Seminars

Neural network to help you

On 15 February, another AYSS seminar was held at the JINR Scientists' Club. Founder of AiBrain and the Communication Agency "Nikitin Studio" Egor Nikitin made a report "Artificial intelligence that thinks with you".

After having interviewed the participants to understand the extent to which they use artificial intelligence (AI), Egor started his speech. "AI is a continuation of natural intelligence, a mathematical model of our intelligence," he said. "How to effectively use neural networks, how to turn AI into a universal assistant? The era of hype ends, the era of mistakes, training and competent implementation begins. You will learn the basics of the basics - how we work with neural networks." According to Egor, AI does not make people smarter. He makes those who know how to think faster. He revealed four key principles of working with neural networks. Namely: a neural network is an assistant to a professional, not a replacement; the result depends on the quality of the task; the main thing is the ability to ask questions and to set tasks; improving outcome is a strategy, not an accident.

What is the right approach to working with neural networks? In tool selection; AI skills needed to understand why AI is lying; understanding limitations and safety; ability to formulate prompt. You don't know what prompt is? I didn't know before the seminar either. This is an extended question that allows neural networks to more accurately meet the task.

"If you only use Chat GPT, you only use 15 percent of AI's capabilities," Egor said. "Neural networks develop a person, strengthen the brain." For a year and a half, neural networks have been able to access the Internet, connect a search engine that functions differently from a neural network. Neural networks can be wrong, "hallucinate", require editing. 90 percent of users work with neural networks as with search engines and their capabilities are much wider. And Egor Nikitin demonstrated it on a specific task in real time.

Those who came to the seminar obtained practical advice, learned how to draw up prompts correctly. The basic rule is to always end it with the phrase "Ask me clarifying questions". As a result, AI collects context, specifies the target, builds an accurate answer. The second is to "Call an expert", that is, to force the neural network to criticize itself. It improves the structure of the model, improves its quality. AiBrain has developed 18 such rules and designed them in the form of the "Alphabet of working with neural networks". Most use AI as a chat, professionals - as an operating environment. Egor demonstrated the use of his rules with certain examples, right at the seminar by going to Chat GPT. "I have started with it, it's known me for two and a half years. This is my intensive assistant," the speaker shared. For example, the content editor writes Egor's texts for the telegram channel, completely preserving his style, since he has previously analyzed many of his posts. This, as the speaker formulated, is the intellectual model of Egor Nikitin. "I'm not an IT user, I'm a translator. And in you and in me and even in a child, there is a colossal experience that needs to be uploaded to the knowledge base," he said. And then, you need to specify the goals and context, to determine the role of the expert, to introduce restrictions, to set the style of the answer and you get a system prompt.

"I teach in a school. I argue that neural networks do not lead to degradation, since we should improve their response every time," E.Nikitin shared his experience.

Why do neural networks sometimes "lie"? Due to lack of data, an incorrect context, an incorrectly asked question, for other reasons. Any company retrains the neural network for itself.

"Start right now," Egor urged. "AI is the strengthening of your competencies. In a year, there will be two types of specialists: those who run AI and those who obey AI." The main skill, he believes, is the ability to ask questions, not to give answers. Egor also shared his personal experience of the first use of a neural network. And as the creative director of Nikitin Studio, he said that the company began to work faster, maintaining quality. In the early years, it has won 20 percent of tenders, last year - already 65 percent.

Answering the question about the further development of AI, E.Nikitin emphasized that this is a soap bubble of the Western economy that will burst. (The same idea, confirmed by statistical data, was expressed at the Conference "Mathematics. Computer. Education" by the specialists of the Moscow Central Economic and Mathematical Institute - O.T.). Egor is more concerned about whether he himself will stop developing. Answering another question, he confirmed that Chat GPT is perfect for generating pictures. Egor gave practical advice on writing instructions and even advised how to work with scientific articles. "I am very pleased with such an active discussion," he appreciated the meeting.

P.S. No AI was used to write this article.

Olga TARANTINA