iFlytek and Huawei announced large-scale model to benchmark against GPT4

At the iFlytek Global 2024 Developer Festival, iFlytek announced the launch of a new independent and controllable large-scale model computing power in China, which is based on the Huawei Ascend ecosystem. The aim of this product is to start large-scale model training against GPT-4.

The onstage test reveals in terms of the English task, the scores of iFlytek Spark Model 3.0 and ChatGPT3.5 were 85.1% and 84.3% respectively. The current Spark Model 3.0 has already achieved the accuracy of ChatGPT 3.5 . of transcendence.

Chairman of iFlyTek, Liu Qingfeng said that the new Spark solution has a certain lack of advancements as compared to GPT-4. This includes complex knowledge reasoning, rapid learning with small samples, ultra-long text processing, and cross-modal unified understanding.

However, Liu said that the company needs to benchmark against GPT in order to find out those differences. He also shared the future roadmap of the Spark ecosystem and said that the company will fully benchmark against GPT-4 in 2024 with the release of iFlytek Spark 4.0.

Xu Zhijun, Vice Chairman and Rotating Chairman of Huawei Technologies also delivered a speech on the spot and said that Huawei’s mission and vision are to bring the digital world to every person, family, and organization, build an intelligent world, and build a second choice for the world.

Xu Zhijun pointed out that iFlytek is a pioneer in artificial intelligence and has led the world in intelligent speech recognition technology. According to him, the voice technology of all smart terminals currently used by Huawei around the world comes from iFlytek.