After the huge popularity of Sora, four judgments made by Sam Altman will determine the "battle for the entry point" of AI.
On September 30th, OpenAI released Sora 2 and launched a brand - new app simultaneously, introducing a portrait authorization mechanism.
Within less than a week of its launch, Cameo emojis flooded social platforms. Altman's AI image appeared in countless group chats, moments on WeChat, and creator communities. Sora quickly evolved from a technical demonstration to a phenomenon - level product.
However, this time, OpenAI has initiated the "battle for the entrance" in the AI era.
On October 8th, Sam Altman appeared on the a16z podcast and comprehensively explained OpenAI's strategic direction for the first time: "We won't just release technical demonstrations. Instead, we'll let society experience in advance what's coming. The entrance to AI is no longer a dialog box but the generation of an entire segment of images, and even thinking one step ahead for you."
He revealed that the video is just a prelude, and the real entrance revolution comes from four judgments:
Videos becoming interfaces, models becoming scientists, Agents evolving into "zero - employee companies", and building self - owned AI factories.
These four judgments are determining the direction of the battle for the AI entrance.
Video: The New Eyes for AI to Understand the World
You can regard Sora as an interface that continuously renders videos, a new way to model the world. ——Sam Altman
Sora has become popular, but its significance is not just about generating videos.
Altman's judgment is clear: The value of Sora lies not in the beautiful images but in teaching AI to understand the physical world.
In the past, AI could only read text and view pictures, which was a static form of cognition.
Now, videos enable AI to start understanding actions, space, and causal relationships. This is a qualitative change in the way of cognition.
Altman gave an example: Instead of opening a webpage and asking "What does this passage mean?", you shoot a video in the real world, and AI can automatically understand the picture, know who is moving and what's happening, and even predict "what problems might occur next".
From looking at pictures to reading actions, the dimension of AI's understanding has changed.
However, OpenAI's release of Sora has a deeper consideration: To let society adapt in advance to the upcoming reality. Soon, anyone will be able to use AI to generate videos that are indistinguishable from real ones.
Altman's exact words were:
"The emotional resonance of videos far exceeds that of text. Text can deceive you once, but videos strike at the heart."
When AI videos are everywhere, the impact will be greater than expected. Society must build immunity as early as possible.
From a technical perspective:
- Sora is not just a content tool but a new way for AI to observe the world.
- This pair of "eyes" is becoming a new entrance. Instead of you inputting questions, AI actively observes, understands, and thinks one step ahead for you.
Video is becoming a crucial training ground for AI to reach AGI.
The Value of AI Lies Not in Answering Questions but in Active Thinking
Sam Altman mentioned a change in this interview: For the first time, we see AI starting to come up with new ideas in scientific research. It's not about summarizing others' opinions but coming up with solutions that have never appeared before on its own.
Some capabilities of GPT - 5 have crossed the boundary of daily tools.
It no longer just writes emails and polishes copywriting. Instead, in mathematical, physical, and biological research, it provides derivation paths that even scientists haven't thought of.
Altman said:
"We used to think that the Turing test was the ultimate standard for AI. What happened? It passed the test without us even noticing. The real major turning point is when AI starts doing 'things we can't do'."
For example:
- Helping calculate complex formulas in physical research
- Finding new proof ideas in mathematical problems
- Helping to establish hypothetical models in life sciences (even if it's not 100% accurate, it dares to propose)
In the past, we asked AI: What does this mean?
Now it starts to actively tell us: Maybe we can think like this.
Altman's standard for AGI is clear: Only when AI can make scientific discoveries can it be considered true general intelligence.
Now, it has already begun.
Many people still have the impression that AI can only write code and draw pictures. But within OpenAI, the researchers of GPT - 5 are already trying to make it a research assistant or even a research partner. It won't replace scientists, but it can be an "inspiration provider" available 24/7.
He believes that:
"AI's involvement in science may be the most profound change in the next few years."
And what does this imply?
The role of AI has changed: It is no longer just passively answering but actively observing, looking for clues, and proposing possibilities.
From "waiting for you to ask" to "thinking for you".
And this is already happening.
From Zero - Code to Zero - Employee, Agents Reset the Starting Point of Entrepreneurship
We're really betting on when the first zero - employee company will emerge.
