Catch up with ChatGPT? Don't worry, set up a flag for iFlytek

This article is the original content of the signature account of the NetEase News NetEase Featured Content Incentive Plan, Mr. Dao Youli. Without authorization from the account, unauthorized reproduction is prohibited

On the domestic model track, another contestant, iFlytek, has arrived.

On May 6th, iFLYTEK held a press conference and officially launched the iFLYTEK Spark Cognitive Model. In terms of universal capabilities, the Spark Cognitive Model supports multi style, multi task, long text generation, multi-level, cross language understanding, and open knowledge Q&A in a wide range of fields. At the meeting, iFLYTEK also simultaneously released five application achievements, including iFLYTEK AI learning machine, iFLYTEK hearing, iFLYTEK smart office book, iFLYTEK smart cockpit and iFLYTEK open platform.

The technical specifications of the Spark Cognitive Model cannot be fully understood from on-site demonstrations, but iFlytek seems ambitious. Liu Qingfeng bluntly stated, "On October 24th, during the iFlytek Global Developers Conference, Spark is expected to fully benchmark ChatGPT: to surpass ChatGPT in Chinese and reach a level comparable to it in English.

Coincidentally, the next day when Zhou Hongyi was visiting Dongfang Selection, he said this sentence in the live broadcast room, "If you don't imitate for two years and say you surpass, it's called boasting." I don't know if he meant anything.

In recent years, with the fad of artificial intelligence and the difficulty of commercialization, the filters of iFlytek from the outside world have gradually disappeared, and the voices of doubt have also increased. Nowadays, on the big model track, iFlytek has arrived late, but catching up with the popularity of the big model cannot dispel these doubts.

Raising stock prices to appease hunger?

Recently, some netizens discovered that the iFlytek Spark model may contain content such as "I was developed by OpenAI" in some Q&A sessions, thus questioning its shell ChatGPT. Subsequently, iFlytek denied this claim.

The reason for refutation is also somewhat self deprecating: iFLYTEK believes that if it were a nested ChatGPT, it would not be possible for the iFLYTEK Spark model to have a faster response speed than ChatGPT; There will be no situations where the iFlytek Spark model outperforms ChatGPT in terms of text generation, knowledge Q&A, and mathematical abilities.

From a neutral perspective, whether the iFlytek Spark model is a technical issue that non professionals may find difficult to determine at the moment, but iFlytek also knows some "tricks" in the capital market.

When iFlytek announced the launch of its big model, the company submitted a very ugly financial report to the secondary market. According to the financial report, in 2022, iFlytek achieved a revenue of 18.82 billion yuan, a year-on-year increase of only 2.77%; The net profit attributable to the parent company was 561 million yuan, a year-on-year decrease of 63.94%, marking the first decline in net profit for iFlytek in the past five years. This year, the decline in performance continues. In the first quarter, iFlytek achieved a revenue of 2.888 billion yuan, a year-on-year decrease of 17.64%; The net profit loss was 57.9 million yuan, a year-on-year decrease of 152.25%.

Affected by this, on April 21st, iFlytek's stock price plummeted by 9%. However, just a few days later, the stock price rose again. On April 25th, the Nikkei Frisbee rose by 5%, reaching 61.73 yuan as of 10:57.

Why is the transformation so fast? Because at the 2022 financial report press conference, iFlytek publicly announced that it was about to launch the cognitive big model "Spark", which easily shifted investors' attention from ugly performance to the big model. On April 24th, the company issued a small number of testing invitations, and the stock price immediately rose the next day.

The methods used by iFlytek to attract market attention and enhance investor confidence seem quite skilled. In recent years, against the background of the industry downturn caused by the bursting of the artificial intelligence foam and the generally negative outlook of AI enterprises, iFLYTEK has set ambitious goals several times, trying to re stimulate investors' confidence and stimulate stock prices with different stories.

On February 8, 2021, iFLYTEK proposed at its annual planning conference that it would achieve the goal of "one billion users, one hundred billion revenue, and driving a trillion yuan industrial ecosystem" in the next five years. However, iFLYTEK's revenue in 2020 was only 13 billion yuan. As soon as this news was released, the stock price of iFlytek at the University of Science and Technology (UST) first pulled two limit up boards. Similarly, at the February 2022 annual meeting, Liu Qingfeng announced the launch of the "iFlytek Super Brain 2030 Plan". According to this plan, by 2030, family companion robots will enter thousands of households.

Unexpectedly, iFlytek's stock price rebounded again.

But as these grand and storytelling goals become distant under the impact of reality, they continue to erode the trust of consumers and investors, and at the same time, slowly fade the halo of iFLYTEK's bright AI leading company. This is also why the questioning voices surrounding iFlytek are becoming louder and louder.

IFLYTEK has clearly not given up on similar routines, and the Spark Cognitive Model has not yet been officially opened. Liu Qingfeng has clearly written about iFLYTEK's ambitions, once again raising the expectations of the outside world for its big model. And the secondary market also cooperated abnormally. On April 7th, it reached the highest point of 70.10 yuan since the company went public, which directly doubled compared to the stock price of less than 35 yuan before this round of increase.

Of course, not everyone is optimistic about iFlytek. "The promise is still just a beautiful vision for the company, and the specific performance still needs to be verified by the market." Many investors who emerged from the previous wave of artificial intelligence have become calm.

Never change without its roots

AI once stood on the Internet by virtue of endless new concepts and technologies, but the death of a large number of AI enterprises from the crazy influx of capital to the stage of "squeezing foam" tells us that technology can not be squeezed out on the track unless it is deeply integrated with industry and truly moves towards commercial landing.

This is even more true of large models, and it is difficult to see the technical background of large models from an application perspective.

