The spring of AI cannot save the winter of consumer electronics

This article is from the WeChat official account: Photon Planet (ID: TMTweb), written by Wen Yehao, edited by Wu Xianzhi, and illustrated by Visual China

The era of AI reshaping everything, the story of salvation and rejuvenation is repeatedly unfolding in the field of consumer electronics.

In the past few years, mobile phone and PC manufacturers have become increasingly thin in the cold winter of their respective tracks. Therefore, when the AI wave rises again, players seem to have found a life-saving straw and are rushing to enter:

Throughout the past six months, mobile phone manufacturers such as Huawei, Xiaomi, and vivo have successively developed their own big model paths; Lenovo, HP, and other computer players re register with AIPC; Upstream technology giants such as Intel, Qualcomm, and NVIDIA have also joined in hardware upgrades to enhance the AI computing power of CPUs and SoCs - a collaborative effort aimed at leveraging user replacement needs has already begun.

However, it is currently unknown whether this oasis is real or just another mirage.

Repetitive AI Narrative

In my memory, the first time I heard the word "AIPC" can be traced back to the TV screen era when Oak International was still frequently active.

That is a netbook priced at 1699 yuan, equipped with the Windows CE system, with a resolution of only 640 * 480. The host's powerful words repeatedly promote selling points such as listening to music, watching movies, being free from viruses, and having an extended battery life. However, during the nearly ten minutes of advertising, he never explained the meaning of "AI" in the product name.

Time has passed, and the classic Windows CE system has recently completely bid farewell to the stage of history; During the difficult moments between TV programs back then, as a ghost and beast video, it underwent a new life on video platforms. At that time, "AIPC", which only appeared as a code name, became a prophecy and became a trendsetter in the technological context more than a decade later.

Even with a more pragmatic perspective, AI consumer electronics is not a new story.

Before the outbreak of the big model craze, AI had already been widely integrated into consumer electronics products by various manufacturers in various forms.For example, the machine learning ability behind the FaceID that iPhone X "kills" fingerprint unlocking is based on a neural network engine embedded in the A11Biotic chip.

For example, mobile phone manufacturers have been fond of mentioning computational photography in recent years, which essentially uses AI algorithms to achieve functions such as adjusting shooting parameters, improving image quality, and removing ghosts. Computational photography was also one of the most intimate coupling between AI and the consumer electronics industry before mobile phone manufacturers were embroiled in the big model battle.

However, at the beginning of the rise of computational photography, hardware players and application manufacturers were on the same competitive river - mobile phone manufacturers introduced algorithms into photography systems, while application manufacturers launched a series of photography apps based on algorithms. Without achieving differentiated competition, facing the rapid iteration of software applications, hardware players who mainly sell products are naturally unable to gain an advantage.

In this context,The targeted transformation at the hardware level became the solution for mobile phone manufacturers to overcome difficulties at that time.

In 2017, at IFA2017 in Berlin, Germany, Huawei launched the Kirin 970 with built-in NPU neural network units. With the support of NPU neural network units, the image processing level of the Kirin 970 was significantly improved.

Afterwards, from Qualcomm's transformation and upgrading of ISP image signal processing units to the continuous launch of image chips by various mobile phone manufacturers at this stage, the competitive logic of computational photography has gradually shifted from pure volume algorithms in the past to volume hardware dimensions.

With the increasingly strong binding relationship between AI and hardware, mobile phone manufacturers have ultimately avoided the endless consumption war with application manufacturers - with the support of computational photography, mobile imaging has become more specialized, ultimately ringing the death knell for the once popular DSLR cameras; Application manufacturers have gradually left the field of specialized photography and instead achieved differentiated survival through more "lightweight" personalized filters and templates.

At present, mobile phone manufacturers are frequently using end-to-end models,Its logical core seems to have a parallel connection with the past computational photography.

Offensive is the best defense

In the industry context, big models carry the hope of consumer electronics manufacturers to overcome the cold winter.

Taking the mobile phone industry as an example, since August this year, Huawei, Xiaomi, Honor, OPPO, and vivo have all developed their own big model paths. Even Apple, which has been slow in the industry, is recruiting people and working hard to prepare their own big models.

This means that,In the current era of massive invasion by giants, big models have become an area that mobile phone manufacturers have to follow.

Although it has been more than half a year since the "Thousand Model Battle" began, the actual application scenarios of large models often focus on the B-end, and there is still a certain gap between the market and the C-end. And hardware terminals such as mobile phones and PCs have naturally attracted a large number of model manufacturers as excellent entry points for the implementation of the large model C-end.

