If you are feeling overwhelmed by Originality AI then you have come to the right place. Don’t feel intimidated as it is far from being unbeatable. Just like any other AI detector Originality AI is decently inaccurate when it comes to classifying human written text as AI written and vice-versa. With the techniques described in this post you would be more than equipped to beat this one. But first let us know how do these AI detectors work.
How do AI detectors work?
If you have been followed the ChatGPT buzz or have been using these models you will know a thing or two about how it works.
The primary difference between these AI detectors and generative models like GPT-3 is that the former is a classifier while the latter predicts words or tokens based on what it has seen previously.
These classifiers are also machine learning models and have been trained on a huge dataset of text labelled as “AI written” or “Human written”.
Now the interesting thing is that these generative models can also act as excellent classifiers. You will need to train them which is a much deeper topic and we will discuss it some other day.
One important point to note is that all the famous AI detectors like Originality AI, CopyLeaks, ZeroGPT, etc. have not open-sourced their code so we don’t know what they are doing behind the scenes. It might be that they are employing something completely different which is far more complex.
With that being said let us find how to beat Originality AI.
How to bypass Originality AI?
There are quite a few different ways to bypass Originality AI. Let us discuss a few of them.
1. Use a dedicated paraphrasing tool
Although there are several general-purpose paraphrasing tools like Quillbot, you will have to use an AI text humanizer for it. These tools are different than regular paraphrasing tools in the sense that they are made for the sole purpose of beating AI detectors like Originality AI. While regular ones may have several different purposes like changing the tone of the text, flipping complex words with their simpler alternatives, elongating/shortening sentences, etc.
2. Write using sentences of varying lengths using a particular tone.
Try to vary sentence lengths throughout the text and write it in a very specific tone like Informative and analytical tone. For beating Originality AI it is extremely important to avoid sounding like ChatGPT.
Tying back to my earlier concept of AI detectors being nothing but text classifiers, it is very easy for Originality AI to detect that the tone of your writing is very similar to that of any LLM, if you are using ChatGPT to write or edit content.
3. Use uncommon words
Change some of the most commonly used words with their synonyms to fool Originality’s AI detection algorithms. The role of synonyms is that it adds variety to the text which is more human like since most LLMs have an annoying thing is that they often use a very limited set of words.
Please note that the use of synonyms are only effective when used in conjunction with other techniques.
4. Increase Perplexity and Burstiness scores
Although there is no single tool to know your text’s perplexity and burstiness. You can increase it by simply asking it to ChatGPT. Use GPT-4 for this task as it is far more capable than its predecessors. Also, increase the ‘Frequency Penalty’ and ‘Presence Penalty’ to 1. Increasing these parameters forces these models to avoid repeating the same words and writing about new topics.
5. Use a different perspective
LLMs tend to generate text using only a single perspective. You can present your content in such a way that some sections of your text are written from a completely different perspective. It helps your text becoming more human like.
Why you should not fear AI detectors?
What many people fail to understand in the quest of beating these AI detectors is the fact that these Large Language Models (like ChatGPT, Gemini, Claude, etc.) are highly advanced and those who know how to use it can easily beat these AI detectors. They have very low accuracy and many such AI detectors are so bad that you can beat them by just tweaking the ‘Frequency Penalty’ and ‘Presence Penalty’.
This was the exact reason why OpenAI gave up on their AI text classifier within a few months of its release. So, the thing is the company that created the model failed to accurately tell AI and human generated text apart. Now, you can always argue that they have a massive conflict of interest in doing so but I think they rolled back that classifier because they had far more to lose if their detector goes wrong often. For example, many academic institutions might have switched to using their AI detector since it was coming from OpenAI and would have wrongly punished students.
Conclusion
As it is with any other AI content detector Originality AI is also no different – extremely inaccurate and unreliable – and this is why it is very easy to beat Originality AI. Even Originality AI 3.0 (Turbo 3.0 model). Your best bet against any of these detectors is using a dedicated paraphrasing tool. However, you can also use prompt engineering and employ the techniques above to beat it effectively.
Leave a Reply