The most powerful domestic models in 2023, these top four.

ChatGPT is a popular model of general artificial intelligence. According to incomplete statistics, as of August, 2023, there were as many as 156 large models published in China alone. Among these large models, there are some excellent ones and some others who just make up the numbers.

According to the Artificial Intelligence Large Model Experience Report 2.0 released by the Enterprise Development Research Center of Xinhua Research Institute in August last year, the large model evaluation report released by MIT Science and Technology Review, and the Artificial Intelligence Large Model Experience Report 3.0 released by the Enterprise Development Research Center of Xinhua Research Institute in December, at present, the top four major domestic models are as follows:

Evaluation Results of Artificial Intelligence Big Model Experience Report 2.0

Evaluation results of MIT Science and Technology Review

Evaluation Results of Artificial Intelligence Big Model Experience Report 3.0

1. Iflytek Spark Cognitive Model

In the "Artificial Intelligence Big Model Experience Report 2.0", it ranked first with the comprehensive energy score of 1013;

In the big model evaluation report of MIT Science and Technology Review, he won the title of "the smartest" big model with a score of 81.5;

In the "Artificial Intelligence Big Model Experience Report 3.0", it won the championship with a high score of 1775, and won the first place in three evaluation indexes: basic ability index, IQ index and tool efficiency index.

Iflytek Spark Model was released on May 6, 2023, and it has seven core abilities, such as text generation, language understanding, knowledge question and answer, logical reasoning, mathematics, code and multimodal. Since then, the Spark model has been upgraded and iterated many times, and the latest version is 3.0.

In October, in order to deeply understand the practical application of related large-scale models, the National Research Institute of the State Council Development Research Center conducted a research on the evaluation of the application ability of large-scale models in different industries, aiming at evaluating the performance of large-scale models in different industries, analyzing potential application scenarios and modes, comparing the industry performance of China’s large-scale models with that of world-class large models, and on this basis, making policy recommendations for the high-water development of China’s large-scale model industries.

The selected big models include: Spark Big Model 3.0, ChatGPT, GPT-4 and a domestic big model.

In the evaluation industry, knowledge-intensive production service industry (service and industrial design), life service industry with high individual demand (medical care, education and retail) and some manufacturing industries (automobile engineering and computer) are selected, and evaluation questions are constructed based on authoritative professional qualification examinations such as clinical practitioner qualification examination, medical practitioner qualification examination, unified professional qualification examination, professional qualification of motor vehicle inspection and maintenance technicians, and national computer technology and software professional qualification, so as to evaluate the knowledge and skills of the big model in the industry.

The evaluation results show that the comprehensive ability of Spark model has reached the first-class level in the world, and its performance in all seven evaluation industries has greatly surpassed that of ChatGPT, and it is superior to GPT-4 in some industries, and it also performs well in comparison with a large domestic model.

Especially in medical care, education, the performance of iFLYTEK Spark Model is particularly outstanding. The knowledge and language comprehension ability in Chinese have surpassed GPT4 by 5.3% and 4.1% respectively, and the performance of basic ability in education is less than 1%.

2. Baidu ERNIE Bot

In the big model evaluation reports released by Artificial Intelligence Big Model Experience Report 2.0 and MIT Science and Technology Review, Baidu ERNIE Bot ranked second with scores of 1010 and 72.5 respectively, which was second only to iFLYTEK Spark Big Model.

It is understood that Baidu ERNIE Bot was released on March 16th, and its main abilities include literary creation, commercial copywriting, mathematical logic calculation, Chinese understanding and multimodal generation.

Li Yanhong, founder, chairman and CEO of Baidu, said at the press conference that although ERNIE Bot is not perfect in internal testing, there is demand in the market and it must be launched.

ERNIE Bot’s technical principle is based on natural language processing technology and deep learning technology. It learns the knowledge of sentence patterns, language habits, rhetorical devices, text structure, tense voice, etc. through the establishment of a massive corpus and depth learning algorithms, so that it can automatically generate paragraphs and sentences that conform to grammar and semantics according to the keywords and needs input by users.

The realizable functions are similar to those of Iflytek Spark and Shang Tang Consulting. Not only can you chat with users, but you can also quickly generate and process various text contents, such as articles, poems, lyrics, codes, pictures and so on.

