Title: Sam Altman and the Future of OpenAI: A Journey Towards General Artificial Intelligence
Introduction:
In this interview, we delve into the world of Sam Altman, a prominent figure in the tech industry and the former president of Y Combinator. He is also the co-founder and CEO of OpenAI, an organization dedicated to ensuring that artificial intelligence benefits all of humanity. We explore Altman’s experiences, his thoughts on the future of AI, and the importance of open-source technology.
Host:
I first met our guest, Sam Altman, almost 20 years ago when we were both developing different projects with the support of Sequoia Capital. I invested in Uber, while Sam invested in a relatively unknown fintech company called Stripe. It was an exciting time for both of us.
Sam:
You invested in Uber? I’ve never heard of it before.
Yes, indeed.
You should write a book, Jacob.
Host:
Sam and I were part of a small experimental fund at Sequoia Capital, which turned out to be one of their most successful funds. I heard that millions of dollars turned into over $200 million. Sam also worked at Y Combinator for some time and served as its president from 2014 to 2019. In 2016, he co-founded OpenAI with others, aiming to ensure that general artificial intelligence benefits humanity. In 2019, he left YC to join OpenAI full-time as CEO.
On November 30, 2022, things got very interesting. It was the day OpenAI launched ChatGPT. In January 2023, Microsoft invested $10 billion. Then, in a crazy five-day period in November 2023, Sam was fired from OpenAI, and everyone was going to work for Microsoft. Heart emojis flooded X and Twitter, and people started speculating that the team had achieved general artificial intelligence. The end of the world was nigh. Suddenly, a few days later, he returned as CEO of OpenAI.
According to reports, in February 2023, Sam was seeking to raise $7 trillion for an AI chip project. There were also reports of him seeking $1 billion from Masayoshi San and co-founder of the iPhone, Johnny Ive, to create an iPhone killer.
Meanwhile, ChatGPT was getting better and becoming a household name. It had a tremendous impact on the way we work and complete tasks. Reports stated that it reached 100 million users within just two months, the fastest for any product in history.
Take a look at OpenAI’s exponential revenue growth. It was reported that their ARR reached $2 billion last year. Welcome to the All-Powerful Podcast.
Host:
I’ve noticed that the entire industry is eagerly awaiting the release of GPT-5. From what I understand, it might be launched sometime this summer, but the timeframe is quite broad. Can you shed some light on this for us?
Sam:
Releasing such a major new model will take some time. I believe it will be fantastic when we do so. We will carefully consider how to proceed. It might be released in a different way than our previous models. Furthermore, I’m not even sure if we will call it GPT-5.
What I want to say is that since we released GPT-4, especially in the past few months, many people have noticed its outstanding performance. I think it reveals a better understanding of the world, that it’s not just a linear progression from 1 to 7, but rather an iterative process of using AI systems that continuously improve. I believe this is both a better technical direction and easier for society to adapt to. That’s the path we’re heading towards.
Host:
Does this mean that we won’t have long training cycles but instead continuously retrain or train sub-models, Sam? Maybe you can share with us some possible architectural changes for future large-scale models.
Sam:
One scenario you can imagine is that we are continually training one model. That seems like a reasonable approach to me.
What we’re discussing here is different ways of releasing. Are you considering releasing it to paying users first or slowing down the release to create tension with the red team because the risks are too high now? With so many paying users and everyone watching your every move, you must be more thoughtful, right?
Yes, currently GPT-4 is still only available to paying users, but what we really want to do is figure out how to make more advanced technology accessible to free users as well. I think that’s a crucial part of our mission. Our goal is to build AI tools and make them widely available, whether they are free or less expensive, whatever it may be, so that people can use them to create the future instead of relying solely on the magical AGI in the sky to create and pour it down on us. It seems like a better and more inspiring path. I believe things are indeed moving in that direction. So, it’s unfortunate that we haven’t figured out how to make GPT-4 level technology accessible to free users yet. That’s what we really want to do. I must admit, it’s quite expensive.
