT AI Content generator, has released a new free tool to detect AI written Content. This tool to detect AI Content requires a minimum of 1000 characters to work & offers support for multiple languages. There is no percentage score, but rather the tool offers a “likelihood” model that outputs:
- Very unlikely
- Unlikely
- Unclear
- Possibly
- Likely
On testing the tool, we found it very unreliable: human-generated text & AI-generated text were both labeled as, “Possibly”.
Originality can detect text from a variety of AI Content-models: GPT-3, GPT-2, GPT-NEO, GPT-J & GPT 3.5 (DaVinci-003). The tool is not free to use, but promises to return results of 94% accuracy when it comes to detecting AI generated text. Originality charges $0.001 per 100 words. The tool checks for plagiarism as well as AI written Content, & was developed with serious Content-reliant clients in mind. This AI-detector, also offers a Chrome extension so that checkers can assess while they browse the Net.
This tool supports multiple languages.
Touted by the developers as highly accurate, this free tool offers an easy-to-use interface to check if writing is human or AI-generated in a variety of languages. Users simply need to copy & paste the text they want to check in the text box provided on the platform’s Website. Within seconds, the app offers a result, which gives a percentage between “real” & “fake”; with “real” equating to human-generated content.
This tool performed well when we tested it by assessing human & AI-generated Content accurately.
This tool is free to use, but only detects AI-generated text from the older GPT-2 AI model developed by OpenAI. This detector was built by OpenAI & offers fairly reliable results. Huggingface hosts this tool on their Website, & it’s very easy to use : copy the text to check & paste it into the textbox provided on the website. Click, “predict”, that’s all there is to it.
It offers a predictive percentage, which we found fairly accurate when we put it to the test on various human & AI-generated Content.
Built particularly for educators by a young Princeton University computer science student Edward Tian, this AI detector is free to use & requires no login credentials.
GPTZero uses indicators to detect if text is machine generated. These indicators are “perplexity” & “burstiness”. Burstiness refers to sentence structure diversity, while “perplexity” refers to the complexity of the text. Machines are less likely to show diversity, so Tian uses this as a measure of prediction.
GPTZero is easy to use : copy & paste the text into the textbox provided on the Website, or upload a file.
This tool is still under development, but offered satisfactory results when we tested a number of human & AI-generated samples.
In conclusion: The most accurate tool that we found during our tests was Content At Scale, but there is a word limit of 25000 characters. This tool was also the easiest to use; even on a mobile device.
