This may be the conversation that best reflects OpenAI's "true intentions"! Altman: Give us a few months, we're not that crazy, we have plans

Recently, OpenAI released a series of significant updates at DevDay and reached infrastructure cooperation agreements worth over one trillion dollars with giants such as NVIDIA, AMD, Samsung, SK Hynix, and Oracle, accelerating the construction of the AI ecosystem. This is seen as a key step for OpenAI in transforming from a leading AI research company to a core infrastructure and service platform in the AI era, with industry insiders calling it "a full-stack gamble that defines the future."

In this regard, the well-known technology analysis firm Stratechery pointed out in its in-depth interview that OpenAI's move reflects its CEO Sam Altman's clear "unified vision"—to build a ubiquitous general artificial intelligence. The execution of integrating venture capital thinking into the company's strategy, making "ultimate YOLO (You Only Live Once)" style bets across infrastructure, products, and research, is its core capability in leading the AI wave.

On October 8, 2025, Ben Thompson, founder of Stratechery and a well-known technology analyst, had an in-depth conversation with Sam Altman, co-founder and CEO of OpenAI. In the interview, they systematically explored core topics such as OpenAI's unified strategy, the logic behind the trillion-dollar infrastructure cooperation, Altman's unique investor mindset, the platformization ambition of ChatGPT, the blockbuster logic of Sora, and the future of AI and copyright.

Stratechery analyst Ben Thompson pointed out in the interview that OpenAI is positioning itself as "the Windows of the AI era," aiming to be both the main interface for user interaction and the core driver of underlying infrastructure. Although Sam Altman is cautious about historical analogies, the vision he articulated aligns closely with this judgment: a personal AI assistant that deeply understands users and provides cross-platform services.

Regarding the core strategic characteristics of OpenAI, the interview revealed the following three points:

Full-stack layout under a unified vision: The goal clearly points to AGI and recognizes that achieving it requires unprecedented massive investments in computing power infrastructure, user products (such as ChatGPT), and cutting-edge research simultaneously.

Investor mindset driving capital allocation: Altman admitted that his venture capital background is "very useful" for operating OpenAI, allowing him to deeply understand how to strategically allocate capital and resources in an exponentially growing world.

Ecological restraint in platform strategy: In promoting ChatGPT to become a platform, OpenAI chooses to empower partners rather than control the user interface, avoiding short-term decisions that are "more user-friendly but detrimental to the ecosystem" to maintain long-term trust.

When discussing the competitive landscape with cloud giants, Altman exhibited a perspective that transcends zero-sum games.

"We have many large-scale, well-funded, and powerful competitors. I don't think this will be a winner-takes-all market." Altman stated that future AI services will be a hybrid model that integrates consumer and enterprise, where users will establish a relationship with a core "AI assistant" and interact with it through ChatGPT, APIs, and various devices.

The following is a summary of the key points from the interview:

OpenAI's Unified Vision

B: We only have 30 minutes, and I'm worried that if I try to read out all the transactions and announcements from OpenAI last month, it will take up the entire time, and I feel that what I'm saying might not be an exaggeration. To save everyone's time, could you summarize the overall idea behind all these announcements—from what you announced at DevDay to these infrastructure deals? Is there a sentence or a paragraph that can encapsulate everything?

SA: We are working to build powerful AI, AGI, and superintelligence, and then deploy it in a way that truly benefits humanity, allowing people to use it for various purposes. This requires a significant investment of resources in infrastructure, products, and research. That is our goal.

B: Today I presented my own understanding of this unified theory on Stratechery. You are positioning yourselves as the Windows of the AI field—I, as an analyst, would take this as a compliment rather than a derogatory term—on one hand, you provide user interfaces, or at least APIs for all terminal interactions aimed at consumer and enterprise users; on the other hand, you are also the target customers of this large-scale infrastructure construction. This is somewhat similar to the relationship between OEM manufacturers and processor manufacturers back in the day; do you think that's a fair statement?

SA: I always find it difficult to understand historical analogies because I get caught up in the differences.