Sam, I think two major factors that people often discuss are potential costs and latency, which to some extent limit the rate of killer applications. Then, I think the second factor is the long-term capability built in the open-source world versus the closed-source world. I believe the enthusiasm in this field lies in the passion of its open-source community. An incredible example is what we did for Devon with a crazy demo about five to six weeks ago. Then, some young people released the project under an open license like OpenDevon, based on the MIT license. It performed exceptionally well, almost on par with other closed-source projects. So, perhaps we can start discussing from this point, what is the business decision behind keeping these models closed-source? What do you think the future holds in the next few years?
Regarding the first part of your question, speed and cost are crucial to us. I don’t want to give a timeline on when we can significantly reduce latency because research is challenging, but I believe we can achieve it. We want to dramatically reduce latency and costs. I believe it will happen. We’re still in the early stages of scientific understanding, unable to fathom how it works. Additionally, we have all the engineering aspects in our favor. So, I don’t know when we’ll have intelligence so cheap that it’s immeasurable and so fast that it feels instant to us and everyone else. But I believe we can reach a reasonably high level of intelligence. It’s important to us and to users, as it will unlock many things.
Regarding open-source and closed-source, I think both have their merits. We have already open-sourced some projects, and in the future, we’ll open-source more. But our mission is really about advancing artificial intelligence and finding ways to broadly distribute its benefits. We have a strategy that seems to resonate with many people. Obviously, it won’t work for everyone. And it’s a massive ecosystem. There will continue to be open-source models and people building in that way.
One area I’m personally interested in, when it comes to open-source, is having a really good open-source model that can run on my phone. I don’t think the world has good enough technology yet to develop a good version of it. But it seems like a very important thing to do at some point.
Would you do it? Would you release it?
I don’t know if we will or if someone else will.
What about Llama 3?
Llama 3 running on a phone?
I guess there could be a 7 billion parameter (phone) version.
Yes. I don’t know if it would be suitable for a phone, but…
It should be suitable for a phone.
But I don’t know, I’m not sure if it’s suitable. I haven’t played with it. I don’t know if it’s capable enough for what I’m thinking here.
So, when Llama 3 is released, I think the biggest takeaway for many people will be, wow, they’ve caught up to GPT-4. I think it won’t be equal in every aspect, but overall, it will be very, very close. The question is, you just released 4 not long ago, you’re developing 5 or doing more upgrades to 4. I’d like to hear your perspective on how you stay ahead in the open-source environment. This is usually a challenging task. What are your thoughts?
Our goal is not to create the most intelligent set of weights that we can create, but to create a useful intelligence layer for people to use. The model is just a part of it. I believe we will stay ahead in this aspect, and I hope we’ll be far ahead of the rest of the world in this regard. But there’s a lot of other work in the entire system, not just the model weights. We have to build lasting value the way any other company does, in a traditional sense. We have to come up with a great product and a reason to stick with it and deliver it at a favorable price.
When you started this organization, the goal or part of the discussion was that it’s too important for any one company, so it needs to be open. Then it shifted to no one can see it because it’s too dangerous, so we need to lock it down because you’re worried about it. I’m curious to know if that’s true. Cynics would say it’s a capitalist move. So, I’m wondering what led to that decision from the openness originally. The world needs to see this. Closing it is really important. Only we can see it. So, how did you come to that conclusion?
We released ChatGPT in part because we wanted the whole world to see it. We’ve been trying to tell people that AI is genuinely important. If you go back to October 2022, not many people believed that AI would be as significant as it turned out to be or that it would genuinely happen. A large part of what we’ve been trying to do is put the technology in people’s hands. Now, again, there are different ways to do that. I think there is indeed an important role to say, for example, this is how we’re doing it.
