GPTRadar AI Content Detection Review

AI-based developments have rapidly emerged recently, and many new tools are regularly announced. For instance, the recent release of ChatGPT has spurred interest in developing AI-based content detection tools.

While many users are excited about the potential use cases of ChatGPT, others have concerns about preserving the integrity of written content. AI content detection tools now exist to address this issue, providing people with an invaluable resource for assessing whether or not a piece of content is made by a human or AI.

GPTRadar is one such tool that launched early this 2023. It was created by Alessio Nittoli and Antonio Cappiello of Neuraltext.com and trained on GPT-3. It can analyze a text’s perplexity, burstiness, and distribution probability.

Let’s explore its features further.

Features

• Compliance with guidelines: GPTRadar aims to ensure your texts fully comply with the guidelines Google and other search engines provide.
• Trained on GPT-3: GPTRadar has trained its model using GPT-3, the most popular AI-writing technology.
• Model accuracy and perplexity scoring: GPTRadar places a text sample in either the “Likely AI-Generated” or “Likely Human-Generated” category, then gives an accuracy scoring (or confidence scoring). It also produces a perplexity score that tells how well a language model predicts words based on the ones that came before.
• Token probability visual demonstration: GPTRadar presents two visual demonstrations for token probability. The first one highlights text and shows the potential AI-generated areas or phrases. The second shows a 3D graph of the token probability distribution.

Pros

• Provides a perplexity score
• Gives a confidence score on its results
• Text highlighting with specific problem areas
• Visual results for a token probability distribution

Cons

• Limited accuracy
• No URL or domain scanning
• The scoring system may confuse beginners
• Not free forever

Testing GPTRadar’s Accuracy

Considering GPTRadar uses different parameters for its detection, comparing its accuracy rate with other AI content detection tools will be difficult. So for this testing, we can only observe whether or not GPTRadar catches AI-made content.

In this experiment, we used text samples entirely generated by Jasper.AI, a widely used AI writing assistant for blog creation and copywriting.

To provide a comparison, we will also show the results of Originality.ai on the same samples. Originality.ai is one of the best AI detection tools available because of its high accuracy rate and training on various language models.

Without further ado, let’s see the results!

As you can see, instead of giving a probability ratio for AI-generated content, GPTRadar focuses on general statements and shows the accuracy of these statements.

For example, the tool declares the first sample “Likely AI-Generated” and shows a 64% accuracy. This means GPTRadar is 64% confident that the sample belongs in the given category (in this case, AI-generated).

Looking at the results, you’ll notice that GPTRadar thinks three samples are “Likely Human Generated.” However, they’re only 50+% sure about it. For the rest of the samples that the tool thinks are AI-generated, only two of them have a confidence rate of more than 60%.

On the other hand, Originality.ai uses a probability ratio to determine how much of a text is likely made by a machine. This method is common among AI detection tools and is easier to understand.

Among its results, only the sixth sample is a miss, with the tool detecting only 2%. The rest are at least above 60%. So for this test, we’ll declare Originality.ai the winner based on its results and ease of use.  

GPTRadar Compared to Other AI Content Detection Platforms

So how does GPTRadar fare against other AI content detection platforms? Let’s look at these factors.

• Ease of Use: While GPTRadar has unique metrics for detecting AI content, such concepts could be more beginner-friendly. The tool may be better suited for people with advanced knowledge in AI and natural language processing.
• Pricing: GPTRadar advertises itself as free but is not entirely free. After using the tool a few times, users may have to create an account and pay a small fee for every use.
• URL and Domain Scanning: GPTRadar has yet to offer URL or domain scanning. This is a minor downside since other popular tools can scan entire websites for potential AI-generated content.
• Minimum Text Limit: GPTRadar requires a minimum of 400 words to analyze. That means it cannot be used for copywriting samples. Meanwhile, other tools can process as low as 200 words.

GPTRadar’s Advantages

GPTRadar’s parameters are unique and can be helpful once users are familiar with them. Aside from the general statements and the perplexity score, the tool also shows visual results for token probability.

The token probability here refers to the likelihood of a model predicting a word. Green means that AI rarely would have generated that word, while red means the word is likely to be predicted by the language model.

Aside from highlighting words within the text, GPTRadar also shows a 3D graph depicting the token probability distribution. The tool defines this parameter as “a measure of how often a language model generates a word, based on the context.”

Both are helpful details to know if you know how language models work. It can provide insights on how to write content that tools won’t mistake as AI-generated.

But these are advanced concepts that need to be understood first before GPTRadar can be utilized. Therefore, the tool may not be recommended for beginners.

Conclusion

There is merit to GPTRadar’s unique method of analyzing text. It provides insightful details on how a language model performs, giving people a better understanding of what AI-generated content looks like.

However, its general statements and other metrics may be difficult to comprehend for some users. Only use it when you have a deep understanding of how AI works.

Overall, GPTRadar offers a valuable resource for quickly assessing the authenticity of text written with GPT-3 models. It’s definitely worth checking out if you’re looking to stay up-to-date on the latest advancements in AI technology.