At the press conference, we saw that iFlytek showcased the abilities of Spark in language comprehension, knowledge Q&A, logical reasoning, mathematics, code, and multimodality, while emphasizing the application of the "Spark Cognitive Model" in four fields: education, office, automotive, and digital employees. But the business landscape outlined by iFlytek for the Spark model is far from that, but rather "1+N", which is a universal cognitive intelligence model. N is the implementation of the big model in various fields, including education, office, automotive, human-computer interaction, and healthcare.

This is consistent with the business framework established by iFLYTEK initially based on intelligent voice technology, that is, platform+track. The platform refers to the open platform. iFLYTEK provides data and solutions, and the application scenarios are handed over to other enterprises. The track is the combination of AI technology and different vertical fields, further applying technology to different scenarios.

Both intelligent speech and AI large models are actually low-level and universal technologies. Therefore, they can integrate with many industries and find a wide range of application scenarios, which naturally means great commercial value and potential. As a provider of technical services, open platform, software and system services are obviously more imaginative than hardware equipment deep in the downstream of the industrial chain.

Internet giants such as Alibaba and Baidu are positioned as AI technology bases in the big model race, making it easier to create sufficiently high technical barriers.

However, as a technology-based company, iFlytek has taken the opposite path or deviated from the original path. Looking back at 2007, iFlytek's revenue from application products in the industry only accounted for 17.02% of the total revenue. However, its main revenue focus has shifted towards the application product end and is now concentrated in one area of the education industry.

This has to be attributed to the fact that iFlytek's technological advantages and capabilities are far from sufficient to support its commercial progress in pure technical services, and it can only turn to downstream hardware devices for monetization. The slow growth of iFLYTEK's open platform business is the most intuitive performance.

By the end of June 2022, iFLYTEK's open platform has opened 496 AI capabilities and scenarios, gathered 3.43 million developers, and had 1.52 million applications in total. However, from the financial report, this did not create much revenue growth for iFlytek. In the first half of 2022, the business's revenue was 1.34 billion yuan, only an increase of 4.15%, accounting for 16.78% of the company's revenue, a decrease of 3.68 percentage points from 20.46% in the same period in 2021.

So, it is worth considering whether iFlytek's massive business system built around the Spark cognitive model will ultimately rely on hardware to make money, or whether it will be labeled as a big model to make hardware sell well. Unlike other companies' release demos for large model launches, iFlytek's launch is clearly more like a product launch.

Hardware revolutionaries are not easy to become

In China, most AI companies face a common dilemma, which is that they cannot find a way to make profits until a certain technology is truly mature and implemented. Like iFlytek, which has been heavily reliant on government subsidies and later attempted to improve profits through "stock trading", iFlytek is still fortunate compared to other AI companies. In its vast business expansion, iFlytek has found a breakthrough in education. Firstly, it gained recognition from the G-end with its own advantages, thereby expanding its ToB business in front of various universities, and then pushing hardware products to the C-end through the schools.

Due to the fact that the main revenue sources of domestic AI enterprises are the G and B end, iFlytek's hardware business on the C end precisely compensates for the instability of the company's G and B end businesses, which is its biggest advantage in outperforming other AI enterprises in terms of performance. According to the financial report, the intelligent education field is an important battlefield for iFlytek's commercial monetization, accounting for over 30% of iFlytek's revenue.

However, in terms of hardware sales in the education field, iFlytek, which is technologically leading, has not yet surpassed traditional leading manufacturers that have already occupied the market. For example, there are many educational tablet products on the market for learning machines. Currently, the top five learning machine manufacturers in terms of shipment volume are still old brands led by Bubugao, Shulang, and Youshixue, excluding iFlytek.

Moreover, in the education hardware market, what is unfavorable for iFlytek is that mobile phone manufacturers such as Huawei and Xiaomi, as well as online education giants such as TAL and Homehelp, are now vying for the cake of education hardware.

The launch of the big model can be said to be a killer weapon for iFlytek to enhance its hardware product capabilities. In February of this year, iFLYTEK launched the AI learning machine T20Pro and stated that ChatGPT like technology will be launched in May, taking the lead in using AI learning machines. At the press conference, Liu Qingfeng announced that the T20 series of learning machines had launched the first batch of Chinese and English essay correction (learning machine end) and generation (family elder end), personalized and precise interactive learning for mathematics learners, and English speaking accompaniment functions.

From the on-site demonstration, it can be seen that, such as grading, learning opportunities are like real-life teachers, delicately outlining and commenting on articles, focusing on standardized vocabulary, skill recognition, structural analysis, content understanding, language performance, writing suggestions, and optimization references, achieving basic grading, high-level grading, and improvement suggestions.

Not only for learning machines, but also for other products such as recording pens, translation pens, and office books for C-end users, it is reported that they have been integrated into the Spark model. But the question is, has the hardware supported by big model technology achieved a qualitative upgrade in product strength, and can it help iFlytek break the market pattern on the hardware track and overtake on curves?

From the information disclosed at the press conference alone, it cannot be seen that the application of large models in educational scenarios requires more time to explore and verify. Moreover, if the universal large model can be used to achieve a more humanized interactive experience of educational hardware, this is undoubtedly the biggest breakthrough. However, it is obvious that iFlytek is not the only one doing this. iFlytek also faces competition from top companies such as XRS and NetEase Youdao.

It is unknown to the outside world how iFLYTEK's technological strength and accumulation are similar to the model being expanded.

In any Internet outlet, the uproar will dissipate before we can see the truth under the foam. iFLYTEK is enjoying the concept dividend of the big model now, and we need to be alert to the possible backbiting in the future.