But the fact is, we have so many people using the free version of ChatGPT, we don’t run ads, and we’re not trying to make money from it. We’re just putting it out there because we want people to have these tools. I think it has done a lot of work, provided great value, taught people how to fish, and made the world truly think about what’s happening here. Now, we still don’t have all the answers. Like everyone else, we’re figuring it out as we go along. I think we’ll change strategies multiple times as we learn new things.
When we founded OpenAI, we really didn’t know how things would unfold, what we would make, or what products we would create. I vividly remember the first day, and we were like, okay, now we’re all here. Getting everything ready was challenging. But now, what’s going to happen? Maybe we should write some papers. Maybe we should stand around whiteboards. We’ve been hard at work, figuring out what the next steps are, one step at a time. I think we’ll continue to do so.
Just to confirm once again, so I don’t mishear it, my understanding is that regardless of whether it’s open-source or closed-source, all these models, regardless of what business decisions you make, will progressively get better in terms of their accuracy. Not every company has the capital, but assuming there are four to five, like Meta, Google, Microsoft, or maybe even a startup. They’re all operating on an open network. Then, soon, the accuracy or value of these models might transition to these proprietary training data sources, the ones only they can see. So, how do you see this unfolding?
Our goal is not to make the most intelligent set of weights we can make, but to make a useful intelligence layer that people can use. The model is just a part of it. I believe we will stay ahead in this aspect, and I hope we’ll be far ahead of the rest of the world in this regard. But there’s a lot of other work in the entire system, not just the model weights. We have to build lasting value the way any other company does, in a traditional sense. We have to come up with a great product and a reason to stick with it and deliver it at a favorable price.
When you started this organization, the goal or part of the discussion was that it’s too important for any one company, so it needs to be open. Then it shifted to no one can see it because it’s too dangerous, so we need to lock it down because you’re worried about it. I’m curious to know if that’s true. Cynics would say it’s a capitalist move. So, I’m wondering what led to that decision from the openness originally. The world needs to see this. Closing it is really important. Only we can see it. So, how did you come to that conclusion?
We released ChatGPT in part because we wanted the whole world to see it. We’ve been trying to tell people that AI is genuinely important. If you go back to October 2022, not many people believed that AI would be as significant as it turned out to be or that it would genuinely happen. A large part of what we’ve been trying to do is put the technology in people’s hands. Now, again, there are different ways to do that. I think there is indeed an important role to say, for example, this is how we’re doing it.
But the fact is, we have so many people using the free version of ChatGPT, we don’t run ads, and we’re not trying to make money from it. We’re just putting it out there because we want people to have these tools. I think it has done a lot of work, provided great value, taught people how to fish, and made the world truly think about what’s happening here. Now, we still don’t have all the answers. Like everyone else, we’re figuring it out as we go along. I think we’ll change strategies multiple times as we learn new things.
When we founded OpenAI, we really didn’t know how things would unfold, what we would make, or what products we would create. I vividly remember the first day, and we were like, okay, now we’re all here. Getting everything ready was challenging. But now, what’s going to happen? Maybe we should write some papers. Maybe we should stand around whiteboards. We’ve been hard at work, figuring out what the next steps are, one step at a time. I think we’ll continue to do so.According to sources, you may or may not be able to access certain information, or others may have access to it that you do not. Do you think this will lead to a competition for data, akin to an arms race, once open networks allow everyone to reach a certain threshold?
I disagree with this perspective. I firmly believe that it will not turn into a competition for data, especially when models become intelligent enough. It shouldn’t be solely about acquiring more data, at least not for training purposes. While data may be important because of its inherent value, the most important thing I have learned from all of this is that it is difficult to confidently predict the trajectory of the next few years. So, I am not inclined to make an attempt at the moment. What I can say is that I anticipate the emergence of numerous highly capable models in the world. It feels like we have stumbled upon a new fact of nature or science, or you could call it a fact that we can create. I don’t think it is meant to be taken literally, but it’s more of a conceptual point. Intelligence is just an emergent property of matter, like physical rules or something. People will figure it out. However, there will be different approaches to designing these systems. People will make different choices and come up with new ideas. I am certain that, like any other industry, there will be multiple methods, and different people will prefer different approaches. Some people like iPhones, while others prefer Android phones. I think the same effect will be seen here.
