After the tremendous success of Deepseek Ai and Manus AI, China is now developing the “Self -Learning and Self -Improving” model of artificial intelligence (AI), which will not require human intervention in updates and maintenance. The frightening concept can be directly of the science fiction dust, a reality in the near future, given the step of China in the artificial intelligence race.
Earlier this year, China disrupted the artificial intelligence industry by launching the famous artificial intelligence of artificial intelligence, as it acquired global attention due to its cheap operating rates. Now, Team Deepseek has collaborated with Tsinghua University to upgrade the artificial intelligence model, by converting it into a well -known average machine that does not require any human being for help.
According to reports, researchers from Deepseek and experts from Tsinghua University, a high school in Beijing, wrote a paper about a new way to use reinforcement learning, a method in which AI learns to obtain rewards for favorable answers, to make artificial intelligence models work better and faster while using a less computer capacity.
What is China’s self -improvement?
Improving artificial intelligence is an artificial intelligence model that can improve in what it does alone over a period of time. Instead of the need to constantly update human beings, this type of artificial intelligence learns from its mistakes, then discovers how it can improve and set itself to perform tasks more accurately or efficiently.
It is like how a person can practice and improve something always without the need for a teacher.
The company stated that it is working on a new way to make artificial intelligence models as humans want. The company is implementing a technology called reinforcement learning, which has worked well to train artificial intelligence to carry out specific tasks, such as playing games or solving small problems. But it was difficult to use this method for the largest and most general tasks.
To address this, Deepseek comes to something called “self -control”. This artificial intelligence technique helps to improve itself by discovering what is true and what is wrong.
According to ReportsThe search paper showed a new approach that has proven to be better than other methods of tests by providing better results while using less computer energy. In simple phrases, it is similar to teaching artificial intelligence to learn more intelligent and faster with less effort.
The new Deepseek model of artificial intelligence, Deepseek-Grm, is designed to learn and improve in a wide and flexible way by obtaining rewards to do a good job. The company plans to share these forms with everyone for free by making them open source so that anyone can use, study or modify them without the need to pay or request permission.
However, Deepseek did not say exactly when they would launch their main model of artificial intelligence. They remain the date of release is not clear at the present time.
Moreover, American technology Openai also plans to launch O3 and 04-MINI AI models within two weeks. San Francisco’s giant alibaba is trying to improve the artificial intelligence model after launching the new artificial intelligence model, after launching the AI model called QWEN 2.5-MAX, in January, which claimed to be better than Openai GPT-4O and Deepseek-V3.
Recently, on April 4, Meta also launched the latest open linguistic style (LLM), called Llama 4.
Also read: What is the Llama 4 WEIGHT AI model that Meta launched?