My thought is that most people will want an AI service, and this service needs to be useful to them throughout their lives. So you would use ChatGPT, but you would also want it to integrate with other services, which means you need to integrate other applications into ChatGPT. We need an API business because you would want to be able to log into services built by others using OpenAI, you would want a continuous experience, and you would want it to still understand you, have your information, and know what information can be shared and what cannot. So we want to build such an AI assistant for people, and that requires a lot of parts to be integrated together.

As for the infrastructure aspect, I spend most of my time on that now. Building enough infrastructure to meet our current needs is very challenging and interesting; for me, it's an exciting new challenge, but there are many factors to consider.

In terms of research, I have never been as confident in our research direction as I am now, which is why we are investing heavily in product infrastructure development.

Yes, this provides a good framework for the three topics I want to explore

Infrastructure Transactions

B: Let's start by talking about infrastructure. Imagine announcing a deal that could increase your partner's market value by 20% or more—how does that feel? For anyone who can release a press release containing OpenAI elements, it's like turning stone into gold.

SA: It's very strange. This is a very new phenomenon. I don't think it will last long, and I even think it's a bit silly, but we have to adapt. Three years ago, we were just a research lab, ChatGPT hadn't been launched, and we didn't have to consider the market impact. We've made rapid adjustments in recent months, but it's really strange.

If people want to understand the construction of artificial intelligence—let's not call it a bubble; let's use another "B" word (Buildout)—do they basically need to understand you? This seems to be the other side of the question, but if you're going to be a unified layer that allows people to have artificial intelligence and you're going to influence everything from infrastructure to applications, aren't you the core? By core, I mean it's complementary, but it's actually a double-edged sword.

SA: We have many large-scale, well-funded, and powerful competitors. I don't think this will be a winner-takes-all market, far from it.

B: Do you think there will be a winner-takes-all situation on the consumer side?

SA: I bet there won't be. Of course, it would be great if we could capture a large market share, but I'm not sure. We will definitely strive to create a product with a massive user base that brings value to many users, and we have ambitious plans in this regard. But I think you'll see artificial intelligence—my favorite analogy for artificial intelligence, and also my favorite historical analogy, is the transistor, which I think we've discussed before—it will permeate all consumer products and enterprise products.

B: If these dynamics are true, then isn't the worst practice not to spend as much money as possible, or not to spend the money that others are willing to give you?

SA: I don't know if that's the worst practice. We've been trying to explain this to the outside world for a while, but I think it wasn't until recently that the outside world really took our perspective seriously. We will invest heavily in infrastructure; we are going to make a company-level bet that now is the time to do so. Given our current R&D, business, and product situation, and what we see in the landscape, is this the right decision? We will find out, but this is a decision we are about to make.

B: Yes. Byrne Hobart recently wrote a book called "Prosperity," which expands on Carlota Perez's theory of bubble investment dynamics. I think his most important addition is that you need a lot of parallel investments and developments, and bubbles are what allow these investments and developments to happen simultaneously. Do you think this is a burden you have to impose? Not just that things need to happen simultaneously, but they actually can't happen in sequence at all?

SA: Yes. You have to manufacture electronic components, build all the physical infrastructure, purchase power equipment, and everything outside of data centers. You need to build chip manufacturing capacity on racks, have consumer demand, and have companies willing to pay for it. This involves many aspects and requires simultaneous consideration.

B: Yeah. What do you think about chips? You just announced a partnership with AMD. Does Jensen Huang know that NVIDIA will indirectly invest in AMD? (laughs)

SA: I don't know.

B: Well, the issue is that both NVIDIA and AMD's chips come from the same company, so there is another independent entity in the value chain, which is TSMC. Do you think it is necessary, responsible, or an opportunity to also expand the market there? Speaking of Intel—

SA: I hope TSMC can expand its capacity.

B: Did you think I was asking about multi-chip suppliers?

SA: I think it is necessary to encourage TSMC to increase its investment and capacity.