Let’s go back to the first part and focus on cost and speed. All of you at NVIDIA have a bit rate limit to your throughput, and I believe you and most people have already announced how much capacity can be achieved, as it is the maximum capacity they can produce. What needs to happen on the board to truly lower the cost of computation, speed up calculations, and obtain more energy? How do you assist the industry in addressing these issues?
We will certainly make tremendous progress on algorithms. I don’t want to underestimate that. I am very interested in chips and energy. However, if we can make the same quality models twice as efficient, it’s equivalent to doubling our computing power. I think there is a lot of work to be done there. I hope we can really see those results. Aside from that, the entire supply chain is very complex. There is the capacity of logic factories. How much HBM the world can produce. How quickly you can get permits, pour concrete, build data centers, and then have people wire all the lines. And finally, there’s energy, which is a huge bottleneck. But I think the world will step up when it has such great value for humanity. We will strive to make this goal a reality faster. The possibility definitely exists, although I cannot give a specific probability. But I believe, as you said, it is a massive foundational breakthrough. We already have more efficient ways of computing. However, I don’t like to rely on it too heavily or spend too much time thinking about it.
Regarding devices, you mentioned models that can be installed on phones. Obviously, whether it’s LLM or SLM, I believe you are considering this. But will the device itself change? Does it need to be as expensive as an iPhone?
I am very interested in this. I like new forms of computation, and every major technological advancement seems to bring something new. The level of excellence in phones is incredible, so I think the bar is set very high here. Personally, I think the iPhone is the greatest technological product in human history. It really is a fantastic product.
So, what comes next?
I don’t know. It would be great if something surpasses it, but I think the bar is set very high.
I have been collaborating with Johnny Avie, and we have been discussing various ideas, but I am not sure if the new device needs to be more complex or if it actually just needs to be cheaper and simpler. Almost everyone is willing to pay for a phone, so if you can create a cheaper device, I think the barrier to carrying a second device or using a second device is quite high. Therefore, considering that we are all willing, or most of us are willing, to pay for a phone, I don’t think cheaper is the answer to solve the problem.
So, what is the answer then?
Could there be a dedicated chip that can run on a phone and drive AI models of phone size effectively?
There might be, but the phone manufacturers will definitely do that, and it doesn’t require a new device. I think you have to find genuinely different interaction paradigms that can be achieved with the technology. If I knew what it was, I would be happy to start researching it now.
Now, you can use voice in applications. In fact, I have set the buttons on my phone to directly access the voice application of ChatGPT, and I use it with my children, who enjoy talking to it. There are latency issues, but it’s really great.
We will make it better. I think voice is a hint for the next thing. For example, if you can make voice interaction really good, it feels like a different way of using a computer.
However, just like the problems we have encountered so far, like why it doesn’t respond, and it feels like CB, like over and over. It can be really annoying to use, but when it gives you the right answer, it is excellent.
We are working on solving this problem. Right now, it is clunky, slow, and feels unsmooth, unreal, or unnatural. We will make it all better.
What about computer vision? You can choose to wear related devices. You can combine visual or video data with voice data.
Nowadays, AI can understand everything happening around you. You can ask ChatGPT questions like “What am I looking at?” or “What plant is this?” I must admit that this ability is very powerful. It is clearly another hint.
However, whether people choose to wear glasses or use some form of device when needed, it will raise many social and interpersonal issues, and wearing computer devices can become very complicated.
We have seen this with the use of Google Glass. People may encounter difficulties when performing tasks. I forgot some specific situations.