On October 17th, 2023, ERNIE Bot was upgraded to version 4.0, and the four abilities of understanding, generation, logic and memory were improved.

threeShang Tang to discuss.

In the Artificial Intelligence Big Model Experience Report 2.0, it tied for third place with Zhipu AI-ChatGLM, with a comprehensive score of 983, and ranked third in the big model evaluation report published by MIT Science and Technology Review, with a score of 68.3. It scored 1746 in "Artificial Intelligence Big Model Experience Report 3.0", ranking second, and ranked first in all 10 big models in the dimension of quantitative measurement of emotional intelligence. In the evaluation, it selected the future quadrant of big model market together with iFLYTEK Spark Big Model and Zhipu AI-ChatGLM.

The meaning of "consultation" is "consultation can solve all problems".

"Consultation" actually points out the core of the ChatGPT model, that is, users can cultivate the problem-solving ability of the big model through "magic" in multiple rounds of dialogue.

It is reported that Shang Tang Consulting was officially launched in April this year, which is one of the earliest products in China based on the 100 billion-parameter large language model, and has been iteratively updated.

Around the Big Model and R&D System Based on SenseCore Shang Tang AI equipment, Shang Tang established an engineering system to support the rapid iteration of the big model. The software, system and hardware are all doing R&D cooperation with the goal of serving the rapid iteration of the big model, realizing agile, low-cost and large-scale model iteration, which is helpful for Shang Tang to find the best and most effective production formula of the big model.

4. Zhipu AI-ChatGLM

In the Artificial Intelligence Big Model Experience Report 2.0, it tied for third place with Shang Tang, with a comprehensive score of 983, and ranked third with a score of 1729 in the Artificial Intelligence Big Model Experience Report 3.0. In the final evaluation, it was selected into the future quadrant of the big model market together with iFLYTEK Spark Big Model and Zhipu AI-ChatGLM.

ChatGLM is an open source and bilingual ChatGPT-like language model developed by a team led by Dr. Zhang Peng, a leading innovation engineer in Tsinghua University. It was released on March 18th.

On October 27th, Zhipu AI launched a brand-new ChatGLM3 model at the 2023 Computer Conference (CNCC). This latest model adopts innovative multi-stage enhanced pre-training method to achieve more adequate training.

Zhang Peng, CEO of Zhipu AI, personally released the new product and showed the latest features of the product through real-time demonstration. According to reports, through the use of richer training data and optimized training programs, the performance of ChatGLM3 has been significantly improved. Compared with the previous ChatGLM2, MMLU has increased by 36%, CEval by 33%, GSM8K by 179% and BBH by 126%.

In addition, ChatGLM3 has been iteratively upgraded for GPT-4V, and some new functions have been introduced. Among them, CogVLM-, which has the ability of multimodal understanding, has achieved a leading position in more than 10 international standard graphic evaluation data sets; Code Interpreter, a code enhancement module, can generate and execute codes according to users’ needs, and automatically complete complex tasks such as data analysis and file processing. Web search enhancement module WebGLM- Access search enhancement, which can automatically find information related to questions on the Internet, and provide links to related documents or articles when answering.

Summary:

Although in 2023, domestic large-scale models developed by leaps and bounds, and many benchmark products emerged, which performed well in industrial applications, we must also see that the development of artificial intelligence technology is a continuous process, which requires continuous technical research and development and innovation.

Exploring the landing mode of empowering different industry scenarios will become the direction of the rapid development of large-scale model enterprises in China, and the industry application value will also become the core index for judging large models. Only when there are thousands of AI native applications that solve the real needs of production and life, can the big model really penetrate into every corner of the social economy, help the industrial upgrading of all walks of life, promote the rapid economic recovery, and profoundly change people’s lifestyles.

According to public data, the global artificial intelligence will continue to grow at a high speed, and the market scale will continue to expand. It is estimated that by 2032, the global artificial intelligence market will reach 1.3 trillion US dollars.

Under the background of full of vitality and opportunities, the rise of artificial intelligence big model will continue to lead the progress of science and technology, and it will play a role in technological innovation, commercial application and industrial transformation. With the continuous expansion of application fields, it brings new possibilities for the development of various industries.

Disclaimer: The market is risky, so you should be careful when choosing! This article is for reference only, not as a basis for trading.

Article complaint hotline: 182 3641 3660 complaint email: 7983347 16@qq.com.