At the GigaConf 2024 technology conference, Sber and SberDevices introduced the GigaChek AI detector, which can detect texts written by a neural network. Sber press service reports this.
As Sergey Markov, General Manager of the Experimental Machine Learning Systems Department at SberDevices, said, GigaCheck can determine with high accuracy whether a text was written by a person or an artificial intelligence system.
According to him, 220 thousand texts from news sources were analyzed in the last month using the solution. It was found that an average of 6% wrote generative neural networks. He also noted that more than a third of the content published in some major publications is written this way.
“Until now the possibility of creating a text detector that worked like this was thought to be extremely low. Many experts considered this task practically unsolvable. Developing such a service requires serious resources and large amounts of data. But even unlimited resources may not be enough,” Markov added.
GigaCheck is extremely accurate and fast, is available via API and can be integrated into any third-party application. Working accuracy is 94.7%.
Markov emphasized that the detector of OpenAI, which is considered one of the leaders in the development of artificial intelligence systems, could not achieve high accuracy, which makes the development of Sber important.
GigaCheck now offers a generalized assessment of authorship. If a text contains both human-written and AI-generated parts, authorship is determined based on the dominant content. In the near future, GigaCheck will also be able to detect which text parts were created by LLM.