Investor Mindset

B: These deals are worth astronomical sums; I remember the Financial Times just estimated the total to be as high as one trillion dollars. Who do you think will foot the bill? Is this related to the nature of these deals? Do you guarantee to purchase the end products, so these companies need to invest?

SA: Yes, I expect OpenAI's revenue will be sufficient to cover this cost.

B: But at the same time, these companies must be taking on debt. Do you see yourself as a financial guarantor helping them secure better rates? Would you say, "Look, here is a guaranteed purchase agreement, and we will leverage this agreement"?

SA: Yes, we will provide financing assistance. We are developing plans to provide them with the necessary financing even before these companies generate revenue. We have some good ideas.

B: For a long time, you have been regarded as one of the greatest financiers in Silicon Valley. Did you realize that this skill would play such an important role in this field?

SA: In the early stages of my career, I learned some things, and while I wouldn't say I mastered them, I was pretty good. I didn't think these skills would be so helpful later in operating OpenAI, but it turns out they are very useful.

B: Is this a top priority?

SA: It is definitely one of them.

B: What else is on the list? I'm quite curious.

SA: First of all, if I could do what I'm really good at, I would be an investor rather than an operator. In many ways, I don't have the capability to run a large company, but I have been trained as an investor and truly understand how to allocate capital in a world where exponential growth may occur. I have witnessed miracles multiple times and know how to select projects and talent.

To some extent, OpenAI is indeed investing in a series of entrepreneurial projects, but they are all internal projects, or sometimes products, like Sora. This is quite unusual; I initially thought it wouldn't be beneficial for company operations and even thought it might be bad, but it turns out it is very useful I can dedicate a special episode to explain all the details in this area, but this is another example.

B: Yes, I think this is indeed very interesting. One point is to understand the dynamic relationship between unlimited upside potential and limited downside potential, which is very meaningful. But in this context, what are the things worth investing as much money as possible? Are there things you absolutely would not do, or things you think should not be done?

SA: Many. The difference from making a lot of early-stage startup investments is that I have many product ideas I want to invest in and new things to try, but we ultimately only choose a small portion of them, which is a bit regrettable. But I think it’s important to do the things we consider most important to the extreme.

B: I’m very interested in hardware in this regard. You had an interview with Jonathan Ive yesterday, which was very exciting because there might be new hardware manufacturers joining the competition. What I want to ask is, if ChatGPT, OpenAI, or other similar tools are to become effective assistants for all your devices, they must be able to operate across devices and not be limited to the devices you develop. Have you considered this, like worrying about becoming too reliant on your own hardware or scaring off other potential partners?

SA: It must run on all devices, including browsers and everyone’s phones. Obviously, it must be ubiquitous. I am a device enthusiast myself and have always wanted to make devices.

B: I know, but that’s the temptation: everyone prides themselves on being a device geek, and then when they get the devices, they get a bit distracted and stop thinking, “Actually, no, we need to be ubiquitous.” There’s another large company in Silicon Valley that has probably been trying to poach some of your researchers, and I think they can easily fall into this trap.

SA: Let me try it once; it should be a bit of fun. No, I think we have the opportunity to do something truly extraordinary; it obviously won’t replace other devices, and we need to balance all devices. But the quality of thinking about new hardware has always been very...

B: Stagnant.

SA: Yes, describing it as “stagnant” is very appropriate; everyone only has those few old clichés. They are all trying. I think the iPhone is the greatest consumer hardware product to date, so I understand why we fall into this mindset, but I believe artificial intelligence does show the possibility of creating new things.

B: Let’s delve a bit deeper; one surprising deal is the one with Samsung and SK Hynix. Obviously, memory is a major bottleneck for building these chips in the future. Is this related to the AMD deal? This is also one of the concerns: “Where does AMD get the memory needed for these chips?” Or are you still developing chips in-house?

SA: Give us a few months, and everything will become clear, and we can discuss this issue comprehensively. We are not as crazy as we seem; we have plans

B: To be honest, I think it all makes sense, which makes me a bit scared of this interview. I feel like what you're doing, although it seems a bit chaotic, is actually very organized.