If AI is everywhere, like on people’s phones, what applications can it unlock? Do you have any thoughts on this? What would you like to see?
What I need is a device that is always online, with ultra-low friction, which I can interact with through voice or text, or ideally through other means. It just needs to know what I want and provide a continuous presence to help me through the day. It would have as much background information as possible, like the world’s greatest assistant. It is this presence that makes me better and better.
When you hear people talk about the future of AI, they may imagine it in two different ways, although they may sound similar. But I think there will be significant differences in practice when we design systems. I want something that extends me, like a ghost or another self, or something that truly belongs to me and acts on my behalf, replies to emails, and doesn’t even need to tell me about it. It’s like me but becoming more and more like me. On the other hand, I want an exceptional senior employee. It may know me very well, and I may delegate tasks to it. You can access my emails, and I will tell you the limits. But I think it is a separate entity. Personally, I prefer the way of an independent entity and believe that is where we are heading.
So, in that sense, it’s not you but an always available, always great, super capable executive assistant.
To some extent, it is like an agent that works on your behalf, understands what you want, and predicts what you want, based on my understanding of what you said.
I think there will be agent-like behavior, but there is a distinction between an exceptional employee and an agent.
I want it, and I think what I like about an exceptional employee is that they can challenge me. Sometimes they won’t do what I ask, or sometimes they will say, “I can do that if you want, but here are the potential consequences.” And then this, and then that.
Are you really sure?
I absolutely want that kind of atmosphere, not just giving it a task and having it blindly do it. It can reason and argue. It can reason, and its relationship with me is like the relationship I expect when working with a truly capable person, not a flatterer.
Indeed, if we have tools with reasoning capabilities like Jarvis, it could have an impact on the interfaces of many valuable products we use today. Take Instacart, Uber, and DoorDash, for example. These services are not meant to be pipelines but provide a set of APIs to a multitude of ubiquitous intelligent agents representing the 8 billion people in the world. So, what we need to consider is how to change our understanding of how applications work, the entire experience infrastructure, to adapt to this new world where you interact with the world in an agent-like manner.
I am personally very interested in designing a world that can be used by both humans and AI. I like its interpretability, the smoothness of the handover, and the ability to provide feedback. For example, DoorDash could expose some APIs to my future AI assistant, enabling it to place orders, and so on. I can hold my phone and say, “Okay, AI assistant, please place this order on DoorDash.” I can see the app opening, see things being clicked, and I can say, “Hey, no, not that.” Designing a world that can be used by both humans and AI, I think it’s an interesting concept.
For the same reason, I am more interested in humanoid robots than robots in other forms. This world is designed for humans, and I think we should keep it that way. Shared interfaces are a good idea.
So, you will see patterns like voice and chat replacing applications. You simply tell it what you want, and it knows what sushi you like, what you don’t like, and will do its best to fulfill your needs.
I find it hard to imagine that we will enter a completely different world where you say, “Hey, ChatGPT, get me some sushi.” and it responds, “Okay, do you want to order from this restaurant? What kind, what time, anything in particular?” I think visual user interfaces are great for many things. I find it hard to imagine a world where you never look at a screen and only use voice mode. But I can imagine many things being like that.
Apple tried with Siri. It was said that you could order an Uber with Siri. I don’t think anyone has actually done it because… why would you take the risk of not having it on your phone? As you said, the quality is not good enough. But when the quality is good enough, you actually prefer it because it’s lighter. You don’t have to take out your phone. You don’t have to search your apps and press it. Oh, it logs you out automatically. Oh, wait, log back in. It’s just too painful.
It’s like setting a timer with Siri. I do it every time because it’s really convenient and great. And I don’t need more information. But when it comes to ordering an Uber, I want to see the screen.Several different pricing options. I want to explore how widely applicable this technology is. I even want to know their specific locations on the map because I might choose to walk somewhere. I think by looking at the Uber order screen, I can get more information in less time, whereas if I had to get that information through an audio channel, it would take longer. I like the idea you brought up of observing things happening, and it’s really cool. I think this will bring about some changes, and we will use different interfaces for different tasks. I believe this trend will continue.
