Deep Sick develops artificial intelligence models that improve their self -performance

Deep Sik works with the University of Tsinghua to reduce the training needed for artificial intelligence models in an attempt to reduce operational costs.

After the emerging Chinese company “Deep Seck” rocked the markets by offering a low -cost logical reasoning model in January, it cooperated with researchers from the educational institution in Beijing in a research paper dealing with a new approach in reinforcement learning to improve the efficiency of models.

The new approach aims to help artificial intelligence models to increase the commitment to human preferences by providing rewards in exchange for the most accurate and understandable answers, according to what the researchers wrote.

Reinforcing learning has proven effective in accelerating the pace of artificial intelligence tasks in limited applications and fields. Nevertheless, some challenges have emerged in front of the expansion of its use to include more public applications, the problem that the “Deep Seck” team seeks to solve through what it called “Self-PRINCIPLED CRITIQUE tuning”.

This strategy surpassed the current methods and models in many standard standards, and the results showed improved performance with the use of lower computing sources, according to the research paper.

Artificial intelligence develops self
Deep Sick calls these models “GRMM”- a shortened publication models- and will be offered by open source, according to the company. The other artificial intelligence developers, such as the “Ali Baba Holding”, a giant technology and “Openai”, which is based in San Francisco, is to explore new horizons in enhancing the capabilities of logical reasoning and self -development of artificial intelligence models while performing tasks in actual time.

“Mita Platforms”- its headquarters in Menlo Park, California- Lama 4, launched its latest collection of artificial intelligence models during the weekend, and indicated that it is the first modal of its development using the “Mixture of Experts” structure.

Deep Sick models are largely dependent on the structure of the “Mix Experts” to increase the effectiveness of resources, while “Mita” compares its newly released models with those developed by the Chinese start -up company located in Hangzhou. Deep Seck has not set the possible date for launching its new primary model.