SA: What's so scary about that? Isn't it better than other options?

B: No, because I think there might be a risk of confirmation bias. After talking to you, I feel like I have a deeper understanding of you. I previously complained that your use of APIs might be wrong, even though you always encounter capacity limitations, which has always frustrated me. For example, I would think, "If you put all your resources into XYZ, things would be much better."

SA: Our capacity has greatly improved!

B: Yes. You're right, capacity will definitely improve. It's like complaining that the processor speed isn't fast enough—speed will definitely get faster, and capacity will definitely be in place. So I now appreciate the idea of "Hey, why not try to be all-encompassing and omnipotent?" Opportunities are fleeting, and you only live once.

SA: First of all, this is a once-in-a-lifetime opportunity for all of us, and we will do our best.

B: This is the ultimate version of YOLO (You Only Live Once), YOLO AI; perhaps this should be the title of a new article.

SA: But I believe we have a vision, which I mentioned before. If you look at everything we do from this vision, once we complete this series of new deals and clarify our direction and reasons for development, everything will become clear. People might say this is too ambitious and too risky. But at least, the reasons for doing these things will ultimately make sense.

Applications in ChatGPT

B: So you mentioned before that the emergence of ChatGPT was a surprise. I remember you seemed to mention it a few minutes ago. Looking back, I think I asked you a similar question before, but I've been thinking about it recently. When did ChatGPT transition from being a "surprise" to "this thing actually exists," and then to "actually, it has great potential"? Because we now have so much demand, for example, the API is great, authentication is great, users can log in with ChatGPT, access their memories, etc. But just this week, I started thinking, "Why don't application developers directly integrate their applications into ChatGPT?" Instead of going through the trouble of developing all sorts of strange integration solutions that lead to various problems, why not concentrate all functionalities on one platform where users will be? When did you realize you had the ability to achieve this?

SA: There were two surprises. One is that I expected ChatGPT to perform well, but I didn't expect it to be this good. The rapid rise of ChatGPT is remarkable. The second surprise is that some large tech companies, which I thought were generally quite capable, made some mistakes in the period following that, which allowed us to gain some advantages before others took action

B: But after those two incidents, we developed and executed a strategy for two and a half years, and during this time, our plans have basically progressed as expected. It takes time and must be done step by step, but we have been thinking about how to build the infrastructure, as well as thinking about applications, ChatGPT, and how to unify APIs and consumer businesses, so this part of the progress is basically in line with our expectations.

When mentioning other tech companies, I think most people would first think of Google. I feel that Google is doing quite well now; was its previous company called Meta?

SA: I don't want to name specific companies here.

B: Didn't you say you wanted to have fun? I specifically brought this for you.

SA: Google is doing exceptionally well now, but clearly, they didn't do that when ChatGPT first entered the market.

B: Yes. In the long run, in terms of ChatGPT, is the consumer market larger or the enterprise market? Are you surprised by the high penetration rate of ChatGPT in the enterprise market?

SA: There are indeed some differences. But going back to the example of Google, using Google services at home is different from using another company's services at work.

B: Yes, that analogy is quite apt.

SA: Enterprises require different security measures and different data access permissions. Perhaps in the consumer space, people are more concerned about the quality of healthcare; while in the enterprise space, people are more concerned about the quality of coding, legal document processing, or other aspects. But I believe the degree of integration between the two will be higher than people imagine, and ultimately, you will have the AI tools you use.

B: That's right, I think Google is a great example; there's really no difference, you use Google for everything. At work, on the company computer; at home too. No employer would block Google; they could easily set Bing as the default search engine, but everyone still ends up visiting google.com. So, what is the overall development prospect of OpenAI? This goes back to the API business itself, or do you think APIs are aimed at emerging companies, startups, or enterprise-level products? Or do companies like Microsoft and Azure still play an important role in this field?

