How to Tell if Chat GPT Wrote an Essay – Plagiarism Detection

Chat GPT, or Generative Pre-trained Transformer, is a powerful language model developed by OpenAI. This model is trained on a massive dataset of text from the internet, allowing it to generate text that is very similar to human-written text. However, there are a few key signs that can help you determine if an essay or piece of text was generated by Chat GPT.

One of the most obvious signs is repetitive or unnatural language. Chat GPT is trained on a large dataset of text, and as a result, it may use the same phrases or sentence structures multiple times. This can result in language that sounds robotic or repetitive. The text may lack coherence or continuity, making it difficult to follow the argument or flow of the essay.

Another sign that text may have been generated by Chat GPT is a lack of context-specific knowledge. This model is trained on a general dataset of text, and may not have the same level of understanding of a specific topic as a human writer. This can result in a lack of nuance or detail in the text, making it less convincing.

Another indication is similar text to other text found online. With the help of the internet and the massive dataset, the model may generate text that is very similar to text that can be found on other websites. This can be a sign that the text was generated by a model rather than written by a human. Copy some passages of the essay into a service like Copyscape. This will cost you only a few cents, but it will scan the internet to check for plagiarism. While Chat GPT will usually produce unique text, it sometimes re-uses a few phrases which will allow you to build a case for plagiarism.

Also make sure to check the text for errors, like grammatical errors, spelling mistakes and punctuation errors. A human writer is more likely to catch these errors and fix them before publishing the text. However, language models like Chat GPT may not be as efficient in catching these errors and may produce text with errors in them. At the same time, an essay with absolutely zero typos can also be a red flag. If your student sometimes mixes up “their” and “there” but then turns in a flawless essay, you might have a cheater on your hands.

While Chat GPT is an impressive model that can generate text that is very similar to human-written text, there are a few key signs that can help you determine if an essay or piece of text was generated by this model. By looking for repetitive or unnatural language, a lack of coherence or continuity in the text, a lack of context-specific knowledge, similar text to other text found online and errors, you can be more confident in identifying text that was generated by Chat GPT.

Here are a few more tips for spotting plagiarism:

1.  Check for similarities: Use plagiarism detection tools such as Copyscape or Grammarly to check for similarities between the text in question and other sources. These tools will highlight any matching text and provide a percentage of similarity.

2.  Look for direct quotes: Direct quotes, if not properly cited, are a clear indication of plagiarism. Look for text that is enclosed in quotation marks or text that is identical to a source.

3.  Check the citation: In addition to direct quotes, it is important to check the citation to ensure that the author has properly cited any paraphrased text.

4.  Investigate the language: Look for unusual or specific language and terminology that may be specific to a particular source. This can be a sign that the text has been copied from that source.

5.  Check the formatting: Compare the formatting of the text in question to the formatting of the source material. If the formatting is identical, this may be a sign of plagiarism.

6.  Check for patchwriting: Patchwriting is a form of plagiarism where the author changes a few words or phrases from the source text but keeps the structure and wording similar. This is a subtle form of plagiarism and requires close examination of the text.

7.   Be aware of self-plagiarism: Self-plagiarism occurs when an author reuses their own work without proper citation or permission. This can be a tricky form of plagiarism to spot, but it's important to be aware of it.

It's important to note that plagiarism can occur unintentionally and is not always a malicious act. Therefore, it's crucial to provide resources and education to help students and writers understand how to properly cite and give credit to sources.

Why a Chat GPT plagiarism detecting program will probably never work

Developing a program that can detect plagiarism generated by Chat GPT will be a challenging task for several reasons. Namely, students are always finding new ways to defeat plagiarism detection tools. As technology advances, so do the methods that students use to cheat. This means that any program designed to detect Chat GPT plagiarism will always be playing catch-up, trying to keep up with the latest techniques and tricks used by students to cheat.

Another reason is that there will always be more students than teachers or programmers trying to stop them. With billions of students globally, it is impossible for teachers and programmers to constantly monitor every piece of text that is submitted. This means that students will always find ways to cheat that are not detected by plagiarism detection tools.

Additionally, Chat GPT is highly sophisticated and can generate text that is very similar to human-written text. This makes it difficult for plagiarism detection tools to distinguish between text that was generated by Chat GPT and text that was written by a human. As a result, plagiarism detection tools may have a high rate of false positives, flagging text as plagiarized when it was actually written by a human.

Finally, it's worth mentioning that even if a program that detects Chat GPT plagiarism is developed, it may not be practical to use it in practice. The cost of running such a program would be high and it would require a lot of computational power. It would also be difficult to implement such a program in schools or universities, as it would require teachers to have specialized knowledge and skills to use it.

Update: OpenAI has released their own plagiarism checker. It can be found here: AI Text Classifier - OpenAI API As expected, it’s not very good. Most of the text I pasted in returned a result of “unclear.” As AI advances it will become increasingly difficult to detect text that was written or partially written by AI.