01
In the past two weeks, we have rarely talked about the development of the AI industry.
Because all kinds of "explosion, subversion, innovation" industry information is too dense, everyone is talking about AI, and there is no nutrition.
But we've been watching changes in the industry and the top players, especially as OpenAI has become very aggressive over the past week.
They did three things in a row, each of which was a big move for the entire industry, and we talked about them one by one.
02
The first is the plugin.
We've been talking about ChatGPT and GPT-4, but what's the problem?
The first is that their data stayed in September 2019, and the data was not collected after that, and if the timeline of the question exceeded the limit, many answers would be made up.
This leads us to repeatedly confirm it, or even train it by knowing the answer in advance.
The second is that GPT is a large language model, and the reason why it is called a large model and not a small model means that the field and data it covers are wide, broad and wide, not refined and beautiful.
GPT is difficult to accurately answer the problem of individual needs, let alone directly give us a satisfactory solution for a specific work and life scenario.
It must go through multiple, in-depth questioning and training to improve GPT logic with multiple rounds of prompts.
This leads to a high time cost for us to train for a certain scenario, and the deeper the field and scene, the more difficult it is to train.
If we could have a framework that directly fills in the blanks, how good would it be to give GPT keywords?
Now, it's on its way to realization.
On the latest OpenAI website, GPT has fully opened two features to Plus users (GPT-4 paid users).
One is "Web browsing".
GPT-4 gets rid of the limitation of data and will search on the Internet on its own.
GPT-4, which can be networked, not only expands, but also further summarizes and refines the information searched in real time.
The second is "plugins".
GPT-4 provides 72 third-party plug-ins, all of which are provided by the user, although strange, but all have one thing in common, that is, they are suitable for different segments and scenarios.
For example, do math problems, weather, recipe recommendations, travel guides, book air tickets, and so on.
While not every one of these plugins works well, it means two important things:
First, GPT has been upgraded from a tool to a platform system, just like a mobile app store, and various plug-ins are like apps, turning GPT into a solution expert in subdivided fields and subdivided scenarios.
Second, only GPT-4 has opened the plug-in function, if you want to use it, you must become a paying subscriber, in the same way, let's think deeply, can some excellent plug-ins be charged like mobile phone applications?
If these plugins have product and shopping recommendations (which will definitely exist), can they still charge advertising fees?
Don't underestimate one plugin, which represents the emergence of two new profit models.
03
The second is the mobile APP.
On May 15, OpenAI just launched the plug-in, and on the 18th, OpenAI officially released the iOS version of ChatGPT, which is currently only available to Apple users in the United States to download, but OpenAI is already developing a general version for Android and Apple users.
We asked the mobile version of ChatGPT to answer the difference between it and the PC version.
In terms of basic algorithms, computing power, and data, the mobile version and the PC version are the same, but the reply will be more concise, and at the same time, we can also use the text conversion of the mobile phone voice assistant to achieve voice communication.
Another thing we noticed is that in the mobile version, OpenAI released another subscription payment channel, and users can upgrade their subscriptions directly on the phone.
This means that OpenAI is about to complete the layout in the two mobile ecological camps of iOS and Android, and more than six billion smartphone users around the world have access to ChatGPT and GPT-4.
You know, the current number of active users of OpenAI web version is more than 100 million, if the mobile application is opened, this number will get a terrifying exponential increase, which is a very certain thing.
Combined with the "plug-in function" we just talked about, what problem does the plug-in solve?
Problems with monetization models and segmentation scenarios.
And what problems does "GPTApp" solve?
It is the problem of the number of users, and the commercial closed loop of one link is basically achieved.
But what's missing?
AI regulation issues, and this, excuse me, OpenAI and the U.S. government also thought about.
04
Third, regulatory issues.
On May 16, Sam Altman, founder of OpenAI, the "father of GPT," appeared before the U.S. Congress.
Let me remind you that Sam is sitting in the same position as TikTok CEO Zhou Shouzi last month, and Zuckerberg, Sundar Pichai (Google CEO), Bill Gates and other Internet giants have sat in this position, and they have all been sharply questioned by lawmakers.
The exception is Sam, who seems to have come to receive the award and enjoy his "meritorious moment".
The lawmakers also smiled the whole time, and the whole hearing was harmonious.
Lawmakers and Sam avoided addressing key AI regulatory issues, such as the openness of training data, AI infringement, and attribution of false information.
So what did they talk about?
Sam volunteered to ask the U.S. government to regulate us, set up specialized government agencies, issue industry licenses, set safety standards, have independent experts audit, and legislate on AI.
For the emerging industry of AI, this set of processes is no problem.
However, Sam said that this agency should not only manage OpenAI, but also manage American companies, but also rise to the level of international security, becoming a "international atomic energy" agency, fully regulating the global AI industry.
At the same time, what Sam has to say about the monopoly problem;
"The innovation born of AI may be the most important technological revolution, and we have become one of the few companies that are ahead, but we still have to pay huge R&D expenses, and if there are fewer companies involved, then it will be easier for the government to regulate us."
Lawmakers kept smiling at the answer.
I would say that if you control the rules, you control the competition.
05
In 4 days, OpenAI did 3 major things in a row, solved the problems of profit model, number of users, and industry supervision, and OpenAI almost cleared the obstacles to the comprehensive development of ChatGPT, showing a strong aggressiveness.
Thinking of the Tencent shareholders' meeting on May 17, our column teacher Shan Renxing, teacher Song Zi, asked Ma Huateng a question on the spot:
ChatGPT is very popular, Tencent is currently also doing mixed element model, but at present, it seems that Tencent is still very low-key compared to other companies, I want to know whether Tencent is holding back a big move, or adopting some other strategy?
Ma Huateng replied:
At first, we thought that this was an opportunity that the Internet would not meet in a decade, but the more we thought about it, the more we felt that this was an opportunity that would not happen in hundreds of years, just like the industrial revolution, so we felt that AI was very important.
But this does require a lot of accumulation, Internet companies are doing, we are also immersed in research and development, but there is no rush to finish early, and the semi-finished products are displayed.
For the industrial revolution, taking out the light bulb a month earlier is less important in terms of a long time span.
The key is to do a solid job of the underlying algorithms, computing power and data, and more importantly, the scene landing, I believe we have many scenarios that can be landed.
We're still doing some thinking, and I feel like a lot of companies are in a hurry right now, and it feels like it's to boost the stock price, and we've not always been in that style.
In addition, we feel that many scenarios can naturally apply technology, and we are not in a hurry to immediately go to the stage to grab a seat, the seat is already there, the scene, users, and services are already there.
Therefore, it is more about whether we can finally come up with the best products and technologies, and we feel that this is a more critical part.
It is hoped that as Ma Huateng judged, in the next development of the AIGC industry, Chinese companies can occupy a place and win in the high-tech competition.
—
Responsible Editor | Luo Yingfan
The pictures in this article are all from the Internet