Under the AIGC wave, the "model" force is sweeping the technology context.

While Baidu, Ali and other giants are playing cards one after another, players such as Zhihu, Xiaohongshu and Yunzong Technology, whose volume is slightly less, are also trying to squeeze into the track - the table of big models is becoming more and more crowded.

It is foreseeable that the big model track will not escape the process of reshuffling. In contrast to the giants who are bound to fight to the end, the players on the halfway point seem to be more careful.

Waist big model, each has its own way to play

The "waist big model" is not an exact definition.

At this stage, most of the domestic large model products have not been baptized in the market in depth, so it is impossible to clearly distinguish their advantages and disadvantages, but if we look at the differences between the players behind the large model in terms of volume, resource investment, technical background, application scenarios, etc., we can still simply delineate certain boundaries.

Take Baidu and Ali for example, as the absolute giants of Chinese Internet, they have deep resources and technical reserves, and what they think about is not how to bear the high cost of large-scale model training, but how to cover more fields with big model products, so as to preserve the name of "BAT" in the next round of technological evolution.

On the contrary, another group of players, due to their small size and limited resources, focus their so-called big models on specific areas and develop big models to match their business needs.

In April this year, Zhihu released "Zhihaitu AI" Chinese big model to the public, and so far, it has already developed "hot list summary", "search aggregation" and other big model application functions within the site. In March this year, Xiaohongshu was revealed to have set up a secret team of big models, and internally, it is also quite a horse race to promote AIGC applications in multiple departments at the same time.

It can be seen that, compared to the large model products that are mostly launched in the search engine and enterprise service fields, the waist large model is more inclined to vertical application scenarios.

For example, Zhihu, as the leading Q&A community in China, has both a rich Chinese corpus and a unique grounding scenario, so we can see that whether it is the "search aggregation" and other in-station applications being tested by Zhihu, or the next phase of the so-called "empowering creators", the core of all of them is the "AIGC". "The core is to optimize the user experience without departing from its original system.

Another example is 360, since this year, both Zhou Hongyi and the 360 company itself, have told many stories around the 360 Smart Brain class ChatGPT, but the story is beautiful, eventually have to return to reality.

Not surprisingly, at the end of May, 360 wise brain big language model was upgraded to 360 wise brain - visual big model. It is reported that 360 intelligent brain - visual big model can be widely used in the long-tail target detection, store patrol, caretaker, equipment inspection and other scenarios - round and round, the end is still familiar security.

A number of differentiated routes, quite a bit of "everything can be a big model" of the meaning of the players can not wait from head to toe, armed with a large model, either "change the world", or resist this round of technological change, to become " The "100-year old store".

Although the big model is a product of technological evolution, not all players can embrace it casually. If you enter the game blindly, you will not only be trapped in it, but also the battle of ten thousand models will become a huge bubble that will eventually burst under the test of the market. It may be time for the big model players who live in the middle of the mountain to change the way they live.

Big model is not a universal key

Big model is not a universal key, it is more like a double-edged sword. Not to mention the burden brought by the huge resources required for the development and maintenance of large models, as well as compliance issues such as data security, many players have not been able to find their place in the application scene alone.

At this stage, the ChatGPT-like dialogue scenario is the "star scenario", for example, and players are trying to replicate ChatGPT and constantly comparing the gap between their own applications and ChatGPT. Take the aforementioned 360 Smart Brain Big Language Model as an example, at the beginning of its release, there was no difference between it and other similar products in the track; the same is true for the Cloud from the end of the cloud from the big model.

I do not know, players in this track within the pie is extremely difficult, blindly into the field is very likely to fall into the dilemma of "repeat manufacturing wheel". A wide range of C-end users, although also interested in the domestic version of ChatGPT, but the majority of users will lose their novelty after a short period of intensive experience. As Zhou Hongyi said, ordinary people may not have the motivation to use the big model, which is still a productivity tool by nature.

In other words, although the big model has solved the problem of content creation to a certain extent, it has not immediately established a new business order - the reason why Ali and Baidu dare to continuously increase the code of the big model, in addition to having a rich C-sided scene, more from their respective enterprise service heritage. Perhaps because of the logic of this layer, 360 will be its big model with other business mix, back to the security field.

This means that the more solid ToB roots of the player, better able to embrace the big model. Conversely, there is no ToB gene players, it is possible to get lost in the wave of large models or cut.

Take C-focused content communities as an example, whether it is the "aggressive" Baidu Post, the "warm" Xiaohongshu, or the former "forced Hu", the value of content communities has always been more than just information. The value of the community is not only limited to the information itself, but also the people who make up the community and the unique atmosphere they create together.