Are there any developers building applications and experiences on OpenAI that have impressed you? Do you think this is a very interesting direction, even if it’s just a toy application? But have you pointed out and said that this is really important?
This morning, I came across a new company, which is actually not even considered a company yet. It’s like two people working on a summer project, trying to eventually become an AI tutor. I have always been interested in this field. There are a lot of people doing great things on our platform. But if someone can deliver it in a way that you really enjoy, and they used the phrase I like, which is that it would be a Montessori-level reimagining of how people learn things. But if you can find this new way for people to explore and learn new things themselves, I personally am very excited about it.
Devin, you mentioned a lot of coding-related things earlier, and I think that’s like a really cool vision of the future. I think healthcare should have a pretty big change because of it. But personally, what excites me the most is faster and better scientific discoveries. Although GPT-4 clearly hasn’t had a big impact on that, it might accelerate things by improving the productivity of scientists.
Sam… that would be a victory. The training and building of these models is different from language models. For some people, obviously, there are a lot of similarities. But many models have this kind of architecture from scratch, and they are applied to these specific problem sets, these specific applications, like modeling chemical interactions. You would definitely need some of those.
I think for many of the things we’re discussing, what we lack is the ability to have models that can reason. Once you have that capability, you can connect it to chemical stimuli or anything else.
Yes, that’s an important question I want to discuss today, the concept of model networks. People often talk about agents as if there’s a set of linear invocation functions happening. But one thing that arises in biology is the presence of networked systems with cross interactions, the aggregation of systems… the aggregation of networks produces outputs, rather than one thing calling another thing that calls another thing. Do we see specialized models or network models emerging in this kind of architecture that collectively solve larger problem sets using reasoning? Are there computational models that can do things like chemistry or arithmetic, and other models that can do other things, rather than one purely general model ruling them all?
I’m not sure how much reasoning can be transformed into a broadly generalizable form. I am skeptical of this and it’s more of an intuition and hope that it could be. It would be great if it could. But I’m not sure…
Let’s take protein modeling as an example. We have a lot of training data, protein images, and sequence data, and based on that, we built a predictive model, and we have a process and steps to achieve that goal.
Have you ever imagined a universal AI or a great reasoning model that can figure out how to build submodels and solve problems by acquiring necessary data, and then solving the problem…
There are many ways to achieve this goal. Maybe it would train a text model, or maybe it would just know a big model. It could select the additional training data it needs, ask questions, and then make updates.
I guess the real question is whether all these startups are going to go out of business. Because many startups are working in this pattern of getting specific data and then training new models from scratch with that specific data. And then it just does that one thing. And it works really well on that one thing. It’s better than anything else.
I think you can start to see a version of it.
When you talk about biology and these complex networked systems, I can understand that because I recently got very sick. Now I’m better. But it’s like the body is being defeated one system at a time. And it’s like you really see, well, this is cascading. It made me think of what you were talking about in biology, like you don’t know how much interaction there is between these systems until things start going wrong. It’s kind of interesting.
But I was using ChatGPT at the time trying to figure out what was going on, anyway, I would say, I’m not sure about this. And then I published a paper without even reading the paper, just like in the context. It says, oh, that’s something I’m not sure about, now I think of it this way. So it’s like a small version of what you could say, I don’t know this thing, you could add more information, you don’t need to retrain the model to add that context here.
So these models for predicting protein structures, for example, yes, that’s the foundation. And now there are other molecules on AlphaFold3. Can they do it? Yes, that’s basically the best general model that comes in and gets training data and then figures it out for themselves world?
Maybe you can give an example, can you tell us about Sora? Your video model can generate stunning dynamic images, dynamic videos, and what’s different about the architecture there, whatever you’re willing to share, how does it stand out?
