{"id":33960,"date":"2023-04-11T16:57:42","date_gmt":"2023-04-11T13:57:42","guid":{"rendered":"https:\/\/hitmedia.su\/?p=33960"},"modified":"2025-10-13T20:57:06","modified_gmt":"2025-10-13T17:57:06","slug":"gonka-mozgov","status":"publish","type":"post","link":"https:\/\/hitmedia.su\/en\/2023\/04\/11\/gonka-mozgov\/","title":{"rendered":"Brains’ Race"},"content":{"rendered":"Reading time<\/span> 11<\/span> minutes<\/span><\/span>

According to\u00a0research<\/a> conducted four years ago by MarketsandMarkets, by 2024, the size of the neural networks market was expected to grow to nearly $300 million. In early 2023, media outlets had already published news<\/a> that Microsoft would invest $10 billion in a chatbot developer based on artificial intelligence – OpenAI. The number of various neural network services is steadily growing, and there is an<\/a> increasing<\/a> discussion<\/a> in the media space about what professions they can replace. In this article we will tell about the most interesting applied abilities of artificial intelligence.<\/p>\n

Translators<\/h3>\n

Neural networks have long been used<\/a> in automatic translation services such as Google Translate, Yandex Translator and Microsoft Translator. However, the real breakthrough in this field was made by a small German company DeepL with its eponymous translator.<\/p>\n

Neural networks can be built based on different architectures<\/a>, each of which is better suited for a particular task. Finding the right combination of algorithms is the key to making the AI work more accurately. Most translators have been using recurrent neural networks until recently, but they have started to be replaced by a new machine learning model, Transformer<\/a>, which Google is actively using<\/a>.<\/p>\n

DeepL does not disclose its translation technology, although early campaigns reported the use<\/a> of convolutional neural networks, which are designed to recognize visual images rather than text. The official site also says<\/a> it is using some of the mechanisms of the previously mentioned Transformer architecture. The developers’ own tests<\/a> and comparisons<\/a> made by ordinary users show that DeepL does translate much more accurately because it is able to capture context:<\/p>\n

\"\u0421\u0440\u0430\u0432\u043d\u0435\u043d\u0438\u0435<\/a>
The comparison is taken from gadgets-news.ru, with the incorrect translation in bold and the inaccurate translation in italics.<\/figcaption><\/figure>\n

It is worth noting that the translator from Google has been significantly improved after the introduction of the Transformer architecture, but DeepL translates virtually error-free.<\/p>\n

Among the neural network services, there are some that can translate and dub video, removing the speaker’s voice and leaving the original noises. For example, Neurodub AI can recognize speech in 70 languages:<\/p>\n

\"The<\/a>
The working window on the Neurodub site. The video was taken from the author’s personal archive. The corrected error and the pop-up warning-line are circled in red.<\/figcaption><\/figure>\n

Sometimes the neural network makes mistakes, but they can be corrected manually. In the picture above you can see that the AI did not recognize the name of the university correctly and heard “neck” instead of “HSE”. After editing the text in the side menu, you need to click “Apply changes” and the video will be re-voiced correctly. Unfortunately, the service is paid (1 minute – 1 dollar), but you can try the first 15 minutes in test mode.<\/p>\n

Photos, pictures, logos<\/h3>\n

In March 2023, the developers of Midjourney, a popular neural network for generating images, shut down free access<\/a> due to abuse with deepfakes<\/span>3<\/sup><\/a><\/span>. Social media users spread pictures of arrested Donald Trump and Pope Francis in a fancy down jacket.<\/p>\n

\"A<\/a>
A fake photo of Pope Francis generated by Midjourney.<\/figcaption><\/figure>\n
\"The<\/a>
The fake about Donald Trump’s arrest. The photo was circulated on social media.<\/figcaption><\/figure>\n

In addition to Midjourney, there are many other services for generating images, such as Stable Diffusion, Craiyon, and Starryai. These neural networks create images based on keywords, so-called \u00abprompts\u00bb<\/span>4<\/sup><\/a><\/span>. If they are not used correctly, the AI may misunderstand the user’s request and generate something hard to recognize.<\/p>\n

Neural networks are also capable of creating logos. The principle of these services is much simpler than the previous ones: just enter the name of your brand, choose some colors and icons you like and you’re done. Here are a few examples generated by Brandmark.io and Looka:<\/p>\n

\"The<\/a>
The logos are generated based on the name of the author of the article, the theme “University” and a few pastel colors.<\/figcaption><\/figure>\n

And the neural networks at Colorize.cc and MyHeritage.com can improve the quality of a photo or even color an old black and white one:<\/p>\n

\"Photo<\/a>
Photo from the personal archive of the author of the article.<\/figcaption><\/figure>\n
\"Photo<\/a>
Photo from the Colorize website.<\/figcaption><\/figure>\n

Write a text<\/h3>\n

ChatGPT is probably the most famous chatbot that can write even a diploma.<\/a> It is based on the GPT-3.5 language model and is used to generate texts that look like their author is a human.<\/p>\n

