Spark model catches up with ChatGPT? iFLYTEK may have been in development for less than a year

author:New Fortune View V
Spark model catches up with ChatGPT? iFLYTEK may have been in development for less than a year

"What would Adam Smith say at the 2023 annual meeting of the China Association of Listed Companies?" On May 27, at the annual meeting (council) of the China Association of Listed Companies and the 2023 China Listed Companies Summit, Liu Qingfeng, chairman of iFLYTEK, said "difficult" to the big model of iFLYTEK Xinghuo during the live demonstration.

The model "recognized at a glance" and replied, "It has been more than 250 years since he (Adam Smith) died in 2023, so it is impossible for him to participate in any activities in reality." ”

This "dripping water" answer not only stunned many listed company executives on the scene, but also gave Liu Qingfeng more confidence to catch up with Chat GPT.

Not long ago, Liu Qingfeng said that iFLYTEK Xinghuo's language comprehension ability is not only far ahead in China, but also only one step away from ChatGPT, only two points away from 100 points, and will surpass it before October this year.

With ChatGPT detonating the global technology field and capital market, artificial intelligence has become a hot topic in the A-share market. As one of the leaders of the artificial intelligence sector, iFLYTEK naturally also enjoyed a wave of dividends from soaring stock prices.

Data show that as of the close of trading on May 27, iFLYTEK's share price of 55.84 yuan rose by 70% during the year, and the highest increase was more than 100%. This fully reflects the market's pursuit of the Spark model.

But looking at iFLYTEK's 2022 annual report and 2023 Q1 financial report, among them, Q1 deducted non-attributable net profit fell by 331.82%! It seems that strong R&D capabilities have not translated into actual results. Although there are many companies in the market that have suffered long-term losses but insist on R&D and eventually become industry giants, is iFLYTEK such a company?

Spark large model development or less than a year

In the context of artificial intelligence, especially large models, which are highly sought after, iFLYTEK's frequent release of news about the Spark model has also triggered heated discussions in the market.

Spark model catches up with ChatGPT? iFLYTEK may have been in development for less than a year

In the recent record of investor communication activities, the company announced the milestone plan of the Spark model:

June 9: Break through open-ended Q&A (i.e. real-time Q&A), gain knowledge through class search plugins, and present it in a more user-friendly way. In addition, upgrade the ability to multi-round dialogue, and the mathematical ability will be raised to a new level.

August 15: Upgraded code capabilities to make developers and partners efficient and convenient, and iFLYTEK has already used code generation capabilities internally. At the same time, the multimodal interaction capability is officially opened to customers.

October 24th (iFLYTEK Global Developer Festival): Through scientific and systematic evaluation methods, it can benchmark ChatGPT in the ability of general cognitive large models, surpass it in Chinese, and reach the same level as it in English.

Doesn't that seem exciting? However, according to industry analysts, the complexity and scale of large models directly depend on the amount of data required for model training.

So to build a large model, you must be able to collect, cleanse, and label large amounts of data, and the data must be representative. This requires the company to have sufficient computing resources and computing power, powerful network architecture and algorithms, a large number of R&D teams and experts in professional fields, including developers, data scientists, researchers and domain experts, etc., to continuously update, optimize and improve models. "These require sufficient R&D time and data collection." The above industry insiders believe.

As for benchmarking or even exceeding ChatGPT, ordinary investors do not understand very well in the field of cutting-edge technology of artificial intelligence, so simply compare the investment in research and development expenses. According to public reports, OpenAI has invested billions of dollars in chatGPT research and development over the years and will continue to invest. According to iFLYTEK's annual report, R&D expenses in 2022 will be 3.111 billion, accounting for 16.53% of the current revenue, which is not as much as the sales expenses invested in the same period (3.164 billion).

What's more, the R&D expenses of more than 3 billion yuan are not all spent on the Spark model. What is more noteworthy is that in the 2022 half-year report, iFLYTEK's business panorama introduction did not mention the Spark model. In less than a year since the launch was completed, rapid research and development progress has been made. This alone makes OpenAI bow down, let alone catch up before October?

Spark model catches up with ChatGPT? iFLYTEK may have been in development for less than a year

iFLYTEK business panorama Source: The company's 2022 interim report

Compared with iFLYTEK, OpenAI looks too time-consuming and expensive. But looking back at iFLYTEK's research and development achievements over the years, it really can't make investors really believe that they can surpass ChatGPT in a short period of time.

In addition, the latest news about ChatGPT is that on May 16, 2023, Beijing time, Sam Altman, founder and CEO of OpenAI, attended a hearing held by the US Congress on the theme of "AI Regulation: Rules for Artificial Intelligence". After experiencing the launch of ChatGPT, the global explosion, and becoming the hottest company in the field of AI, the other side of OpenAI is facing controversies in US regulators, public data attribution, intellectual copyright attribution and ethics.

If the Spark model can really surpass ChatGPT, I don't know if it is prepared accordingly, how to deal with the doubts from the regulatory level, public data ownership disputes, intellectual copyright and AI ethics? After all, you can't "just follow people to eat meat, and not accompany people to be beaten", right?

Can't stop subsidies?

Following Liu Dong to look up at the bright starry sky of the Spark Model, let's look down at the actual performance of iFLYTEK.

In 2022, iFLYTEK achieved operating income of 18.82 billion yuan, a year-on-year increase of 2.77%; The net profit attributable to shareholders of the listed company was 561 million yuan, down 63.94% from the same period last year. The net profit attributable to shareholders of the listed company, net of non-recurring profits and losses, was 418 million yuan, down 57.31% from the same period last year.