For example, although Xiaohongshu has not yet been able to break the shackles of commercialization, the community atmosphere has always been well maintained. However, with the AIGC wave sweeping in, painters who used to receive praise for their work are now often questioned in the comments, such as "is it painted by AI? The "AI beauty" surrounds Little Red Book, which also injects a touch of suspicion into the community.

The platform is down to introduce large models into the content creation, the original intention is to stimulate the creativity of users, but inevitably will also be stripped of the original "human nature", and thus have a profound impact on the community atmosphere. This is perhaps the reason why Xiaohongshu is hesitant in the field of big models.

Compared to Xiaohongshu, Zhihu's series of actions show that it is more eager to embrace this round of technology trends. It is worth mentioning that for the aforementioned issue of content ecology, Zhihu CEO Zhou Yuan has said that the starting point of Zhihu's discussion and sharing is people, and the end point is also people, and AI is just the link and assistant.

In this regard, Zhihu is no different from those content players who want to tell stories by AIGC but are afraid of touching the sensitive nerves of creators. The emergence of big models has changed the rules of the game, and no matter how much it "appeases" them, the battle between AI and real creators is inevitable.

Therefore, one of the reasons for Zhihu's crazy layout is that it has the "courage" to sink the ship, after all, the humanistic and warm community atmosphere that could be destroyed by the big model has gradually disappeared over the years - Zhihu, with its marketing copy flying around, is obviously not afraid of the AIGC. The company is obviously not afraid of the "industrial waste water" of the AIGC era.

In addition to the business level exploration, another reason for Zhihu to enter the big model is perhaps its deliberate pursuit of "new". Looking at the AIGC track, there are few players who uphold this logic.

Meta-universe 2.0?

In the technology context, the motivation for players to develop and invest in large models and launch AIGC applications is obviously not only for business-level considerations.

For example, as early as before the public testing of the cloud from the "from the tolerance" large model, it has announced that it intends to raise additional capital of not more than 3.635 billion yuan for the cloud from the "industry genie" large model research and development project, trying to seize the AIGC this life-saving straw; and Zhihu. As a "high-quality Chinese corpus", in fact, as early as before the announcement of the big model, the stock price has risen because of the "ChatGPT concept stock".

Although Zhihu lacks the technological background, it does not prevent itself from applying the aura of "cutting-edge technology" through its "money power" - Zhihu The main character of the "Zhihaitu AI" Chinese model is not actually Zhihu, but the face-to-face intelligence led by Zhihu. Perhaps to strengthen the binding relationship, recently Zhihu CTO Li Dahai will also be the director and CEO of Mianbi Intelligence.

The players are competing to enter the big model track, which is undoubtedly a dramatic "technology drama", but behind the curtain, there are more different realities faced by different players - for players such as Cloudy and Zhihu, AIGC is both a round of bets and a In contrast, TME, which tells the story of AIGC, sees AIGC as a "placebo", bringing a touch of brightness to a business with few stories.

For example, Soul, after its unsuccessful listing in the U.S., submitted a prospectus to the Hong Kong Stock Exchange in June last year, planning to go public in Hong Kong. Since Soul was in the heyday of the meta-universe at that time, Soul used "meta-universe" to wrap its past business and changed its slogan from "Follow your soul to find you" to "Young people's social meta-universe". Universe".

Now, Soul's plan to go public in Hong Kong is at a critical period, but the concept of meta-universe has gradually cooled, from the concept of technology to marketing concepts, leaving only the so-called "insiders" in various industry meetings and dinner parties to talk about themselves. In this context, Soul once again turned the direction, from the "meta-universe" to the intelligent dialogue robot "AI Gou egg" and other AIGC products, itself has also become the first batch of ecological partners of Baidu Wenxin a word, behind the calculations played The calculation behind it is self-explanatory.

It is worth noting that, compared to the meta-universe of dialing concepts, AIGC applications require more technical support. Therefore, for those players whose volume is not enough to support the development of large models, but also eager to tell the story of AIGC, access to the program of large manufacturers may be the only choice.

In the final analysis, except for the game industry, internal cost reduction and efficiency of enterprises, and a small group of people who can make money in extremely segmented areas such as selling classes and ChatGPT accounts, the AIGC track has not yet run a mature and replicable business order.

Throughout the players who continue to enter the game, some regard AIGC as a former meta-universe, as a "tool" to serve specific demands; some want to bet on a bet to get the key to open the door to the gold mine. Unbeknownst to them, in this gold rush, selling water is the business that can make money at this stage - the game of domestic AIGC, the final probability is still the game of several ToB giants.