SA: I want to emphasize again that we are striving to create an AI service that can assist you anytime and anywhere, rather than a dichotomy between ChatGPT Enterprise and APIs. I believe you will establish a connection with this AI tool: sometimes you use it in ChatGPT, and it does a lot of work for you; sometimes you use Codex in the terminal; sometimes you use its API in other services.

But I think in a few years, it will be like this: "Well, I have an entity that does a lot of useful work for me across all these different services," and I’m happy to have APIs, happy to have ChatGPT, happy to have my new devices, but you will feel that you have established a relationship only with this entity that helps you

B: At the ChatGPT conference, you showcased a series of excellent partner applications, and you've made quite a few attempts in this area. I'm enthusiastic about all these applications, partly because I can understand your overall vision. You first launched the connectors, or I can't remember what the first one was, it seems like it was something like plugins, and then you launched GPT.

SA: The ChatGPT plugins didn't succeed. The GPT itself is fine, and the usage rate is surprisingly high, but mainly used internally within companies or in personal workflows. I hope this new feature will be more useful; if not, we will continue to try.

B: So, with these applications, do you feel like you've returned to the original intention of "we have so many users, it would be better if they could use Zillow directly in the app"? Users go elsewhere for services, and Zillow says, "Oh, I would prefer them to use our app; we spent so much time developing this app." Do you feel you have the authority to lead? "It's a better experience for users, and since users are here, if you're not there, someone else will be"—you do have that authority; leading doesn't sound great, but if the user experience is better, it's beneficial for everyone, right?

SA: No. My early career training was also very useful in this regard. We could have made a version with a better user experience, but it would have been terrible for partners.

B: What would that look like?

SA: Take Zillow as an example. If you say, "Hey, ChatGPT, help me find all the houses that meet these criteria," and we say we want to control the user interface, what would happen?

B: Oh, right. So the Zillow app doesn't even have a display layer; you can only see the search results.

SA: Yes. But I strongly feel that by doing this, the entire ecosystem would benefit, especially allowing new startups to rise quickly. So the approach we take is to establish a close connection between users and other websites. We will directly call them by name and recommend them; they will take over the user interface and link their accounts. We could have done something more user-friendly but detrimental to other companies, and I really don't want to do that.

Instant Checkout

B: Yes. Regarding instant checkout, I find one point very interesting. Speaking of analogies, when you first launched this feature, some people discussed how certain social networks abandoned in-app checkout because the conversion rates were low, while e-commerce sellers wanted users to use their websites directly. Is this somewhat like a monopoly of existing businesses? They have some effective methods, and you think you can do better. You provide tremendous value to long-tail users, which is worth the investment. If you prioritize users who offer instant checkout services, assuming other conditions are the same, and your survey results are unbiased, then users have no choice but to accept this feature?

SA: There are two points. First, I don't understand why other websites have given up on instant checkout. As long as merchants can connect directly with consumers (that's what we do), the user experience is better, and it's more beneficial for merchants, which is good for everyone. Second, maybe we will learn from others' lessons and realize we were wrong in this regard, but for now, instant checkout seems really good.

Regarding the second part of the question, I want to say that if we can't do our best to provide the best answers for users in anything we do, then users' trust in ChatGPT (which is currently very high overall) will plummet. Therefore, we have every reason to avoid that situation.

The reason people like ChatGPT is that they don't like it in the same way they like products from other large tech companies. I think even when ChatGPT makes mistakes or has hallucinations, you know it's trying to help you, and you know your interests are aligned.

B: Yes. I suddenly realized that I'm very excited about this topic; I love everything related to the long tail. My business falls into the long tail category in many ways, and I've always been a loyal supporter of Meta, one reason being that I believe they provide immense long tail value. It helps merchants find customers and show them products they might not have even thought they wanted. After returning to the U.S., I was pleasantly surprised to find that Instagram's advertising effectiveness was simply unbelievable. I thought it was already good, but now it seems even better. I can hardly believe how much I've bought, like home goods and such. It feels like ChatGPT does the exact opposite—it allows consumers who only have a vague idea of what they want to precisely find specific products they've never heard of before. My thought is that because you want to achieve 100% trust, you won't really profit from these transactions. If I find a customer and they genuinely want me to complete the transaction on the website, you can totally accept that. Tech companies generate a lot of consumer surplus, which is normal. The real profit comes from if your product is available at 10 retailers, then if you are a retailer using ChatGPT for instant checkout, and ChatGPT can earn some revenue from it, then those retailers are your priority customers. What I mean is, that's my understanding of your approach.

