＂AI+ hardware＂, ByteDance gave a new answer.

Wen | Sanyi Life

ByteDance won’t make AI mobile phones, but will make AI hardware. Recently, it was reported that ByteDance’s PICO is developing several AI wearable devices. Different from the vagueness of Internet giants about new projects in the past, ByteDance gave a straightforward affirmative answer. "Bean bag business is exploring the combination with smart wearable devices to provide users with a more natural and convenient interactive experience."

AI+ hardware is undoubtedly a popular track in the current AI field. Not to mention whether the two products, AI Pin and Rabbit R1, are superficial, but they have indeed caused extensive discussions, and a large number of consumers have expressed their support for them with real money, which is enough to see that AI hardware is not a false proposition. What’s more, Meta has set a good example for the industry, proving that AI hardware may be really promising.

Just a few days ago, Meta reorganized its Reality Labs department. In addition to the meta-universe department composed of Quest head-mounted series and Horizon, it also set up a new "wearable device" department.

Meta’s purpose of splitting Reality Labs into two parts is not only to lay off employees, but also to see the value of more AI hardware products because the sales of Meta Ray-Ban smart glasses far exceed expectations. The newly established wearable device department of Reality Labs is more like Meta’s ratification of the market achievements of Ray-Ban smart glasses.

As a smart glasses product jointly launched by Meta and Ray-Ban, the first generation product of Meta Ray-Ban, Ray-Ban Stories, was released in the autumn of 2021, but the market performance of this product at that time was extremely poor, and the sales volume in the whole life cycle was less than 300,000 units. Then, the second-generation product Ray-Ban Meta, which appeared with Quest 3 last autumn, made a turnaround. According to the estimation of a third-party organization, Ray-Ban Meta smart glasses have sold more than 1 million sets. In Meta’s last earnings conference call, Zuckerberg even mentioned that several styles of Ray-Ban Meta have been sold out.

In fact, the design language of Ray-Ban Meta is similar to Google Glasse, and it is similar to ordinary sunglasses in appearance, except that two 12-megapixel cameras are added to the frame, speakers and five microphone arrays are added to the temples, and Qualcomm Snapdragon AR1 Gen1 is equipped.

At first glance, the design of Ray-Ban Meta seems to be no different from mainstream AR glasses, but what really makes this smart glasses stand out is the Meta AI that Meta joined this spring. Based on the Llama 2 model, the multi-modal AI function of Ray-Ban Meta can not only translate speech, but also translate what users see in real time, and realize a conversation similar to ChatGPT. Users don’t need to take out their mobile phones and simply look at their glasses, so they can let Ray-Ban Meta complete the functions of searching, navigating and taking photos through voice.

"GPT on the bridge of the nose", this is the label that some overseas media put on Ray-Ban Meta. Its success lies in providing users with fashion attributes and adding AI capabilities, so that users can enjoy the practicality and entertainment of Meta AI more conveniently. The significance of Ray-Ban Meta is to prove the feasibility of AI hardware, and to give friends the information to continue to introduce AI hardware products. To know that the biggest headache for AI model manufacturers at present is commercialization, so how to let users use it first is very important.

In fact, the AI ​ ​ big model itself is the result of "big bricks flying", and its technical threshold is not as high as everyone thinks. Otherwise, after the appearance of OpenAI’s ChatGPT, the domestic market will enter a state of hundred-mode war in just half a year. Today, a year and a half later, a big model like Claude 3.5 Sonnet, which surpasses the latest GPT-4o of OpenAI, has also appeared. Although there are many large models, there are not enough users who can really use the capabilities of large models in real life.

So that in order to find possible application scenarios, domestic large model manufacturers not only frantically price, in order to attract the favor of third-party developers, but also "put satellites" one by one. For example, on the same day that ByteDance was going to confirm the news of AI hardware, Tencent’s AI macro model application "Tencent Yuanbao" announced that it supported the super-long word processing capability of a single document with a maximum length of 10 million words, and Kimi, the dark side of the previous month, announced that it would test the Context Caching. Why do big model manufacturers come up with this kind of "performance art"? In the final analysis, the ability of the big model itself is not strong enough, and it is not enough to reshape the mode of using PCs and mobile phones.

The current situation is that the big model is adapting to mobile phones and PCs, not the other way around. Similar to Google’s cooperation with Samsung to provide AI support for Galaxy phones and OpenAI’s cooperation with Apple to provide AI capabilities for iPhone, it is only an expedient measure for big model manufacturers. Because AI represents the future to a certain extent, this view is already the greatest common denominator of the technology industry, and manufacturers with technical strength such as Apple and Samsung will inevitably make their own big models in the future.

AI hardware is different. Its significance lies in creating the Application scene of the big model out of thin air, thus avoiding the big model as an app falling into a close struggle with competing products on mobile phones and PCs. It is obviously a feasible way to drive the success of software with the success of hardware.

Then the question is, what will the AI hardware in ByteDance look like? AI mobile phone must be excluded as the wrong answer. Considering PICO’s own experience in manufacturing VR equipment, it is undoubtedly possible to imitate Meta to build a ByteDance version of Meta Ray-Ban and play "Copy To China".

Of course, considering that just a month ago, ByteDance bought Oladance, a domestic earphone manufacturer, for about $50 million. The big model +TWS headphones also seem to be a good combination. After all, compared with smart glasses with higher recognition, thanks to the popularity of TWS headphones, it seems more acceptable for domestic users to go out with an AI headset.