OpenAI, no longer an "evangelist"

When OpenAI's chief technology officer, Mira Muraty, released a desktop version of ChatGPT and a new flagship model, GPT-4o, OpenAI's development entered a truly singularity moment.

OpenAI's short 26-minute press conference still caused a lot of waves.

Whether it's ChatGPT-4o's completely free ride, or the launch of the PC desktop version of ChatGPT, all of them show us that OpenAI is ahead.

However, if you just blindly brag about OpenAI's lead, it will not do much good.

The reason is that OpenAI does have a relative lead in many aspects, and such a lead is not only reflected in the gap between it and Google, but also in the gap between it and Chinese AI players.

Through OpenAI's press conference, what we need to see more is the profound changes that are taking place in the current AI industry, and to find new opportunities that are suitable for the current development trend.

If we must find the inspiration brought to us by this short press conference of OpenAI, I prefer to see it as a direct manifestation of the fact that AI is walking on the road of the Internet.

Starting from this, AI will truly become an existence that is recognized, used, and accepted by more and more people from a laboratory confined to a laboratory.

If we regard OpenAI before ChatGPT-4o as an evangelist, then OpenAI after ChatGPT-4o is more like a practitioner.

一句话，OpenAI正在open AI。

In the future, we will see more and more scenarios and industries begin to see the shadow of AI appear, and we will see more and more players begin to join in the process of popularizing and implementing AI.

Recognizing this, we may be able to better grasp the deeper meaning that OpenAI's short 26-minute press conference brings us.

One

OpenAI's "King Bomb" thrown out at such a press conference is actually the launch of ChatGPT-4o, a completely free application.

On the surface, one of the important reasons why OpenAI is completely free is that its user growth has entered a bottleneck period, and in order to expand more users, OpenAI has begun to acquire new users through free methods.

Since the launch of ChatGPT last year, it has reached a peak in May 2023, with 1.8 billion web visits, but has since experienced a decline in traffic. In March 2024, ChatGPT's number of visits remained stable at 1.8 billion, and there was no significant growth.

When OpenAI further lowers the threshold for use through ChatGPT-4o, which is completely free of charge, the first thing that will have an immediate effect is the growth of users.

It is undoubtedly a very good attempt for OpenAI, which has fallen into a bottleneck period of user growth.

If we regard ChatGPT-4o's complete free operation as an attempt by OpenAI to increase the number of users, then ChatGPT-4o's application of more scenarios to the scenario application of device-side mode and code generation capabilities this time is to expand the growth of OpenAI's user base to a broader field.

This is actually similar to the development of the Internet industry.

We all know that back before the Internet was popular, people's use of the Internet was not as widespread as it is now, and even in many cases, people had to pay for the use of the Internet, just like the use of AI now.

Obviously, it is difficult to unleash the greater development potential of the Internet through payment, and it is difficult to achieve an era called "Internet".

Therefore, in order to release the greater development potential of the Internet, in order to make the Internet truly achieve an era, what must be done is to make the Internet services that used to be charged for free, and really use such a way to achieve the maximum popularization of the Internet. On this basis, let's look for the "Internet+" business model.

It can be said that free has truly made the Internet a kind of "infrastructure", and free has also promoted the continuous maturity and improvement of the "Internet +" model.

When it comes to AI, it's actually the same.

There is no doubt that OpenAI has shown us the power of AI through ChatGPT, and let us see the positive impact on efficiency improvement.

However, in many cases, people's perception of AI is still done under the condition of charging, and it is still constrained by charging.

In such a situation, it is difficult for AI to become an infrastructure, and it is even more difficult to release new and greater development potential.

Therefore, if we must find the specific performance of AI on the road of the Internet, it is completely free, and it is undoubtedly the most direct manifestation to promote the maximum popularization of AI and give birth to more new business models.

When OpenAI tears open the hole for free, we may see more players join it in the future, so as to truly bring people into a new stage of development where AI can be used by everyone and everything.

Two

In addition to ChatGPT-4o's completely free cost, OpenAI showed us more at the press conference the model capabilities, benchmarking, model safety, and limitations of GPT-4o.

In terms of model capabilities, before GPT-4o, the average latency of ChatGPT's voice mode conversations was 2.8 seconds (GPT-3.5) and 5.4 seconds (GPT-4).

Now, with GPT-4o, OpenAI has trained a new model end-to-end across text, visuals, and audio, meaning that all inputs and outputs are processed by the same neural network.

At the traditional benchmark level, GPT-4o achieves GPT-4 Turbo-level performance in text, inference, and coding intelligence, while also being at the top of the multilingual, audio, and visual capabilities.

GPT-4o achieved a new high score of 87.2% on 5 MMLU (General Knowledge Questions), far surpassing Google's Gemini Pro 1.5 and Ultra 1.0, as well as its own GPT-4T and GPT-4.

In terms of model safety and limitations, GPT-4o has built security into the cross-modal design by filtering the training data and refining the model behavior after training. and the creation of a new security system to protect voice output.

Based on an assessment of cybersecurity, CBRN, persuasion, and model autonomy, GPT-4o did not score above medium risk in any of these categories, and the team continued to mitigate new risks found.

If we want to find the information that ChatGPT-4o these characteristics convey to us, in the final analysis, it is to allow more users to use ChatGPT-4o faster, safer, and more efficiently.

To put it simply, these evolutions of OpenAI on ChatGPT-4o are precisely for the better commercialization of its products.

Through this, we can also see that the development of AI is on the road of the Internet.