SA: First, regarding Instagram ads, it really made me realize that maybe ads aren't always that bad. I love Instagram ads; they bring me value, and I've discovered things I would never have found otherwise and bought quite a few things. I genuinely like Instagram ads. I think there are many aspects of Meta that I respect, but their excellence in this area was an unexpected surprise for me. Besides that, I've always thought of online advertising as a kind of tax

B: I think the problem is that search is essentially a tax. Usually, organic search results will contain what you want, and then I have to spend money on ads to rank at the top. I have always supported Meta, and I think this is the advertising model we should welcome.

SA: I agree.

B: So, in this context, how do you view your business opportunities?

SA: I believe we can definitely develop some great advertising products that benefit users and positively impact our relationship with them. But I don't know exactly what that is yet; I'm not at the point of saying, "This is our advertising model."

B: But affiliate marketing seems obviously profitable; you don't have to worry about it eating into your advertising business.

SA: Yes, it looks like an obvious win.

Sora

B: Back to the unexpectedly discussed topic of ChatGPT, you expressed great confidence in Sora's success in a post (I'm not sure who wrote it; you referred to OpenAI), and I didn't think much of it at the time. I was completely wrong; I completely overlooked this. But do you really think you have established a repeatable mechanism for successful development?

SA: First of all, the team here is very talented. The best way to continuously create popular products is to have truly talented people who work hard, conduct in-depth research, and build great products around those research findings.

B: Secondly, Fidji Simo recently joined us; she is an outstanding leader and excels in many areas, product being just one of them, but she is indeed very impressive in product. Handing over all product-related work to her makes me a bit sentimental and nostalgic, but I have many other things to do, so it's okay. However, the thought of having someone much better than me in charge of products, and knowing she will continuously launch hit products, excites me greatly.

B: You just mentioned the brand halo effect, which is that people trust ChatGPT, and I think that's very important. If ChatGPT has a problem, people will say, "That's okay, try my little brother." How important is this in the mechanism for developing hit products? You launched Sora, and people think it's cool, are willing to give it a chance, and will forgive any shortcomings. Your competitors may be completely different in this regard.

SA: I don't think so. It's just a good product; we captured the key and grasped the core dynamics. OpenAI is not a very well-known brand; ChatGPT is a very well-known brand, but the logo of Sora is shaped similarly to that of ChatGPT, which may be overlooked by everyone.

B: I missed that; I didn't notice.

SA: There may not be many people outside the tech circle, but I hope some of them will notice. However, I think we just seized the opportunity, so it spread because people thought it was cool. We won't launch products in large quantities—when we launch a device, I believe people will buy it because it's cool, not because of the brand halo, but I believe the brand halo will help us

B: Have you considered integrating Sora into ChatGPT, or was it clear from the beginning that it would be a standalone application?

SA: We did consider it, but it's a completely different way of thinking. ChatGPT is a very personal thing for people.

B: Plus, with some social characteristics, it's like—

SA: I think the feeling at the time would be, "Oh, do I still want to share my deepest issues?"

B: Exactly. Did Sora greatly accelerate the necessity for a monetization model? I can't imagine how much it would cost. Can it solve the problems it creates, like providing better advertising potential?

SA: As things develop, some interesting things will emerge. I'm not sure, Sora might have a suitable advertising model; I can immediately think of many great ideas. However, many people just use it to create funny memes and send them to their three friends, and no advertising model can support a user base of that scale. I think, in some cases, people might need to pay for several generations.

