Experts of the online project "Call me back" gave advice on identifying AI content

The content created by users using artificial intelligence (AI) technologies, both for harmless entertainment and for conducting large-scale scams, can be recognized by special programs or an attentive recipient following a number of tips. This was announced on April 27 by experts of the online project "Call me back", created in 2022.

"In most cases, users post AI content online for entertainment, for example, they create social media accounts of people who do not exist and post generated images to them. But attackers can also use technology to deceive, mislead or commit a crime," the experts are quoted as saying on the website. mos.ru .

It is emphasized that people who plan to break the law often use neural networks to generate videos that feature acquaintances of a particular victim or well-known people. To detect fraud, the recipient of such content needs to pay attention to the details. In such clips, as a rule, the movements of the lips of the person depicted on them do not always coincide with the words he actually utters, and his pupils do not react to light and do not focus on any objects, experts warned.

The person in such an AI video may have an ideal face, unnatural fingers, move slowly and broadcast emotions that do not match the atmosphere of the message itself. Against its background, objects often disappear or appear, and the proportions themselves are sometimes violated.

In order to recognize an artificial photo to the viewer, you also need to pay attention to the look of the person depicted, fix strange and atypical shadows, as well as uncharacteristic strands of hair or folds of clothing. AI generators, creating such materials, may make mistakes in the signage in the photo or "think out" something that should not have been on them. After receiving such a file, it is important to analyze the metadata as there is no information in the neuro—photo about which camera the picture was taken on, and there is also no data about the file parameters and the date when it was created.

Speaking about AI texts, the staff of "Call me back" stated that such materials often lack logic, and the paragraphs repeat the same thought. In addition, AI, even despite imitating human emotions with the help of large language models (LLM, Large language model), is not yet able to defend its position. Such publications contain common metaphors and comparisons, as well as unique real-life examples.

The article added that special programs can help in the process of recognizing such content. Of course, they do not guarantee a completely accurate result, but they allow you to get a preliminary assessment of the authorship of a particular message.

Shah and Mate: the new smartphone relies on realistic shooting

What is interesting about the Huawei Mate 80 Pro flagship

Representatives of Roskachestvo on April 24, speaking about the breakthrough that is being recorded in technologies for creating deepfakes, pointed out the difficulties it generates. So, according to their observations, in order to generate a specific character, people interested in this began to resort to several programs at once. They "glue" the image of a person in several stages, and then sometimes use it for commercial and criminal purposes. The authorities recommended that Russians who do not want to be misled come up with a code word with which to disable an AI copy of a person.