iFLYTEK released version 2.0 of the Xinghuo cognitive model and jointly released the Xinghuo all-in-one machine with Huawei

"Many people of insight and industry leaders in China are very anxious, in this wave of general artificial intelligence era, is there any opportunity and hope for China? iFLYTEK from December 15 last year to launch the research and development of Xinghuo cognitive large model '1+N', '1' is the benchmark ChatGPT general cognitive large model, 'N' is the application of various industry tracks, today we not only have full confidence in the algorithm, but also in the computing power has sufficient guarantee, I think China in the key strategic link of general artificial intelligence, stand! On the afternoon of August 15th, the 2.0 release conference of iFLYTEK Xinghuo cognitive model with the theme of "liberating productivity and unleashing imagination" was held in Hefei, and Liu Qingfeng, chairman of iFLYTEK, made a passionate introduction.

Compared with the 1.0 version on May 6 and the 1.5 version on June 9, what features have been jumped in Spark 2.0?

[Code capability will greatly lower the threshold for entrepreneurship in the digital economy]

According to reports, the abilities of Spark 2.0 continue to improve, with 100 points as a benchmark, 72 points in text generation, 78 points in language comprehension, 70 points in knowledge quizzes and answers, 60 points in logical reasoning, and 72 points in mathematical ability. "After 60 points, it can empower many areas and improve work efficiency." Liu Qingfeng said.

A number of actual measurements were carried out on the spot, a welcome speech was written at the press conference, a mathematical problem involving trigonometric functions, equations and equal difference series, and chatted after being set to high emotional intelligence, high IQ, and senior foodie personality, and Spark 2.0 performed well.

After serving these "starters", let's take a look at the "code ability" of one of the "main dishes".

Code ability is a key dimension supporting the wisdom of cognitive large models, and it is a "hard-to-head" ability, which is regarded as an important indicator of the intelligence of large models. "Code capabilities will greatly reduce the threshold for entrepreneurship in the digital economy, and you don't need to be a programming master, as long as you have imagination and a deep enough understanding of application scenarios, you can start a business." Liu Qingfeng introduced.

In the field measurement, Xinghuo version 2.0 has well completed the coding tasks such as "volley handwriting", "snake mini-game", "using Python language to draw a three-dimensional diagram of the saddle face equation, and setting the gradient color". In the previous comparison of real application scenarios, the "code generation" and "code completion" capabilities of Xinghuo version 2.0 have surpassed ChatGPT, and are very close to ChatGPT in terms of "code correction", "code explanation" and "unit test generation". "By October 24 this year, the code capabilities of these five dimensions will surpass ChatGPT." Liu Qingfeng said.

iFLYTEK released the newly developed "Intelligent Programming Assistant" based on this code capability, and its code capability can be seamlessly integrated into the integrated development environment. A few weeks ago, more than 2,000 iFLYTEK employees tried the "intelligent programming assistant", with a code adoption rate of about 38%, a coding efficiency increase of more than 30%, and a 15% increase in overall efficiency.

Liu Qingfeng gave an example of a school whose academic affairs office wanted to build an application to solve the problem of irregular leave for students leaving school. Although they do not have any programming experience, the teachers of the Academic Affairs Office only need to talk to the "intelligent programming assistant", and the assistant can automatically analyze the requirements, generate data models, apply pages and processes, and quickly complete the application construction and launch. "The traditional development cycle for such an app was about 17 days, but now it can be completed in just one day through conversational development."

【New upgrade of multimodal capability】

Multimodal capability is another "main course". This is iFLYTEK's established long-term strategy for artificial intelligence technology, and it has won 17 international authoritative evaluation champions in the multimodal field in the past three years.

Through multiple tasks, the capabilities of Xinghuo version 2.0 in image description, image understanding, image reasoning, image recognition, image generation and virtual human synthesis were demonstrated on site.

In the actual measurement, a picture of a polar bear and a penguin is vividly described in version 2.0: "The polar bear is standing on the ice while the penguin is standing next to it, and they seem to be observing each other, or maybe just enjoying this cold moment." When asked what was strange about the picture, Spark 2.0 found that "polar bears live mainly in the Arctic, while penguins live mainly in the Antarctic region." Therefore, it can be speculated that the penguin may have migrated from the Antarctic region to the Arctic region, or that they may have appeared in zoos or oceanariums. "Migration from Antarctica is not reliable, but it is possible to be present in zoos or oceanariums at the same time.

Write creative copywriting for Huangshan Maofeng and create Mid-Autumn Festival blessing videos. "iFLYTEK Smart Work", which has more than 3.75 million users, has also been completely upgraded, "The traditional dubbing + editing, which takes at least one or two hours, can now be solved in 5 minutes with the help of AI." Liu Qingfeng said.

Committed to creating a "spark language companion" for speaking sparring, the number of topics has increased from 73 to 393, intelligent evaluation has been added to the speaking mock test, and the situational communication also supports picture/document custom scenarios.

[Developers increased by 282% year-on-year]

In Liu Qingfeng's view, the key elements of the deep application of cognitive large models in the industry are: safe and controllable, scenario-driven and exclusive models. The current content security issues of the cognitive big model are mainly dirty corpus and "a serious nonsense", and iFLYTEK, which is responsible for the construction of the "National Engineering Technology Center for Speech and Language", has formed a three-dimensional guarantee mechanism for this. In terms of computing power security, iFLYTEK and Huawei are jointly building a computing power cluster for ultra-large model training. At the press conference, iFLYTEK and Huawei jointly released the Spark all-in-one machine, which allows enterprises to deploy large models on the platform of independent innovation.

According to reports, the Xunfei Spark cognitive big model has promoted the vigorous development of the developer ecosystem, and the number of developers has increased by 282% year-on-year, of which the number of "enterprise services" is the largest. "Whether it is a team or an individual, you can not understand professional development experience, as long as you are creative and imaginative, understand user needs, and gain insight into customer needs, you can generate the required applications with one click through our platform, which greatly reduces the threshold of innovation and entrepreneurship."

"When it comes to the future, whether you think it works or not, you're going to be right." Liu Qingfeng quoted a famous quote to express his determination and iFLYTEK, "Although the Xinghuo general model has many capabilities that are already stronger than ChatGPT, but overall it is still benchmarking against it, October 24 will be a milestone node, Xinghuo's Chinese ability will surpass ChatGPT, and the English ability will be comparable to it." Next year, we will benchmark against GPT4. ”

Source: Author: Huang Haihua