Next gen cybersecurity paradigm towards artificial general intelligence: Russian market challenges and future global technological trends

Next gen cybersecurity paradigm towards artificial general intelligence: Russian market challenges and future global technological trends

연구 분야: Artificial Intelligence

학회: Journal of Computer Virology and Hacking Techniques

The growing distribution of large-language models (LLMs) shifts a cybersecurity paradigm. Typical emergent abilities of LLMs, such as in-context learning, instruction following, and step-by-step reasoning, which have not been presented in smaller models, enable LLMs to infer how to perform a new downstream task from a few examples in the context without training, follow the instructions for new tasks without using explicit examples, and solve many complex math tasks and reasoning problems. And one of the effective ways to enhanced the LLMs reasoning ability to solve complex tasks performance is prompt engineering, which allows to develop and optimize prompts for LLMs to return the better solution. Using LLMs for cryptography instead of prescribing specific cryptographic algorithms assumes end-to-end adversarial machine learning based only on a secrecy specification represented by the training objectives; as a result LLMs can learn to use secret keys to protect information from other neural networks, learn how to perform forms of encryption and decryption, and also how to apply these operations in order to meet confidentiality goals. The article deals with the description of the role of LLMs for cybersecurity, along with the knowledge gaps involved in technology, and challenges issues that require detailed consideration for future prospects. At the same time, the authors consider LLMs as a transition stage on the path to artificial general intelligence.