Table of Contents 64q43
Since the launch of the Chat GPT, more and more people are using the tool to write texts and sign them as if they were their own. While the tool has helped professionals from different sectors who use it with discernment to have faster deliveries, texts produced entirely with AI are a problem that we have to be careful to face. 294i2e
Distinguish between a text written by THERE (artificial intelligence) and text written by a human is becoming an increasingly difficult task as these tools learn from the information added by each , but there are still some features that make the presence of AI in creating texts more transparent. Check out some tips:
Human text vs synthetic text 4u2mu
At first glance and to the uninitiated, it may seem impossible to distinguish text generated by AI from text written by a human. However, the construction of the text and its elements provide clear clues to identify that we are reading a text generated by generative artificial intelligence. See below some highlights.
Dramaticity 1w2e42
Text written by AI tends to add a degree of drama that is not common in a human text, as well as excessively formal and difficult words, which given the plural nature of readers on the internet is not very common.
In an article published in Medium by the 'Paul Is Positive', he compared results given by some AI tools with texts written by humans — both on the same topic — and this was one of the main points of his research.
The overly dramatic and extravagant sentences are a good indication that the piece was not only written by the AI, but was also accepted by someone who knows nothing about writing.
Paul Is Positive, Medium
Originality and depth 1r3b2e
The author also emphasizes that the text written by AI lacks originality and depth, having a shallow and generalized result on a given topic. The same text, when written by humans, tends to have more information and be richer in details, since when we write an article we need to be based on real knowledge about the topic (the result of a lot of research), in addition to differentiating by our own learning experiences when a text really clarifies a doubt, something that artificial intelligence cannot do.
Hallucination and repetition k655c
One of the biggest problems with text written with AI is “hallucinations” – a term used by the artificial intelligence industry that describes when it generates fake news.
In an article written by CNN, where the prompt used asked that the article be written in the text voice generally used by the newspaper, these 'hallucinations' happened several times. The newspaper also noticed that the text had a lot of repetition of words such as the phrase “increase in demand” and “waves of increase”.
This repetition is something that we humans know we need to avoid, but artificial intelligence cannot capture the need to avoid reusing the same term in the same sentence.
This repetition was also noted by the Google brain, research unit deep learning of artificial intelligence within the Google AI:
“If you have enough text, a really easy tip is that the word 'the' occurs a lot.”
Daphne Ippolito, senior researcher at Google Brain.
Imaginative 3p245c
In the same article, the tool was also asked to create an article about a fictitious but plausible product. This is something that a human could do without great difficulty, but, in the text written by ChatGPT, the result was not only shallow, but with extremely fanciful information that does not coincide with each other.
The tool also demonstrated problems in creating humorous or creative texts in any way, whether creating a recipe or a joke to be published on the website. X (old Twitter). In these cases, we can notice that your “hyper-dramatic gene” emerges even more, calling a non-existent drink a 'classic', when that is not the case.
Lack of basic knowledge is also quite common and perhaps one of the easiest ways to understand the use of AI. As you point out Daphne Ippolito, senior researcher at Google brain:
“If you look at a lot of generative texts and try to figure out what doesn't make sense in them, you can get better at this task [of differentiating human vs AI text]… One way is to take implausible statements, like AI saying it takes 60 minutes to make a cup of coffee".
Daphne and her team created an online game that helps train the human eye so that they can detect text written with AI. Check out: https://roft.io
ZeroGPT 6w1f5b
This AI detector is one of the best known and used worldwide, as well as being one of the simplest. Upon entering the website official, you just need to add the text – which can be up to 15.000 characters in the free version and up to 100.000 in the paid version – and click on the 'Detect Text' button.
The site then performs the check and responds with the percentage of the text detected as written by AI, highlighting the most likely in yellow. As we can see in an experiment carried out by introducing the same text you are reading, it detects that it was written 100% by a human:
We asked ChatGPT to rewrite the introduction so that we could use the results to test the ZeroGPT tool and we came up with this text:
We used the above text in ZeroGPT to test its efficiency and had a positive result:
The platform came to the conclusion that 94.74% of the text had been created by artificial intelligence, not just signaling the action phrase at the end of the sentence – very common in human texts.
These examples show the performance of the tool, which is available for free. There are 2 paid plans: Pro and Max. The Pro plan promises a difference of 2 thousand Prompts in ZeroCHAT-4, search of up to 100 thousand characters per detection, verification of 50 files in batch, manage a report for AI detection and an ad-free experience. All this for USD$7,99 per month.
In addition to all the differences mentioned above, the Max plan offers 3,5 thousand Prompts in ZeroCHAT-5, verification of 75 batch files, as well as an increase to 7 thousand in the summary it offers when found AI (1,5 thousand in the plan free and Pro) and 3 thousand words to paraphrase the AI (300 on the free and Pro plan). All this for USD$18,99 per month.
copyleaks 286p23
Another AI detector we tested to check the same text used above was Copyleaks. To use it, just enter the official website, click on the 'Try It Now' button and it will open a second page where you need to click on the 'text' button and add the text you want to check. This was our result:
The tool was able to correctly predict that all content was written by AI in the free plan, but the tool also offers a paid plan that guarantees to detect AI in more than 30 different languages, API access, detection of the most popular AI models like ChatGPT and Gemini (with more than 99% accuracy), in addition to allowing 2 s per . All this for USD$9,99 per month.
smodin l4s4n
The AI detector, Smodin, also uses percentages to deliver results, as well as highlighting in red the parts of the text that it believes were written by AI. To use this tool, simply access the Smodin website, click 'Start now for free', click 'Free Content Detector' and add your text. Check out our result:
The tool was also able to predict that 100% of the content was created by artificial intelligence in its free version. This tool, however, is much broader than others that are limited to detecting AI and plagiarism. It offers the option of rewriting articles, summarizing them, translating them, as well as completely different functions such as giving grades for work and making your school homework.
The tool works on a credit basis, with the free version only offering 3 credits per week. Paid plans increase these credits among themselves, in addition to increasing the number of characters that can be checked at a time – as is the case with the other two tools. Plans range from USD$15 in the essential plan, costing up to USD$79 in the professional plan, depending on the amount of demand the customer has.
Conclusion 60664y
While we see the lines that separate the real world from the virtual becoming increasingly intertwined, the human touch is still something that artificial intelligence cannot replicate, and in this moment of intelligent technology, this is the biggest tip on how to differentiate a synthetic production from a human. However, this is usually only apparent in long content, as in short sentences or factoids it is more difficult to notice the difference.
The misuse of technology that we have witnessed in recent years should not deter us from using AI as a tool in our research when necessary. But the companies behind them may have to surrender, creating indicators in their own results that make their authorship clear to avoid a future where all information is created by artificial intelligence – especially since it creates from knowledge made by a database dated data, or from 'predictions' of most commonly used and how this can lead to fake news distributions. In addition to a lot of unnecessary drama in texts that should have an explanatory tone.
See also other features
reviewed by Glaucon Vital in 7 / 3 / 24.