Since ChatGPT landed in the mobile phone market, ERNIE Bot, iFLYTEK Spark, Zhipu Qingyan and other large model products have followed, successively entering the mobile phone application store. The giants are rushing in, and if mobile phone manufacturers do not take action, they can only carve up the remaining leftovers from application manufacturers.

However, the Battle of the Big Model is different from the battle of mobile imaging in the past. As the most popular track in the current technological context, there are numerous giants participating in the battle, and the competitive pressure exerted on hardware manufacturers is unparalleled - most hardware manufacturers do not have the ability to forcefully compete with technology giants in terms of model parameters, computing power, and corpus.

In this context,The end-to-end model has become the key for consumer electronics players to leverage their "home advantage".

Looking at the big model gameplay of various mobile phone manufacturers, as well as Apple's recruitment information on "implementing compression and acceleration of big language models in the inference engine on Apple devices", it seems that the combination of end-to-end big models and cloud computing has become an unavoidable topic.

From a technical perspective, AI applications such as ChatGPT and Midjournal have models deployed in the cloud, while terminals such as smartphones and PCs can only serve as media for content presentation. And this makes it difficult to avoid many limitations such as long "reflection arc", inability to use offline, and privacy security.

And the integration of end to cloud involves stuffing localized and lightweight large models into terminal devices - when faced with simple needs, real-time offline reasoning and rapid response are achieved through the end to side large models, while complex problems are handed over to the cloud.

To this end, Qualcomm and MediaTek have also cooperated to transform and upgrade their processors, and established close cooperation with mobile phone manufacturers to support the operation of large models on the mobile end.

AIGCOffensive is the best defense

True Singularity or Pseudo Demand

The author of "The Innovator's Dilemma" and Harvard Business School professor Clayton Christensen once said, "Innovation is initially insignificant and a joke, but suddenly one day it meets the soft spot of consumers and rapidly develops into a ruler

The logic behind this sentence of "from quantitative change to qualitative change" has always been a belief in the technological context. In many cases, this is just players' self flattery - carrying the so-called technological romance, but in the end no one buys it, only touching themselves.

For the consumer electronics industry, the big model seems to be the "life-saving straw" that triggers a wave of changes and reshapes the cycle. At this stage, no one can answer where the "killer applications" are? And how much are users willing to pay for the current large model application?

The ups and downs of ERNIE Bot, iFLYTEK Spark and other products in the C-end market confirm that,The perception and actual demand for technological succession in the consumer market seem not as strong as players imagine

On the other hand, whether it's a large model phone or Lenovo's AIPC, the underlying logic is to leverage the replacement demand of the existing market through AI.

The same pit, players have stepped on the folding screen once.At that time, the smartphone market was severely homogenized, coinciding with the iterative upgrade of flexible screen technology. Hungry manufacturers immediately launched foldable screen products and continuously packaged them with new demands.

In the early stages of folding screens, apart from the widely criticized issues of creases and hinges, there were also many mismatches and immaturities in software and applications, which made the consumer market consider them as "chicken ribs". The Pandora Magic Box has been opened, and the players have all entered the game. They cannot retreat, so they have to carry the folding screen with the supply chain and solve problems such as hinge, adaptation, and weight step by step.

Until today, after multiple rounds of evolution, the shipment volume of foldable screens has achieved significant growth. Even so, foldable screens have not been able to reverse the decline of the smartphone market. In the PC field, even with NVIDIA's continuously iterative and upgraded DLSS technology, what it can do is only to keep the PC alive rather than save it, failing to save the declining PC market.

Perhaps it is difficult to wait for the next "iPhone moment" in the smartphone, PC, and other tracks - consumer electronics manufacturers aim to cultivate the next faster horse, but what the consumer market wants may be a car.

In other words,If the development of the C-end of the large model is tightly tied to hardware, the desire for replacement in the stock market will be difficult to support the popularization and prosperity of AI, and the next entry-level hardware that can fully enjoy the demographic dividend is the silver bullet.

Fortunately, the close correlation and cross application between consumer electronics technologies have injected deeper and more diverse existence value into hardware manufacturers committed to large model research and development. Although the results have not yet been revealed on the shallow shore of time, this move has undoubtedly planted the seeds of foresight and laid a solid foundation for the future.

This article is from WeChat official account: Photon Planet (ID: TMTweb), written by Wen Yehao, edited by Wu Xianzhi

This content is the author's independent viewpoint and does not represent the Tiger Smell stance. Reproduction is not allowed without permission. For authorization matters, please contact [email protected]