When the epidemic is still in a "small upsurge", entrepreneurs are not idle. Just a few days ago, in a house in Tongzhou, Beijing, the former employees of several large factories were planning for AIGC to take a piece of the track. BAT is not as keen on track competition as before, but now their idea is more defensive. A former employee of BAT who would not disclose his name said so to Logco Planet. Of course, this is also an opportunity for entrepreneurship. The following content will answer for you.

1、 The real reason for not placing re note

At the moment, no one seems to deny the popularity of this field in the overwhelming number of AIGC articles. Entrepreneurs are looking for opportunities, big factories are working hard, and the next grand meeting seems to be coming. According to incomplete statistics, on the track of AIGC, more companies have joined the entrepreneurial level. But faced with the huge momentum, the giants seemed afraid to pay attention to the outcome.

This is first reflected in the cost. At the level of large factories, the first problem is that no effective profit model can be found. A former BAT employee shared the digital planet. In his view, it is difficult for ordinary entrepreneurial enterprises to maintain such high expenditure, as is the case for large factories.

The giant's attitude towards AIGC seems to be more inclined to obtain a "sense of participation" than the past "Soho". This is also reflected in the promotion of parameters by some giants. Although some manufacturers claim to have trillions of parameters, according to our research, their models are too sparse, and their estimation effect is the same as hundreds of billions. After repeated inquiries, insiders still refused to disclose the name of the factory and the specific situation of the project.

In fact, in addition to the number of parameters, the intelligence level factor also has a parameter "density". In other words, in sparse models, although there are many parameters, they are all "sesame seeds". In some specific aspects, the "big model" cannot even catch up with the "small model".

At present, Baidu's Wenxin model is recognized as one of the products with the largest promotion and relatively complete business system among major manufacturers. Compared with other large factories, the "bottom-up" Baidu Wenxin model is a typical "top-down" case. From the perspective of other employees in large factories, Baidu's intention to overtake at AI curves is understandable, but I don't know the specific level of commercial realization, so I can only wait for time to check.

2、 Domestic AIGC has natural hard injury

Obviously, for investors, the story of "becoming a Chinese version of OpenAI" is very attractive. In other words, from the perspective of domestic entrepreneurs, it does not aspire to the application level, especially other languages. Therefore, for Chinese enterprises, there is a natural obstacle between dream and reality - distrust among domestic enterprises.

The relationship with enterprise service industry is basically similar. Application oriented enterprises need to conduct secondary development based on the interface provided by "base oriented" enterprises. At first, it was not bad. In the cloud computing ecosystem, manufacturers can make profits by selling server cloud resources and other infrastructure. However, the problem is that in AIGC, compared with manufacturers, manufacturers usually have no effective way to make profits.

As a result, "base" manufacturers are particularly jealous of the huge profits at the application level. In the view of some observers, the practice of large factories may be subverted by the "three-tier structure" model that OpenAI believes in, so that large factories face consumers and become a "two-layer model". If this idea becomes a reality, for countless people, AIGC is undoubtedly a sword of Damocles for development enterprises.

For startups, the difficult question is how to find their own ecological location. However, in China, practitioners are slightly pessimistic about whether there is a win-win basis between basic manufacturers and application manufacturers.

3、 Opportunities seem limitless, but they are actually limited

What kind of projects do investors like? Someone gave his own opinion: we should have a deep accumulation in the industry, understand the algorithm, understand the industry, be flexible in change, and the team should be young.

This may be a rough judgment standard, but in terms of details, in the process of many enterprises participating in the competition, it is not easy to choose the future star. In other words, from a purely investment point of view, the AIGC track is very risky.

At present, to the known extent, a more reliable landing model is to replace the game industry with some art outsourcing work of AIGC. Because the best Wensheng Diagram algorithm is open source. If you have data and industry resources, it will spread quickly, said one practitioner.

A developer said: I still think the big model has a ceiling. The things of the big model are very complex and there are many things designed. As a technician, he thinks that AIGC is full of technical achievements, and even a little "disdainful" of small model products. He gave a more incisive judgment that AIGC investment tide may be a gust of wind, and the final industry pattern may evolve into a large factory doing its own foundation and application. This logic believes that after AIGC reduces the cost of data training, the balance of industry competition will tend to giant technology enterprises, and the situation of SMEs will be extremely awkward.

If the era of big factories comes again and the industrial pattern develops, I think for big factories, the R&D middle ground will come back. In his judgment, the last generation AI craze gave birth to the "Research Institute" model, which is more suitable for GPT-4. In the new stage, the responsibility of the Research Institute is to study the big language model, and then let each department apply it in different fields.

summary

To sum up, in his mind, for entrepreneurial projects, at this time, he is just "fighting" the time window. The pursuit of short-term profits will reduce enterprise barriers. In fact, only for the game scenario, there are not many model parameters, which are affordable for startups, but the problem is that no one knows how long the business can last. In other words, it is difficult to determine whether it is a business model or a business model.