In the first quarter of 2023, the performance continued to decline. The current operating income was 2.888 billion yuan, a year-on-year decrease of 17.64%; The net loss attributable to shareholders of the listed company was 57.8953 million yuan, turning from profit to loss year-on-year; the net loss attributable to shareholders of the listed company, net of non-recurring profits and losses, was 338 million yuan, down 331.82%.

The company gave some explanations for the collapse of performance, such as the Spring Festival factor, the special social environment factor in the first quarter of this year, being included in the US entity list, and the large model research is in the investment period.

Spark model catches up with ChatGPT? iFLYTEK may have been in development for less than a year

But aside from the short-term factors in the first quarter of this year, the company's performance growth rate has actually declined as early as 2022, Q1 2022, mid-2022, Q3 2022, the end of 2022, and Q1 of 2023, the company's non-net profit year-on-year growth rate is 37.73%, 33.21%, -17.04%, -57.31%, -331.82% respectively. It has been declining for five consecutive quarters, and the magnitude of negative growth is accelerating. In this case, in addition to external factors, I am afraid that the company itself must also be found.

According to the financial report data of previous years, from 2019 to 2022, the government subsidies included in the profit and loss of the current period were 412 million, 426 million, 439 million and 1 billion, while the company's attributable net profit in the same period was 819 million, 1.364 billion, 1.556 billion and 561 million, and the proportion of government subsidies to the attributable net profit was 50.31%, 31.23%, 28.21% and 178.25% respectively. It can be seen that iFLYTEK's dependence on government subsidies is still quite high, and it even exceeds the attributable net profit in 2022.

In the first quarter of this year, the company said that the cash received from other operating activities decreased by 64.30% compared with the same period of the previous year, mainly due to the decrease in government subsidies received in the current period. So if government subsidies continue to decline in the future, what will companies do?

In fact, to measure whether a company has a strong own hematopoietic capacity, just look at cash flow. Results without net operating cash inflows can often be mere paper wealth.

According to this year's Q1 financial report, the company's net cash flow from operating activities was -1.668 billion, net cash flow from investing activities was -280 million, net cash flow from financing activities was 817 million, and cash and cash equivalents decreased by 1.135 billion from the beginning of the period. This also confirms the above problem, the company's hematopoietic capacity to bring net cash inflow through operation and investment is seriously insufficient, and must rely on financing and government subsidies to bring positive cash inflow.

Of course, iFLYTEK has another channel for cash flow – financing from stock market investors. According to incomplete statistics of Oriental Wealth, iFLYTEK has paid dividends of 1.25 billion yuan since its listing, refinanced about 10 billion yuan from the capital market, and the total amount of major shareholders and other completed and planned cash out has exceeded 4 billion.

AI ecology is inseparable from selling products?

From a business point of view, although iFLYTEK has always been known as an AI ecological enterprise, except for voice-related products, the company has not produced new core products that are loud in the market for many years, which is one of the important reasons for the company's weak performance growth.

From the perspective of the company's business structure, more than 30% of the revenue by product category comes from education products and services; Open platform and information engineering each contributed about 14% of revenue and revenue, another 11% came from carrier-related businesses, and other product categories (such as smart finance, smart medical and other businesses that the company had high hopes for) contributed in single digits.

One of the biggest "hobbies" of tech companies is to describe the business as "tall, big, and up", but investors are usually more concerned: How does your company make money? So how does iFLYTEK make money? Let's take the description of the smart education business in the 2022 mid-year report as an example to disassemble it.

iFLYTEK's description of smart education business is: "In the field of education, under the guidance of the concept of 'artificial intelligence helps education, teaching according to aptitude achieves dreams', iFLYTEK provides accurate and convenient services for students, teachers and education administrators at all levels through the deep integration and application of artificial intelligence core technologies and education and teaching scenarios, realizes data accumulation in the teaching and learning process, helps teachers and students reduce burdens and increase efficiency, and promotes educational progress." ”

At present, iFLYTEK smart education products have been applied in 32 provincial-level administrative units across the country and overseas markets such as Japan and Singapore.

It is understood that the business has built a business system for three types of customers in G/B/C: the G-end business is mainly based on the construction of cities, counties and districts, covering the overall solution for teaching various schools and users in the region; The B-end business is mainly based on school construction, including school-oriented smart classrooms, big data precision teaching, English listening and speaking classes, etc. The C-end business is mainly based on independent purchases by parent user groups, including AI learning machines, personalized learning manuals and other products.

Spark model catches up with ChatGPT? iFLYTEK may have been in development for less than a year

The cloud obscures a large circle, which is essentially to allow parents to independently purchase educational products such as AI learning machines and personalized learning manuals. According to the annual report, iFLYTEK's education products and services generated revenue of 6.2 billion yuan last year, which is also its most profitable business.

In addition, "the deepening and expansion of the product application value around the school scene is based on the background of the new education reform and the needs of the school, and the hardware of the large screen of iFLYTEK smart window and the small screen of the student terminal is launched and upgraded, so as to build a new classroom space with multi-terminal data integration." "To put it bluntly, or sell products, all kinds of large screens and small screens, electronic blackboards and other hardware, the so-called intelligent, more is to constantly enrich the question bank, customized solutions and so on."

But what do you want to say is that you can develop the world's leading large model with these accumulations? Believe it or not, stock market investors aren't buying too much anyway. On May 18, when Liu Dong released his bold claim to catch up with ChatGPT, iFLYTEK's share price rose only 4.9%, and the day after the news was released, it fell 2.08%.

Today, the world's major tech giants are fiercely competing in the field of big models to seize the first-mover advantage. iFLYTEK also released a milestone plan for the Spark model. We wish the company substantial R&D progress, and also hope that the company can do a good job in a down-to-earth manner and continue to bring good returns to investors and minority shareholders.

(Disclaimer: This article is based on public information and data analysis, and does not constitute any investment advice.) )

Read on