Shulou(Shulou.com)11/24 Report--

On October 24, on the 1024 Global developers' Day, iFLYTEK released iFLYTEK Spark model V3.0 as scheduled. In the words of iFLYTEK Chairman Liu Qingfeng, iFLYTEK Spark 3.0 seven capabilities continue to improve, surpassing ChatGPT as a whole and surpassing GPT-4 in the field.

IFLYTEK Spark can be said to be an alternative among a large number of domestic models. It is not the first release, but it is the only large model product that gives a clear upgrade node, and the iteration time is even accurate to the date. From the release of Spark V1.0 on May 6 to iteration 2.0 on August 15 and the update of Spark V3.0 on October 24 as scheduled, it took only 5 months to complete the transcendence of ChatGPT.

The National Research Institute of Economics of the Development Research Center of the State Council compared Spark V3.0 with other large model products, and gave the evaluation that "the comprehensive ability of spark model surpasses ChatGPT, leading in China and first-class in the world".

Is Spark V3.0 so powerful?

Needless to say, it's a mule or a horse. Today we're going to test the ability of Flying Spark V3.0.

IFLYTEK stressed the improvement of iFLYTEK Spark V3.0 in code capability in the press conference. In Liu Qingfeng's view, software defines everything in the process of industrial digitization and digital industrialization. As the cornerstone of the digital world, iFLYTEK Spark V3.0's code capability has been further improved. In the cross-platform transformation of the "smart classroom" product, it involves the migration from the Windows platform to the Linux platform, with a code volume of up to 200000. According to the traditional development method, it takes three months, but under the spark empowerment, the migration work is completed in one month.

First a simple Mini Game, using Python language to generate a gluttonous snake game code.

IFLYTEK Spark V3.0

ChatGPT:

Both can generate a usable piece of Python code in a very short time, with little difference.

Here's another question and answer for the code:

ChatGPT:

IFLYTEK Spark V3.0

Both iFLYTEK Spark and ChatGPT correctly answered the question of testing code-to-natural language conversion, and both gave line-by-line explanations. ChatGPT explains the purpose of the code, but iFLYTEK takes the code more seriously, pointing out that it lacks the closing parentheses of the function definition.

At the press conference, iFLYTEK also released iFlyCode2.0, an intelligent programming assistant based on iFLYTEK Spark V3.0. According to Liu Qingfeng, iFlyCode2.0 improves efficiency by 50% in the design phase, 37% in the development phase and 44% in the testing phase, greatly improving the efficiency of software practitioners.

The improvement of code capability is only one of the capabilities of iFLYTEK Spark V3.0 upgrade. IFLYTEK Spark 3.0 surpasses ChatGPT in Chinese language capabilities. Let's compare the differences between the two in Chinese content production.

For example, we asked both of them to draw up an outline of a recent interview with Musk.

ChatGPT:

IFLYTEK Spark V3.0:

In comparison, iFLYTEK Spark's outline is more substantial, giving more professional and detailed questions on some key issues, such as Tesla and the electric vehicle industry, such as when it comes to the construction of charging facilities, the depth and breadth of the interview outline are guaranteed, rather than just listing some possible directions of the interview, like ChatGPT.

According to iFLYTEK's understanding, a big model should not only be able to answer questions, but also be able to ask questions, not only to answer, but also to ask questions actively like people.

This is about the progress of iFLYTEK Spark V3.0 in active interaction. Liu Qingfeng said that they have further improved the ability of multiple rounds of dialogue on iFLYTEK's version 3.0, so that the big model is no longer just a simple information consultation, but can ask questions more clearly in multiple rounds of questioning.

This feature is used more deeply in iFLYTEK's newly released iFLYTEK Spark model. In fact, as early as 2017, iFLYTEK Intelligence Medical assistants had already passed the National qualification examination for Medical Practitioners, ranking more than 96.3% of human candidates, and already provided services to doctors in grass-roots and grade hospitals. However, due to the particularity of the scene, iFLYTEK is very cautious in providing this service to the public.

Now, thanks to advances in active interaction in iFLYTEK Spark 3.0, iFLYTEK Spark 3.0 has surpassed GPT-4 in the field, with the release of Feixiao Doctor App. In the demonstration in the press conference, Liu Cong, director of iFLYTEK Research Institute, specially demonstrated the use of iFLYTEK Xiaoyi, which can take the initiative to ask questions, actively enrich patients' background information, and judge patients' condition through multiple rounds of conversations. IFLYTEK calls it everyone's AI health assistant.

In terms of actual measurement, we use the simplest question of "stomachache" to ask ChatGPT and Xunfei Xiaoyi respectively.

ChatGPT:

Although ChatGPT's answer lists a lot of advice, it does not give particularly effective medical information.

Doctor iFLYTEK:

IFLYTEK Xiao Doctor is much ahead in this respect. First of all, it pertinently understands the basic personal information of the patient, and then actively asks the patient about the characteristics, duration and other adverse symptoms of the pain. It was only after many rounds of dialogue that iFLYTEK Xiao Doctor gave a very detailed suggestion, including the possible illness, what department of the hospital is needed, and so on.

It feels like a face-to-face communication with a doctor. For people who usually go to the hospital and don't know what department to hang up, this AI health assistant is really powerful!

In addition, Spark 3.0 brings another major upgrade, the personalized assistant feature. To put it simply, you can give iFLYTEK Spark V3.0 any personality you can imagine, making it have a personalized expression style and emotion. You can even import specific knowledge and historical chat records anytime and anywhere to make the big model form a personalized memory.

It can be said that this is basically all of the personal AI assistants.

According to Liu Qingfeng, within 48 hours after the launch of iFLYTEK friends, the first batch of users have pinched out more than 3000 people, which is evident in their pursuit of personalized Ai assistants.

I have to marvel at the strange imagination of iFLYTEK users, for example, here you can talk about the feud between Autobots and Decepticons with "wasting Brother" in Transformers.

You can also share "the singing of burial flowers" with Lin Daiyu.

Conclusion: iFLYTEK Spark V3.0 deserves to be the first echelon of domestic big models, surpassing ChatGPT is not verbal, but really achieves the lead and transcendence.

But we can't be complacent either. At the iFLYTEK press conference, Liu Qingfeng also mentioned that the gap between domestic large models and GPT-4 must be faced squarely. "domestic large models still lag behind GPT4 in complex knowledge reasoning, rapid learning with small samples, super-long text processing, and unified cross-modal understanding." He said that only the scientific spirit of seeking truth from facts can truly achieve transcendence.

Fortunately, the spark big model training for the larger parameters of the standard GPT-4 has officially started, and iFLYTEK plans to achieve the standard GPT-4 in the first half of 2024.

Let's wait and see!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.