MOSCOW (Pakistan Point News / Sputnik - 19th May, 2023) More than one million people have used a new generative neural network, developed by Russian IT giant Yandex, tentatively called YandexGPT, as part of the Alice virtual assistant functionality, the company said on Thursday.
On Wednesday, Yandex announced it became the world's first to introduce a new-generation large language model into its virtual assistant, Alice.
"Thanks to the Alice neural network, it (YandexGPT model) learned to write texts and offer ideas almost as well as a person who understands the topic ... More than 1 million people used the YandexGPT neural network in Alice in a day," the company said in a statement.
The new feature is available in the mobile application, Yandex browser, stations and smart TVs with Alice VA installed.
"The virtual assistant will write a script for graduation speech, a business letter, offer a travel plan or options for a wedding gift. Though Alice is only learning and can make mistakes in the facts. However, this does not prevent the virtual assistant from creating something new," the company stated.
The YandexGPT neural network embedded in Alice was trained on Yandex supercomputers, the most powerful in Russia and Eastern Europe. The training took place in two stages. In order for the neural network to absorb knowledge about the world, it was first fed publicly available texts such as books, websites, articles, selected using Yandex search technologies.
Unlike previous language models, the YandexGPT neural network was trained on hundreds of thousands of examples of meaningful and well-written answers. To collect and prepare such examples, Yandex used its own technologies and a team of professional AI trainers.
Earlier in the day, OpenAI, the developer of the ChatGPT language model that started the AI chatbot boom, announced it had launched the ChatGPT app for iOS. It also promised to release the app for Android devices soon.
ChatGPT, launched in late November 2022, was met with mixed reactions due to its ability to mimic human conversations and generate unique texts based on users' prompts. Some praised the model for its professional applications, such as code writing, while others criticized it for potential abuse, for instance, in education.
