DeepSeek upgrades its AI model for math problem solving

The Chinese Ai Lab Deepseek has tacitly updated RehearseHis AI system, which has developed the solution of math evidence and theorems.

Accordingly South Chinese morning postDeepseek invited the latest version of Prover, V2, to the AI ​​Dev Platform late Wednesday. It seems to be built on the V3 model of the startup, which contains 671 billion parameters and assumes a mixture of experts (MEE). The parameters roughly correspond to the problem -solving skills of a model, while Moe divides tasks into subcontracts and delegates them to smaller, specialized “experts” components.

Deepseek last updated in August in August and then described it as a custom model for formal theoretical sample and mathematical argument.

In February, Reuters reported This deepsek, which recently published an improved version of V3, a general model, and is expected to update its R1 model “Argumenting” shortly, should be considered for the first time to increase outside of the financing.