ChatGPT is like a "nuclear bomb", the shock wave has not only shocked the entire scientific and technological world, but also achieved the effect of breaking the circle, setting off an AI frenzy around the world. With the popularity of ChatGPT, more and more cognitive intelligence models have been developed and applied, not only abroad, but also many domestic Chinese cognitive intelligence models. On May 9, SuperCLUE, a comprehensive evaluation benchmark for Chinese general large models, was officially released. The agency uses the SuperCLUE protocol to evaluate and rank mainstream Chinese-supporting general-purpose large models on the market.
Chinese What are the evaluation criteria for general large models?
Chinese general large model benchmark is mainly an evaluation standard for the Chinese available general large model, which mainly solves the problem of Chinese the effect of the large model under the current vigorous development of the general large model, including but not limited to the effect of different tasks of these models, the comparison of international representative models and the effect of comparison with humans. A series of domestic and foreign representative models accept the ability test in multiple dimensions under this benchmark, and then the SuperCLUE evaluation list is obtained.
On May 9, the Chinese general large model comprehensive evaluation benchmark SuperCLUE was officially released, which is the authoritative evaluation community in the field of Chinese, and the overall list released by it shows that GPT-4 ranks first, ChatGPT ranks second, and Spark cognitive large model ranks third, although it shows that although there is still a gap between Spark model and GPT, it is already the leader of domestic large models.
SuperCLUE will evaluate the capabilities of the model from three dimensions
The SuperCLUE evaluation list mainly evaluates the capabilities of the model from three different unique capabilities, namely basic capabilities, professional capabilities, and Chinese characteristic capabilities. Among them, the basic capabilities include common representative model capabilities, such as semantic understanding, dialogue, logical reasoning and other 10 capabilities; Professional competencies cover secondary school, university and professional tests, covering more than 50 competencies from mathematics, physics, geography to social sciences; The list of Chinese features targets tasks with Chinese characteristics, including more than 10 abilities such as Chinese idioms, poetry, literature, and glyphs.
The Xunfei Xinghuo cognitive big model has reached the level of GPT 3.5 in terms of dialogue, encyclopedic knowledge, role simulation, computing power, semantic understanding, and logical reasoning. In terms of semantic understanding, the iFLYTEK Xinghuo cognitive large model even got a full score of 100, exceeding GPT-4.
iFLYTEK Spark cognitive large model technology empowers more industries
iFLYTEK's Spark cognitive big model already has seven dimensional capabilities, including text generation, language understanding, knowledge question and answer, logical reasoning, mathematical ability, code ability, and multimodal capability, and has been applied in education, office, automotive, digital employees and other industries.
In addition to the above fields, iFLYTEK Spark Cognitive Big Model will also empower more industries including medical, urban, political and legal, industry, etc., to meet the needs of more professional fields and extend to a broader industrial field.