Yes, so first I’ll say a general thing, you obviously need specialized simulators, connectors, data fragments, and so on, but my intuition. Again, I don’t have scientific backing for this, my intuition is that if we can find the core of general reasoning and relate it to new problem domains, like humans are general reasoners, I think it might unlock things faster, I guess. But yeah, you see, it’s not starting from language models. It’s a model designed specifically for video. However, we obviously haven’t entered that world yet.
For example, to build an excellent video model, you start from scratch, and I guess you use different architectures and different data. But in the future of general reasoning systems, AGI, whatever system it is, theoretically it can achieve the goal by understanding how to do this.
Yes, one example is, as far as I know, all the best text models in the world are still autoregressive models. The best image and video models are diffusion models. It’s somewhat strange in a way.
So, there’s a lot of controversy around training data. I think you guys are the most considerate in that sense, and you have now secured licensing agreements with FT and others. We have to be a bit cautious here because you got involved in lawsuits with The New York Times, and I guess you couldn’t reach an agreement with them regarding training data.
How do you view fairness in fair use? We had a heated debate on the podcast. Obviously, your actions show that you’re trying to achieve fairness through licensing agreements. So, what is your personal stance on the rights of artists who create beautiful music, lyrics, books? Do you utilize those rights to create derivative works and then monetize them? What is fair? How can we allow artists to create content and then decide how they want others to handle that content?
I’m just curious about your personal beliefs because I know you’re a thoughtful person in this regard. I know many others in our industry have not thought deeply about the perspectives of content creators. So, I think different types of people would have very different views.
Regarding fair use, I think we have a very reasonable position under current law, but I think AI is so different. But for things like art, we need to think about it in a different way.
But, let’s say you read a bunch of math knowledge online and learn how to do math, and it seems uncontroversial for most people. Then, there’s another group of people who might have a different view. Actually, to keep this answer from getting too long, I won’t go into it in detail.
So, I think there’s a class of people who would say, well, there’s general human knowledge, you can say if you learn that, it’s like, that’s, that’s open domain or something, if you go learn the Pythagorean theorem. That’s one end of the spectrum. I think the other extreme is art, maybe even more specific, I would say it’s like doing, it’s a system that generates art in the style or likeness of another artist, that might be the most extreme. And then there are many, many cases in between in the spectrum.
I think, historically, the focus of the discussion has been on training data, but as the value of training data diminishes, the discussion will increasingly shift to what happens during reasoning. What the system does, accessing information in real-time from the context, or taking similar actions, what happens during reasoning, and how the new economic models are affected, will be more debated.
So, if you say, for example, if you say, create a song for me in the style of Taylor Swift, even if the model has never been trained on any Taylor Swift songs, you still run into the problem that the model may have read articles about Taylor Swift, may know what she stands for. The next question is, even if this model has never been trained on Taylor Swift songs, should we allow it to do that? And if so, how should Taylor Swift get the appropriate compensation?
I think, first, there should be an option in this case, to opt-in or opt-out, and then there should be an economic model. Taking music as an example, there are some interesting things to look at from a historical perspective, there’s sampling and how the economics around sampling work. It’s exactly the same thing, but it’s an interesting starting point.
Sam, let me challenge you on this point.
What’s different about the example you gave? The model learns the structure, rhythm, melody, harmony, relationships that make music successful, and then it builds new music using that trained data. And humans have listened to a lot of music, and their brains are processing and building all the same predictive models and understandings. These are the same discoveries or understandings. What’s the difference here? Why do you say maybe artists should get unique compensation, it’s not a sampling situation, you’re not even outputting, it’s not storing actual original songs in the model.
Yes, learning the structure.
So I wasn’t trying to emphasize that point because I agree, like humans being inspired by other humans. I was saying that if you say, create a Taylor Swift-style song for me
