The Spanish platform designed to hunt for plagiarism and student work whose real author is ChatGPT

The Spanish platform designed to hunt for plagiarism and student work whose real author is ChatGPT
The Spanish platform designed to hunt for plagiarism and student work whose real author is ChatGPT

With generative AI, anyone with a modicum of skill can impersonate an expert in a subject, sign texts full of data, write university or scientific (or journalistic…) papers and stand out, get brilliant grades, with minimal effort.

So in the intellectual ecosystem there arises the need for a new species that is capable of Catching cut and paste cheatersto maintain a certain “ethical” balance.

This is what the Lithuanian company is dedicated to. I identifywith a specific service in Spain through the website plag.es: catching plagiarism and texts created with artificial intelligence.

In 2011, when AI was only being talked about in a few specialized circles, Arnoldas Viburys y Chorst Klaus They founded a startup called Lingua Intellegens in Vilnius, Lithuania, which is focused on education.

“In 2017 an accelerator invested in our company and we changed the business towards language intelligence”, Klaus explains to DISRUPTORES – EL ESPAÑOL. They changed the brand to Oxsico, which refers to “Oxford similarity checker”, focusing on helping to “maintain academic integrity.”

And now, in the midst of the era of generative artificial intelligence and the powerful tools it brings with it, the company renamed again, now as Identific, in which Klaus appears as founder, while his colleague Viburys is the chief technology officer, presents itself as a hunter of plagiarism and texts generated with AI.

The company presents its Plag platform as a detector of texts in which it has intervened ChatGPTboth in the academic and business fields. His plan is to add the same specificity to other LLM models, such as Google’s Gemini.

AI-generated phrases

Offers services of plagiarism detection; elimination of them in own texts; formatting of texts to conform to the style standards of an institution; correction of incorrect citations and grammatical correction of texts.

In some of these cases they intervene professional human editorsfor example, to refine a paper. They check citations, or “add them if they are missing,” and “if they see that a text is excessive, they can rewrite it a little… without adding anything or changing the original idea.”

The recognition tool uses a color code to highlight sentences that may have been created by AI. The model’s creators claim that it is 99.8% accurate. If it indicates more than 50% probability of AI intervention in a text, it can be assumed with a fair degree of certainty that it was generated automatically.

And how is it technically discovered that this text is the product of artificial intelligence?

“We have a wide collection of texts written by humanswith which we have trained our AI model to recognize it. Although we are still working to make the detection more precise,” says Klaus.

“We add information about the language, such as how many nouns there are [en el texto]in what order, and things like that. We train the model on texts created by humans and by AI, and that makes it easier for it to distinguish between the two. When you upload a text, you can see which parts look more like AI-generated and which ones look more like human-written. It shows us the statistical possibilities of both options,” he says.

Up to 129 languages

In a test carried out by this journalist with his own text, Plag analysed it as free of AI intervention. However, after an automatic translation into English he detected 50% automatism. Klaus has the explanation: “It is because of the translation algorithm. In Lithuanian, the same thing happens when we translate into English. The language is very correct, very structured.”

As for languages, plag.es, which is presented in Spanish, declares itself “fully multilingual” capable of detecting plagiarism and cheating in 129 languages.

They may even appear mixed different languages “within the same document.” The co-founder explains that “in many papers “Two languages ​​can be used, and in conference proceedings, which can reach 300 pages, you can find articles in German, Spanish, French and English. The trick is to divide the document into smaller parts, identify the language in each of them and apply the corresponding AI model.”

The company’s flagship website, identific.com, can be used in a dozen languages, including Spanish, English, French, German, Italian and five other Eastern European languages, as well as being able to analyse documents in the same 129 languages.

“The difference between the two is that Plag is open to anyone. Students and teachers can register and use it for free in some of its functions. On the other hand, Identific is for universities and companies. They have different functionalities,” says Klaus.

“When it comes to students and AI, it’s a complicated issue: most are Using that tool already in schools and we are receiving many queries from teachers about how to work with that,” he adds.

According to a survey of high school students in Spain, Germany and Italy, led by plag.es, 87% of those who have heard about generative AI say they have already used tools such as ChatGPT or Gemini (Google). Of these, 72% have used it to search for information and 63% have used it to find information. to do homework.

In Spain, with a study by TMG Research in 63 secondary schools and consulting 1,006 students, only 6% said contrary to using AI toolseven in the future.

Dealing with AI in education

“The results in Spain are very similar to those in Germany and Italy, but I can’t say whether that’s good or bad,” Klaus admits. “They are using and learning the technology. But on the other hand, there are more risks in the education sector, which needs to develop some methodology to deal with AI.”

He also points out that a survey in Lithuania conducted in November gave “lower” results, although he believes that if it were repeated now “it would show something similar to Spain and the rest of the countries.”

Klaus confesses that artificial intelligence was already used in the student field “even before ChatGPTthat some models already existed. But it was not so widespread and the students were not aware of it nor was it a matter caliente for them. Everything changed with ChatGPT because there was a big marketing campaign.”

“It also changed for us, for train our model”, he continues. “Fortunately, we had a history since 2011, with texts that we knew were of human origin. In the papers current ones you don’t know if they really are.”

That moment meant to them “work 16 hours a daywith eight hours to sleep, training and training the models. I remember it was summer and we spent every day creating a logic to do it, getting the knowledge and testing everything, so that it would fit the scientific papers.”

Since not all services are free, Klaus clarifies that prices may depend on each country: “We have some discounts and, for example, In Ukraine, access is completely free.. For universities in countries where we have more clients, it is cheaper than in those where we have fewer, due to the costs of maintaining the languages.”

Prices according to country

When human editors are involved, “it is a paid service,” because they have to charge for their work. The price “depends on the length and is also different depending on the country, but the general rule is around ten euros per page or something like that”, Klaus specifies.

On the other hand, although Identific does not have offices in the 120 countries in which it is being used, what it does do is “a search for Partners that customers can talk toin some countries where we cannot work remotely. For example, in the Philippines, India, Indonesia… where we cannot participate in public tenders without having a presence there.”

For plagiarism detection, the intelligence model “is its owncreated with a logical tool that is also their own”, although they use Google’s infrastructure to train it.

Another important issue is the management of the intellectual property: “For training we have policies that allow us to use texts for statistical purposes, in which the author cannot be distinguished. And as for universities, we let them decide whether something is correct or not. We do not decide whether a part [de contenidos] violates the copywright. Universities have different opinions about what constitutes correct usage.”

Klaus highlights the usefulness of his tools in all kinds of industrial areas. Even in the field of Innovation and patents“They can show the similarity between ideas, the similar patterns, because we have a large database to make quick comparisons. And we are working on being able to make the comparison between texts in different languages.”

He also stresses that his work is not limited to reviewing texts with AI. “There are parts where we have to use optical character recognition in images to extract text. Some students cheat with images inserted in the document. We also have some mathematical algorithms to compare that do not use AI.”

 
For Latest Updates Follow us on Google News
 

-

PREV Paola, a young woman who was swept away by the current in Zapopan, is located – El Occidental
NEXT Miguel Layún and David Faitelson rant against the organization of Copa America – Publimetro Mexico