Writing texts with neural networks has its own peculiarities: they cannot simply repeat existing phrases. AI generates new texts based on statistical parameters and patterns found in the data sets on which it has been trained. As a result, texts generated by neural networks can be original, but contain errors and inaccuracies. By the way, this paragraph is almost entirely written with ChatGPT: I only had to correct some grammar and speech errors.<\/p>\n

\"The<\/a>
The mechanics of interaction with the chatbot is based on requests and responses. It is desirable to have different topics in different chats.<\/figcaption><\/figure>\n

Some neural networks, such as Rytr, do not work like chatbots, but use a clear user-defined terms of reference. You need to select “tone of speech”, language, “content format” and enter keywords on the site. After specifying the settings, the artificial intelligence will generate a sample text, which it will offer to complete for yourself.<\/p>\n

Artificial Intelligence detectors<\/h3>\n

Due to the emergence of services that can write the text for a person, there was also a need to check the work for anti-plagiarism. The problem is that using ordinary software it is impossible to determine, for example, if a student wrote an essay by himself or herself or if he or she used AI. Moreover, The New York Times pointed out that even writers and teachers cannot distinguish<\/a> ChatGPT writing samples from student papers, so schools in New York banned the use of OpenAI’s AI<\/a>.<\/p>\n

\"The<\/a>
The result of checking the text generated by ChatGPT on the Writer site.<\/figcaption><\/figure>\n

For this purpose, special services have been developed that also work with neural networks. Some of them are AI Content Detector from Writer and AI Text Classifier from the already mentioned OpenAI. So far, however, AI detectors have low accuracy in determining neural networks work. AI Text Classifier, for example, successfully identified only 26% of AI texts<\/a>. In addition, this detector will fail<\/a> if the text is even slightly edited, written by children or not in English, and contains less than a thousand characters.<\/p>\n

Despite minor advances in this area, services that mask the work of neural networks are already being developed in parallel. This is how Stealthwriter AI handled rewriting ChatGPT’s response to my question about whether he would be offended if the author did not mention him in the article (the text is given in unmodified form):<\/p>\n

Since I am only an Al artificial intelligence modeled brain, there are no sentiments or affections inside me that can cause offense. To assist in any way possible and offer information is my purpose. The decision of mentioning my support in the article lies entirely with you. That said, recognizing and attributing sources or tools utilized in your work is a common practice for ensuring transparency and integrity.<\/p><\/blockquote>\n

Original text:<\/p>\n

As an Al language model, I do not have feelings or emotions, so I cannot be offended. My purpose is to assist and provide information to the best of my abilities. Whether or not you mention my assistance in your article is entirely up to you. However, it’s always good practice to acknowledge and give credit to the sources or tools that you use in your work for transparency and integrity.<\/p><\/blockquote>\n

Which version is more like \u00abmade by human\u00bb is a subjective question.<\/p>\n

Artificial Intelligence has indeed reached a new level: neural networks are now able to perform some tasks faster than humans, and sometimes even better. Nevertheless, they still make mistakes, and in some places, they are far from being successful at all. There is no need to be afraid of robots taking over the world. Probably.<\/p>","protected":false},"excerpt":{"rendered":"

Reading time<\/span> 11<\/span> minutes<\/span><\/span>According to\u00a0research conducted four years ago by MarketsandMarkets, by 2024, the size of the neural networks market was expected to grow to nearly $300 million. In early 2023, media outlets had already published news that Microsoft would invest $10 billion in a chatbot developer based on artificial intelligence – OpenAI. The number of various neural … Continue reading “Brains’ Race”<\/span><\/a><\/p>\n","protected":false},"author":1103,"featured_media":33962,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[77],"tags":[251,11],"class_list":["post-33960","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blank","tag-foot3","tag-top"],"_links":{"self":[{"href":"https:\/\/hitmedia.su\/en\/wp-json\/wp\/v2\/posts\/33960","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/hitmedia.su\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/hitmedia.su\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/hitmedia.su\/en\/wp-json\/wp\/v2\/users\/1103"}],"replies":[{"embeddable":true,"href":"https:\/\/hitmedia.su\/en\/wp-json\/wp\/v2\/comments?post=33960"}],"version-history":[{"count":10,"href":"https:\/\/hitmedia.su\/en\/wp-json\/wp\/v2\/posts\/33960\/revisions"}],"predecessor-version":[{"id":33975,"href":"https:\/\/hitmedia.su\/en\/wp-json\/wp\/v2\/posts\/33960\/revisions\/33975"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/hitmedia.su\/en\/wp-json\/wp\/v2\/media\/33962"}],"wp:attachment":[{"href":"https:\/\/hitmedia.su\/en\/wp-json\/wp\/v2\/media?parent=33960"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/hitmedia.su\/en\/wp-json\/wp\/v2\/categories?post=33960"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/hitmedia.su\/en\/wp-json\/wp\/v2\/tags?post=33960"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}