We all know that in the Internet era, we have experienced the DOS system, the Windows system, and later the iOS system, Android system and other operating systems.

If these operating systems are summarized and defined, in the final analysis, one of the ultimate purposes of their continuous upgrading is to make the Internet better commercialized and better integrated with business scenarios.

When the Internet has such an evolution, we have seen a variety of applications derived from the Internet portal, "Internet+" applications and mobile Internet era.

It can be said that one of the most direct results brought about by the continuous iteration and upgrading of the Internet is the popularization of the commercialization of the Internet.

For AI, in fact, it is also following such a development line.

In the final analysis, AI will eventually settle on commercialization in order to truly release its own development potential to the maximum.

Otherwise, the so-called AI is still just a beautiful story for the capital market, and when the enthusiasm of capital is no longer there, especially when the development of AI really needs self-hematopoiesis, its development will still face such and such problems.

Whether it is Baidu's Robin Li, or 360's Zhou Hongyi, or even GSR Ventures' Zhu Xiaohu, they have actually expressed such a view on AI on different occasions.

At the end of the day, the views they express are actually a central idea.

Such a central idea is actually to make AI focus more on scenarios and applications, and more on commercialization, rather than just staying in the laboratory, and not just limited to a niche existence in a limited number of scenarios.

Through the release of ChatGPT-4o, we can see that OpenAI is exploring and practicing the ways and methods of commercialization, and starting with this, we will see more new signs of the development of AI along the path of the Internet.

Three

Looking back on the development of the Internet, it is not difficult to see that one of the important reasons why the Internet has achieved such great development and become a way of life for people is that the Internet has been popularized to the greatest extent.

Today, the Internet has become a way of life for almost everyone.

The advent of a new era dominated by live broadcast and short video has brought the penetration rate of the Internet to a new height.

For AI, if it wants to achieve new development and become a new way of life, it is necessary to reinterpret the path taken by the Internet in the past, and it is inevitable that the popularization actions made by the Internet in the past will be reinterpreted in its own body.

If we want to find the new revelation provided to us by such a conference as OpenAI, it continues to popularize AI and continue to make AI an existence that can be accepted by both B-side and C-side, which may be another aspect worth paying attention to.

When OpenAI appeared in front of people as a ChatGPT chatbot, it only existed in a scenario like chat, and it only solved the ability of relatively simple logical reasoning.

Today, ChatGPT is no longer a chatbot in the simple sense of the word, it can not only chat conversations, pictures, videos, and even interact with people as quickly as they react.

OpenAI says GPT-4o ("o" stands for "omni") is a step toward more natural human-computer interaction — it accepts any combination of text, audio, and images as input, and generates any combination of text, audio, and image output.

GPT-4o can respond to audio input in as little as 232 milliseconds, with an average of 320 milliseconds, similar to the response time of a human.

Its performance on English text and code matches that of GPT-4 Turbo, with significantly improved performance on non-English text, while the API is also faster and costs 50% less. Compared to existing models, GPT-4o is particularly excellent in terms of visual and audio understanding.

In the final analysis, OpenAI is actually raising the threshold for communication and exchange between people and AI by continuously improving ChatGPT's interactive capabilities, and constantly making OpenAI's products better popular.

If we summarize and define OpenAI's improvement in interaction with the previous development of the Internet, their common ultimate goal is actually to make AI more popular, and truly make AI a new way of life.

Therefore, if we want to find new enlightenment from OpenAI's press conference, it is undoubtedly another aspect that deserves our attention to continuously popularize AI and continue to move forward in the direction of the Internet.

As OpenAI CEO Altman said in his blog after the press conference:

"I'm very proud that we're able to offer the world's most advanced model in ChatGPT for free, all without ads or other distractions. Initially, OpenAI's vision was to develop AI technology and use it for global benefits. However, the reality is that we have developed AI technologies, and others have used them to create outstanding results that benefit the world. As a business, we have a lot of services that we need to charge for, but that doesn't stop us from supporting the provision of top-notch AI services to billions of users around the world."

"The new voice (and video) mode is the best computing interface I've ever used. It gives the impression of AI in a movie, and people can't help but marvel at how real it is. Achieving human-like responsiveness and expressiveness marks a major shift. The original ChatGPT has already demonstrated the potential of language interfaces; And this new technology has made a qualitative leap in experience. It's responsive, smart, fun, natural, and functional. Before, my conversation with a computer had never felt so natural; But now, I finally feel that way. As we gradually add personalization options, access to personal information, the ability to perform actions on behalf of users, and more, I can really foresee an exciting future where we can do more things with computers than we could have done before."

epilogue

If you must look for the information conveyed by OpenAI's short 26-minute press conference, OpenAI is on the road of the Internet, which is undoubtedly clearer.

It can be said that through such a press conference, OpenAI has completed a perfect transformation from an AI evangelist to a practitioner.

For AI, in fact, it needs to be free, universal, and commercialized.

至此，OpenAI开始open AI。

The reason is that only after such a popularization and commercialization, the development of AI is not just a existence that stays in the laboratory, not just a niche existence, but has become an existence similar to the Internet, which can be deeply integrated with thousands of scenarios and industries.

With this as a start, ChatGPT can truly become the "iPhone moment" in the mouth of Nvidia CEO Jensen Huang, and GPT-4o is undoubtedly OpenAI's singularity moment.

-ENDS-

Author: Meng Yonghui, senior writer, columnist, industry observer, well-known KOL, digital economist.