Follow ChatGPT directly? The billion-dollar AI giant is rapidly upgrading large models!

China Fund News reporter Feng Yao

Just over a month after launching the multi-modal large model for the first time, iFlytek has been non-stop upgrading its "Spark Cognitive Large Model".

On June 9, iFlytek announced new progress in its general large model, releasing version V1.5 of the "Spark Cognitive Large Model". This version has made breakthroughs in open question and answer, and has been upgraded in multiple rounds of dialogue and mathematical capabilities. , text generation, language understanding, and logical reasoning abilities have also been improved. In addition, iFlytek launched the Spark APP and equipped it with the "Spark Cognitive Large Model".

When iFlytek was released a month ago, it had planned to reach the benchmark of ChatGPT capabilities by October 24. iFlytek revealed that it plans to launch a multi-modal interactive upgrade of the "Spark Cognitive Model" on August 15.

Aiming at the "three major flaws" to be overcome

On May 6, iFlytek announced the "Spark Cognitive Large Model" for the first time. At that time, iFlytek Chairman Liu Qingfeng set a goal for it: striving to surpass GPT in Chinese and English in English by October 24 this year. reached a considerable level.

iFlytek upgraded and released the V1.5 version of the "Spark Cognitive Model" 34 days later. Liu Qingfeng introduced that this version has made breakthroughs in open-ended question and answer, and has further upgraded its multi-round dialogue and mathematical capabilities, and its text generation, language understanding, and logical reasoning capabilities have continued to improve.

Especially in terms of open-ended question and answer in a wide range of fields, the V1.5 version of "Spark Cognitive Large Model" targets the "three major flaws" that pure large model technology needs to overcome: new knowledge is difficult to update, and fact-based Q&A is easy to "take away". , historical facts and traditional classics are easy to "make up plots".

At the same time, the leap in multi-round dialogue capabilities makes the dialogue experience of iFlytek Spark more relevant to real people. Multi-turn dialogue is a traditional problem with large models, which simply means "no memory".

At the meeting, iFlytek also conducted a live demonstration of the "Spark Cognitive Large Model". When talking about the question of "What are the new trends in artificial intelligence in China?" "Spark Cognitive Large Model" mentioned that on June 3 this year, the Yangtze River Delta Entrepreneur Alliance Industrial Digital Summit released the "Yangtze River Delta (Hefei) Declaration of General Artificial Intelligence" and the "General Cognitive Intelligence Large Model Evaluation System."

In fact, the Spark model was finalized in May, and the answers given by the "Spark Cognitive Model" already include relevant policy trends in June, which also shows that the model is in a real-time update and learning state. It should be noted that the "Spark Cognitive Large Model" further reveals the current gap faced by China's artificial intelligence.

"There is no point in giving the same answers between large models and searches, but to provide constructive solutions through professional knowledge and reasoning capabilities," Liu Cong, dean of iFlytek Research Institute of HKUST, also said bluntly at the meeting. In addition, the "Spark Cognitive Model" successfully answered the mathematics and Chinese questions of this year's college entrance examination at the conference.

Next node: Multi-modal interaction is upgraded again

iFlytek plans to carry out three rounds of iterative upgrades this year, aiming to reach a level comparable to ChatGPT on October 24. In addition to June 9, the next upgrade phase is on August 15, mainly to improve coding capabilities and multi-modal interaction capabilities. Functions in multimodal areas, such as virtual human synthesis and image and text understanding, will be opened to customers in the future.

iFlytek Chairman Liu Qingfeng previously stated that iFlytek’s current coding capabilities are focused on the industrial Internet and many applications within enterprises. In the future, the goal is to allow large models to generate various codes without the need for programmers. But Liu Qingfeng also admitted that there is still a big gap between this function of the Spark model and ChatGPT, and the key function of the next upgrade is also in this area.

Liu Qingfeng revealed at the meeting that iFlytek will also explore more potential artificial intelligence technology routes in more cutting-edge fields, such as game intelligence, brain-like intelligence and neural network models.

In addition to the further improvement of the large model's own capabilities, iFlytek also released further commercial implementation progress of the "Spark Cognitive Large Model" in the fields of learning, medical care, industry, office and other fields, including the launch of Spark APP and Spark Language Companion APP.

At the same time, iFlytek has further targeted the segmented fields and launched Spark Cognitive Model, medical post-diagnosis management platform, Spark Cognitive Model, industrial Internet platform, Spark Cognitive Model, and iFlytek smart screen products. According to industry insiders, this move is intended to promote its commercialization in subdivided fields. The scenarios that are expected to be the first to break through are the above-mentioned medical, industrial manufacturing and office fields.

At the same time, in addition to developing demonstration application products for different application scenarios, iFlytek's Spark Ecosystem, which targets AI developers, large model upstream and downstream enterprises, and entrepreneurial teams, is also simultaneously recruiting ecological partners.

In fact, referring to the development history of OpenAI, the premise of large model development is that the development, training, and application of small models are mature enough. When OpenAI was founded, its products were only vertical small models in the gaming field. After thoroughly understanding the development and implementation of small models, we continued to expand the number of parameters and finally formed the large model GPT3 with 175 billion parameters.

Domestic large models start the "Battle of Hundreds of Models"

Since March this year, domestic general-purpose large models have been released one after another. Among them, Baidu was the first to publish Wen Xinyiyan, and Alibaba followed closely behind and officially announced Tongyi Qianwen. Even scientific research institutes such as Tsinghua University, Beijing Zhiyuan Artificial Intelligence Research Institute, and Shanghai Artificial Intelligence Laboratory also released their own AI universities. Model results.

According to statistics from relevant research reports of Minsheng Securities, at least 30 large models have been unveiled in China. The producers include Internet giants, AI concept listed companies, leading server companies, scientific research institutes and primary market startups, including The parameter scale of the large model is close to or even exceeds the scale of ChatGPT (hundreds of billions).

IDC forecast data shows that China’s artificial intelligence market expenditure will increase to US$14.75 billion in 2023, accounting for approximately one-tenth of the global total. In the long run, innovative iterations of AI technology drive the further implementation of application scenarios, and hot topics represented by AIGC, digital humans, multi-modality, AI large models, and intelligent decision-making bring more imagination and possibilities to the market.

IDC predicts that China’s AI market will achieve a market size of US$26.44 billion in 2026, and the five-year compound growth rate (CAGR) from 2021 to 2026 will exceed 20%. CITIC Construction Investment believes that the domestic upsurge in R&D and application of large models continues to rise, and the development of large models has accelerated across the board. However, the current implementation of the global large model industry is still in the early stages of exploration, and it is necessary to cooperate with downstream scenario companies to establish large model business models.

Editor: Captain

Review: Xu Wen

Copyright Notice

"China Fund News" owns the copyright to the original content published on this platform. Reprinting without authorization is prohibited, otherwise legal liability will be pursued.

Just now, Fang Xinghai made a big statement!

The above is the detailed content of Follow ChatGPT directly? The billion-dollar AI giant is rapidly upgrading large models!. For more information, please follow other related articles on the PHP Chinese website!