Now, you won’t have to blush in front of your employer or teacher because AI wrote the text instead of you. Just make the detectors believe that the author is a human. The article explains how to do this and what the trick is.
How AI Detectors Work
First, let’s figure out what kind of beast the AI detector is. Like a neural network, it works on the basis of language models. That is, the detectors were also trained on a large volume of text – it was created by AI and written by a person. Therefore, such programs can recognize signs of AI writing by template sentences and choice of words and phrases that are usually used by many neural network tools.
The most obvious way to avoid AI detection if you are a student is to get cheap help from the essay writing service where professional writers work. You will receive human-written papers from them.
When you enter a fragment of text into the AI detector, it analyzes the entire context. It determines with what probability AI created the structures and phrases in the text. After the service has summarized the results, you will receive a final opinion on whether you have AI-generated text or not.
How to Identify that AI Generated Text
Today, there is a lot of talk about marking text, images, or other objects that were created by neural networks. The industry even introduced the concept of a “watermark” for neural networks. What is it? In fact, we can draw an analogy with a picture that you get from a photo stock – if you download it for free, it usually has the logo of the photo stock. AI writing tools can also mark their content. In the case of ChatGPT, the watermark is not visible to the human eye. However, ChatGPT generates text using a unique algorithm that can be recognized. The OpenAI classifier can definitely determine this.
The authors of the article Can AI-Generated Text be Reliably Detected? suggest that all llm creators integrate the “Watermark” technology into their models, but such a solution is not yet effective. A user can simply ask another language model to rephrase the text and get AI-generated content that will not be detected by most classifiers. It is not exactly known which language models currently use watermarks, but one can be sure that this method will be used everywhere over time.
Also worth mentioning is mutual information, which shows how much one word in the text depends on the previous ones. If this indicator is at a high level, it means a strong dependence, which is typical for languages with a rich morphological structure. AI detection models trained in such languages take the context into account better when generating or predicting text.
It is important to understand that all AI writing tools are trained on a specific dataset, which may repeat structures when constructing a sentence or using a specific writing style. Each AI writing model also has a “mood” when generating text. However, most of these factors can be simply crossed out by asking another AI writing model to rephrase the text.
How to Avoid AI Detection in Writing
We will show you how not to get into trouble when working on texts using AI writing tools. After all, sometimes teachers can catch you red-handed when you submit a research paper. Serviced can block your article publication, or your text may not appear in search engine results. Scanning and detecting AI-generated content is gradually becoming a problem; no one has accurately and confidently stated how much they are for or against AI-generated material.
But why don’t we do everything in the best possible way so that AI detectors do not catch us? Let’s remember once and for all: what works today may not work tomorrow, and this is no one’s fault. Pay attention to the publication of information and videos; the author is not responsible for the relevance of the information at the time of your review.
The world is changing very quickly, especially in the era of AI writing tools. It would be a stupid decision to simply copy AI-generated content and publish it because this is fairly general writing without a human touch, without its experience and emotionality of writing. AI-generated texts are not interesting to read; there is a lot of fluff, and they feel robotic. This may be useful for making lists and comparisons, but it is not useful for articles. In this block, we analyze in detail how to fix this and pass AI detection from any software.
There are already a lot of options, from expensive to free, to trick AI detectors into believing that a human wrote a piece of text. Let’s look at a couple of them.
Use Third-Party Services
Let’s start with the most radical approach. You can rewrite your content using special services. They will help you bypass some of the most popular AI detectors on the market. The site will rewrite your content and “humanize” it. But, of course, it’s not free; the price starts at $ 9 per month.
For example, BypassGPT is an advanced tool for humanizing AI content, rewriting it to sound more natural and like a human wrote it. Trained on millions of datasets of human content, it performs line-by-line editing and refines AI-generated text using similar word choices and sentence structures that human writers typically use.
This allows you to create authentic-sounding content that can achieve a high level of human writing even in the most stringent AI detectors, such as Copyleaks. The platform can humanize AI-generated text that the author obtained using any AI language model, be it ChatGPT, Claude, Jasper, or GPT-4. It also guarantees 100% uniqueness of the results, which can pass any plagiarism checker, such as Copyscape, Quetext, etc.
