Nowadays, artificial intelligence has become much more common, whether it is in drawings generated from certain words or even with texts that give us details about specific topics, even some tools have already made scripts for television episodes. It is worth mentioning that some hackers have already started using it to carry out their attacks on banks and other entities, and this is precisely because identifying texts is not the easiest thing to do.
As mentioned in a report by the technology media, a new method has been developed that allows estimating the use of advanced language models (LLM) measuring “word repetition” in scientific texts.
The researchers found that in English texts, the use of words such as “delves” and “showcasing” can indicate that a text was generated by this method. They compared the frequency of words from 2010 to 2024, observing a sharp increase in 2023 and 2024, coinciding with the popularity of the tool. It is said that at least 10% of the summaries published in 2024 were processed with artificial intelligence.
A significant increase in the use of certain verbs, adjectives and adverbs in the post-colonial era was also revealed. LLMUsing pre-2023 abstracts as a control group, a change in vocabulary choice was shown. Countries such as China, South Korea and Taiwan have a higher use of indicative words. Detecting usage is not yet that accurate, but the new method could help in certain aspects.
It is certainly a useful tool for making certain summaries, but at this point it cannot be expressed in the way that people with different thoughts would.
Via: Ars Technica
Author’s note: It is difficult to recognize this type of text, but in Spanish there are also certain details that draw attention, and when the same key word is repeated over and over again it is very noticeable.
#Researchers #investigate #identify #AIgenerated #texts