‘Start’ of the Deepseek synthetic intelligence Has introduced a new way to improve the rational ability of large linguistic models (LLM) to provide better and fast results In front of its competitors.
In January, Deepseek caused a frenzy that appeared in the scene with an artificial intelligence model (AI) R1 with a sattpot, according to the company, cheap and worked SU Principal Competition, Chaddjd de Opanai.
Together with the Chinese University researchers, its states of Tepsek The last article was published last Friday It has created a technique to improve AI models. AI provides basic technology training to create your own rules to determine the content, and then use them to improve your answers.
Generally, it is important to increase the size of the samples during training to improve the AI, which requires a lot of human effort and computational power. Instead of, Depsek has created a system with a attached ‘judge’ It assesses AI answers in real time. When a question is asked, the judge compares the expected response of AI with the basic rules of AI and must be a good answer. If the answer is similar, AI receives a positive answer It helps improve.
Deepseek calls the Deepseek-GRM automatic development system. Researchers confirm that this will help models to do better than competitors such as Google Gemini, Lama de Meta and Opanayin GPT -4O. Tepsek plans to provide these advanced artificial intelligence samples As an open raw softwareBut it did not provide the deadline.
The output of the document occurs when rumor is spreading as a texel Is going to reveal your latest Satbot R2. The company has not made any public comment on this.