Make Changes in Punctuation and Spaces
For classic AI detection tools, it is enough to make a few subtle changes to punctuation and spaces to fool the program and make the result more “human.”
Use a Role
To do this, you just need to assign a role to the neural network, as if it were not a neural network that was answering you but a person. You just need to make the right prompt: “Hi, imagine that you are a person, a copywriter. Write a short post on the topic “Why we need AI detectors” in the most human language possible, please.”
Include Personal Anecdotes
Including a personal story, opinion, example, or real-life experience in AI-generated content can make the text less formulaic and boring. Since AI tools cannot tell you about such situations, it ensures that the text has a level of authenticity that detection tools like Copyleaks recognize as human writing.
Recheck the Facts
AI makes up information because it tries to create text that is not correct but similar to human text. For example, ChatGPT often refers to non-existent works.
Add Imperfections
You can add typos and grammatical errors. However, you should be very careful when doing this. Do not overuse them so as not to create the impression that your content is difficult to read or that a bad writer worked on the text.
Generate Small Excerpts
AI has problems with logic in large texts. ChatGPT often produces not a thought but just a set of phrases, so the text ceases to be meaningful. Generating small parts of the text allows you to control the logic.
Don’t Overuse AI Tools
You should avoid creating full-length content with AI writing tools. To avoid AI detection, it’s best to use the programs to create short drafts or some parts of the text. Later, you can refine it using real human expertise. In short, you should use AI tools to supplement parts of the content, not all of it.
Use Selective Method
Carefully study the composition of the databases that detection tools check. There is a good chance of finding a database on your topic that, for some reason, the system has not yet indexed. It is important to make sure that the material you have your eye on is absolutely absent from other databases; otherwise, you may be surprised when you submit your paper.
Synonyms
Replacing words with synonyms is a very popular way to avoid AI detection. There are several options for using synonyms. The first is to figure out how to replace words yourself. You can select an analog for one word in the form of a phrase or a word combination, use contextual synonyms, and express terms through concepts.
The second method to avoid AI detection involves using a synonymizer. That is, you can perform an automatic operation to replace words. In this case, a synonym will be selected for almost every term that has a pair in the application dictionary. Some options may not fit the meaning, and you will have to change them yourself so that the meaning of the expression is not lost.
Online Translator
In order to avoid AI detection, you will need an online translator. It is necessary to translate your text into a foreign language and then translate the resulting text into English.
Thus, you may have new material. Caution with this method does not hurt. After all, if the dictionary does not have a suitable analog in a foreign language, then the semantic coherence of the AI-generated content may be violated.
Conclusion
In conclusion, it should be noted that this article has not covered all methods to avoid AI detection. Everything that has been discussed is relevant for most AI detection tools that classify machine text. Here, we have demonstrated both methods that make minimal or invisible changes to the text and those that allow for paraphrasing.
Technologyrewiew, a group of researchers, evaluated the ability of 14 tools, including Turnitin, GPT Zero, and Compilatio, to detect text generated by OpenAI’s ChatGPT AI system. The researchers wrote short essays on various topics in different languages and then translated them into English using AI tools. They then used ChatGPT to generate additional texts, which they tweaked slightly to disguise the fact that they were generated by AI. They also used an AI paraphrasing tool called Quillbot to edit some of the texts.
Ultimately, it turned out that while the tools were good at identifying human writing, they were less good at detecting AI-generated content, especially after editing. While the AI detection tools identified ChatGPT text with 74% accuracy, that number dropped to 42% when the ChatGPT-generated text was slightly modified.
The study’s conclusion is that existing tools for detecting AI-generated content are not always effective, especially when it has been slightly modified or edited by human writers.
The developers of AI tools also claim that human intervention (namely paraphrasing) allows users to avoid AI detection. However, we do not lose hope that the situation will change dramatically over time. Of course, more and more AI-generated content that is then paraphrased will appear on the Internet. The dataset will grow, and as a result, the quality of AI detectors will improve.