B: If we've been underestimating— you also mentioned that people in the AI field have talked about "AI will help people become more creative," which sounds like a cliché to justify what you want to do, but maybe it really is true. I think that's one of the inspirations I've gotten from the Sora project. In fact, when DALL-E was released, I wrote an article about how to separate ideas and arguments and integrate them into a broader framework of printing, oral communication, reading, etc. I feel that, even though I wrote that article, I still underestimated its importance. Have you always been aware of this?

SA: This is another valuable lesson I learned during my venture capital days. There is a huge potential demand for creative expression in the world, and if you can provide some tools to help people better fulfill that demand—maybe it's hard to create a great TikTok video or any other content in traditional ways, but if you provide some tools that allow people to quickly turn ideas into creative works, it can satisfy a deep-seated human need, and I've witnessed its success countless times.

B: Is this a human need that exists independently of sharing with others? Sometimes, you just want to create for the sake of creating.

SA: Sometimes it really is. People enjoy appreciating others' art, and those who create art also like to showcase their work to others. So, there is a kind of collective appreciation involved. You enjoy creating for others.

This is also another reason why I'm not worried about the long-term impact of AI on employment. We like to do things for others, we like to be recognized and appreciated by others, so we will find things to do that benefit others.

B: Yes, on Sora, we've observed some interesting dynamics, but one thing that was clear from the beginning, and one of the reasons we excitedly got involved and didn't think it would turn into a meaningless stream of junk, is that we hope that if we can significantly lower the startup costs of creation, more people will engage in creation I haven't looked at the data for several days, but in the first few days, about 30% of active users were active creators, which is simply incredible.

This is my insight. And that old 90/9/1 rule. I think maybe this rule exists simply because the barrier to creation is too high; in reality, if given the opportunity, far more than that 1% would create.

SA: I completely agree. We invested a lot of energy in developing this product because it is difficult to create excellent works compared to other AI video creation tools. These tools need to handle a lot of data, and we strive to make the model understand user needs well and create high-quality works. I believe our efforts have ultimately paid off.

B: Do you think social networks have actually ceased to be social networks and have merely become platforms that aggregate various entertainment information? Does this shift provide an opportunity to re-establish social connections?

SA: It's strange, but I always hope we can do more. I hope we can create outstanding entertainment products and outstanding social products at the same time. I feel they are actually quite different. Of course, there are some overlaps between them.

Which one is Sora? Is it a social product or an entertainment product?

SA: It is primarily an entertainment product, but the social use of sharing funny memes among these small groups is more than we expected, and more than I anticipated. However, all social products have more or less leaned towards entertainment products, and instant messaging is the new way of socializing, right?

B: Yes.

SA: I believe there are still many excellent social products that can be developed again, but the incentive mechanisms of the global economy have pulled them all towards B2B entertainment products.

Exactly. But if you are starting from scratch, you have a long way to go before considering these economic incentives.

SA: Maybe next we will develop a great social product.

Copyright and Artificial Intelligence

B: The two recent viral events you experienced can be said to be related to copyright holders. One was the image generation from Studio Ghibli, and the other was Sora's interaction with Pikachu and SpongeBob, sometimes even appearing together with them. You ultimately acted quickly to stop these infringements, but does this really prove that your philosophy of "acting quickly and breaking the norm" is correct?

SA: No. Clearly, before Sora, we communicated with many copyright holders. In terms of image generation, we still did a lot of things that we wouldn't do in videos; it's like you have to respond to the joint development of society and technology. But it turns out that the impact of video on people, especially copyright holders, is completely different from that of static images.

B: Why is that?

SA: I don't know. They are emotionally different. If you create a funny image instead of a real video, the video feels more real and vivid, and it is easier to evoke emotional resonance. Copyright holders want to adopt different approaches. Most copyright holders I have interacted with are very eager to have their content here; they just want to set more restrictions than with images because videos feel different

B: I predict that there may be discussions now about "I don't like the content in the videos," and in a year, or perhaps even sooner, there will be discussions like: "OpenAI is unfair to me, it hasn't put my content in enough videos, we need better rules," because people want to establish a deep connection with their fans.