——Sam Altman
During this 40 - minute interview, Altman mentioned several times that Agents can now really do the work.
You don't need to understand code or form a team. As long as you can write a sentence, AI can automatically handle tasks, generate processes, and complete execution.
After seeing the process demonstration of OpenAI's internal Agent Builder in the background, Altman sighed:
"These things used to take a long time to complete a year ago, but now you can almost get it done in real - time. I feel that I can't come up with ideas fast enough."
In the past, you might need an operator, a customer service representative, a salesperson, and a data analyst. Now, as long as you can clearly describe what to do, AI can help you complete it all:
- Reply to customer messages
- Organize Excel reports
- Search for information
- Write copywriting
- Call external tools to submit results
OpenAI calls it an Agent: a real "AI colleague" that can execute tasks.
It's not an assistant that chats with you but a real execution layer that can accept requirements and produce results.
Altman gave an example: Someone told me today that AI can already complete a whole day's work tasks. It's amazing. Maybe it can't reach the level of "not needing to be managed for a week" yet, but this goal is not far away.
This is no longer just about improving efficiency.
Instead, a brand - new work unit has emerged: One person plus a set of AI can support a complete business.
Altman recalled that in the past, he and his friends bet on when a one - person - operated billion - dollar company would appear; now their new bet is when a zero - employee company will become a reality.
He has witnessed more and more teams using AI for processes, operations, and even product development.
So he emphasized:
"Many of the changes brought about by AI are not about stronger models but about the way people do things."
When AI is no longer just a tool but an execution layer that can continuously execute tasks, the starting point of an organization is reset.
You no longer need "a team" but an idea + a set of AI.
From Model to Entrance, Full - Stack Self - Building Gives the Right to Dominate
OpenAI is no longer just a company that focuses on models.
In the past, it was regarded as a "model company". Its core ability was to train the technologies behind ChatGPT and Sora.
But in this interview, Altman said bluntly: To create truly useful AI, relying solely on models is not enough. We have to build our own infrastructure and control the way users access AI.
This is not just empty talk.
OpenAI is building one of the largest infrastructure projects in human history, including:
- Collaborating with AMD to manufacture AI chips
- Building large - scale data centers with Microsoft and NVIDIA
- Personally promoting the construction of AI power plants to solve the power consumption problem
- Investing billions of dollars to layout the entire chain from underlying hardware to terminal applications
Why invest so much?
Altman's answer is: If the entrance is in someone else's hands, OpenAI will ultimately have to rely on others' platforms.
This has completely changed one of his long - held views.
He said: I used to oppose a company doing everything from start to finish, thinking it was too cumbersome and inflexible. But now I admit I was wrong. We must control the entire chain ourselves.
The logic is simple: Only by controlling the entrance can AI truly be implemented.
Today's OpenAI is a trinity:
- A research team: Continuously breaking the boundaries of models
- A product team: Transforming breakthroughs into user products
- An infrastructure team: Solving problems starting from power supply
The core idea is: To provide truly useful AI to humanity, we have to build this set of things ourselves instead of waiting for others to help.
Where the entrance is, the right to dominate lies.
OpenAI is no longer waiting for others to provide an entrance but choosing to rebuild from the bottom up.
Conclusion: Whoever Thinks Clearly First Will Control the Entrance
On the surface, this battle for the entrance is about the popularity of Sora and the evolution of video models.
But what Sam Altman is really promoting is a brand - new interaction logic: Instead of you opening an app, AI actively understands, responds, and thinks one step ahead for you.
His four judgments point to the same core:
- Video —— Enabling AI to understand the physical world
- Science —— Enabling AI to actively discover problems
- Agent —— Enabling AI to execute complete tasks
- Infrastructure —— Controlling the complete chain from production to delivery
Future AI won't just answer your questions one by one. Instead, it will observe first, reason first, and present possibilities first.
Altman has already started building this future with his own hands.
Everyone who uses AI, develops products, or builds organizations must also make a choice: In the next step, will you let AI wait for your instructions, or will you let it think clearly first and then make your judgment?
This is the real change in the entrance.
This article is from the WeChat official account "AI Deep Researcher". Author: AI Deep Researcher. Republished by 36Kr with permission.