Well, another point I thought of is that you previously mentioned you are not too worried about work because people enjoy socializing and helping each other. Also, take these viral videos as an example; although copyright owners are concerned about artificial intelligence, the reason they become the focus of attention is that they are common and symbolic things that people naturally pay attention to. As artificial intelligence creates more and more customized content, those common and well-known things will become more valuable, which seems to be a huge opportunity.

SA: Obviously, there is a bias in my sample here, and I am sure many people do not wish for this, but the number of copyright owners who contact me saying "this benefits me, I need to ensure I get enough returns" far exceeds those who say "I don't want this." Most of them now want to set rules, such as "I don't want my character to say such offensive things," while some adopt a completely "life is short, enjoy it while you can" attitude. But I think you will see a shift in focus to those who believe "this will bring me value, rather than the other way around."

Feedback and Future

B: I feel that OpenAI sometimes overhypes on Twitter, which may make some people uncomfortable. I feel this might explain why my view on Sora was completely wrong. But is it really a problem of overhyping by people? Or is it because the user base is too large, causing the product to take longer to transition from the lab to practical application? And when the product is actually put into use, it may have already been weakened or far less powerful than expected? Upon reflection, this might be an infrastructure issue.

SA: There are a few points I want to make. First, we might need to hype less on Twitter; we are just too excited. Second, if you look at our current achievements and compare them to the expectations of most people five years ago, it must be said that this is impressive to some extent.

B: You should earn trust. I think that is fair.

SA: I don't want to get bogged down in the specific definition of the Turing test. According to the general public's understanding, we have somewhat passed the test, and in 2020, most people did not think this would happen. So, if in the next five years we really make breakthrough progress in the field of artificial intelligence, with new discoveries and significant scientific achievements, and if our previous promotions may have been a bit over the top, that would indeed be off-putting, and we should reduce such behavior. However, I believe that given the actual progress, we should be granted some leniency.

B: What do you think about the feedback mechanism? I believe that one of the real challenges faced by large consumer applications in the social media era is that even if only a small portion of users are dissatisfied, their complaints can be very loud, creating an illusion of a large user base. How do you view this situation? How do you view feedback on actual effective functions? It seems that ChatGPT-5 has encountered this situation. The simplified version of ChatGPT-5 is excellent. The data you shared indicates a significant increase in user numbers, which has been a major issue in the past, especially for free users. But then, you made a lot of adjustments and modifications in a short period of time.

SA: First of all, listening to user feedback, analyzing data, and making quick changes is a good thing, not a bad thing. We are doing both now and will listen to user feedback.

B: That's right. I think there needs to be a balance between data and listening to a small portion of voluntarily chosen user opinions.

SA: I care more about data, but I believe in this principle: if anecdotes and data are inconsistent, you must work harder to find the answer because anecdotes are often correct.

B: Is it more difficult to manage online sentiment nowadays, as the narrative on platforms like TikTok evolves constantly, and you don't want the narrative of "ChatGPT-5 is bad" to take hold?

SA: That is definitely a challenge. What I have truly learned from the GPT-5 project is that the way active AI users on Twitter use AI is completely different from how most ordinary people in the world use AI.

B: Correct.

SA: It is possible to create a product that meets the needs of both sides, but we have not paid enough attention to this matter yet.

B: Based on this, are we still underestimating the potential of the subscription model? I think, in terms of creativity, people's creativity is stronger than we realize. Are people more willing to pay than is generally thought?

SA: I think ChatGPT has already proven this.

B: I think so too.

SA: As for how far we can push this, I don't know, but at least it's a cognitive update.

B: Yeah, this is all great. Do we have enough power?

SA: As I said, please give me a few more months to get everything in order. I promise to arrange everything properly.

B: Okay. Next, we need to focus on a series of transactions. Sam Altman, it's great to talk to you again. This month has been amazing, and the product launch was very successful. Congratulations!

SA: Thank you very much. Goodbye