The video features an engaging discussion with Mira Marati, the CTO of OpenAI, highlighting the launch of ChatGPT and the transformative impact it has had on the interaction with AI systems. Mira reflects on the unexpected viral success of ChatGPT, which marked a significant inflection point in AI's integration into everyday life, demonstrating how user-friendly interfaces allow individuals to grasp AI's potential better. She delves into how the team at OpenAI was focused on developing GPT-4, but the response to ChatGPT underlined a broader shift towards interacting with AI in innovative ways.

Mira Marati shares her journey from being raised in Albania to becoming a pivotal figure in Silicon Valley, shedding light on her passion for technology and innovation. The video explores Europe's burgeoning potential in AI, emphasizing how companies across the continent are leveraging technologies like ChatGPT for a variety of applications from breast cancer diagnosis to enhanced customer service. Marati also discusses the crucial role of regulation and the responsibility of ensuring AI benefits all while considering privacy and data usage.

Main takeaways from the video:

Please remember to turn on the CC button to view the subtitles.

Key Vocabularies and Common Phrases:

1. generational shift [ˌdʒɛnəˈreɪʃənəl ʃɪft] - (noun) - A significant change that affects an entire generation, often altering the course of history or technology. - Synonyms: (paradigm shift, transformation, changeover)

This, for a lot of people, I think, feels like an absolutely generational shift, one of those very big shifts in technology.

2. inflection point [ɪnˈflɛkʃən pɔɪnt] - (noun) - A moment of significant change or a turning point. - Synonyms: (turning point, milestone, breakthrough)

This is probably the inflection point for how we're interacting with these AI systems.

3. dual use [djuːəl juːs] - (noun / adjective) - Applicable for more than one purpose, especially uses that can be both beneficial and harmful. - Synonyms: (ambivalence, multipurpose, double-edged)

...because we're talking about technology with dual use.

4. empirical study [ɛmˈpɪrɪkəl ˈstʌdi] - (noun) - Research based on observed and measured phenomena, derived from practical experience rather than theory. - Synonyms: (experimental research, observational study, evidence-based inquiry)

Probably the most valuable lesson would be the value of just empirical study.

5. frontier models [ˈfrʌntiːr ˈmɒdəlz] - (noun) - Advanced systems or models at the leading edge of a field, especially in technology or research. - Synonyms: (cutting-edge models, advanced systems, leading-edge designs)

...to regulate the frontier models that have really great capabilities.

6. collective intelligence [kəˈlɛktɪv ɪnˈtɛlɪdʒəns] - (noun) - The shared or group intelligence that emerges from collaboration, collective efforts, and competition. - Synonyms: (group wisdom, communal knowledge, collaborative intellect)

...really pushing our collective intelligence.

7. technical infrastructure [ˈtɛknɪkəl ˈɪnfrəˌstrʌkʧər] - (noun) - The interconnected systems and devices that enable the functioning of technology. - Synonyms: (technology backbone, IT framework, network architecture)

...ensure that we have the public and technical infrastructure for these advancements.

8. decentralized talent pool [diːˈsɛntrəlaɪzd ˈtælənt puːl] - (noun phrase) - A workforce spread across different geographical areas, contributing through a distributed network. - Synonyms: (distributed talent, scattered workforce, integrated expertise)

...funding is reaching more diverse countries, and there's a decentralized talent pool.

9. iteratively [ˈɪtərəˌtɪvli] - (adverb) - Repeatedly performing a sequence of operations until a specific result is achieved. - Synonyms: (repeatedly, gradually, progressively)

...to actually deploy the systems iteratively and do so in a way that...

10. capital intensive [ˈkæpɪtl ɪnˈtɛnsɪv] - (adjective) - Requiring significant amounts of money or investment to operate or develop. - Synonyms: (investment-heavy, expensive, fund-demanding)

OpenAI might end up being one of the most capital intensive startups in Silicon Valley history.

Unveiling Europe's Tech Pioneers through AI's Transformative Power

Every year, to launch the State of European Tech report, we travel across the continent to showcase incredible stories of innovation, in the words of the founders who created them. We also like to showcase incredible Europeans who are building the future that we like to see. If you're wondering where I am right now for this year's launch, let me see if I can give you just a little hint.

Chat GPT, what is the State of European Tech report? The Atomico State of European Tech report reveals several key insights. There are 166,000 tech startups, indicating a robust ecosystem, funding is reaching more diverse countries, and there's a decentralized talent pool.

Ladies and gentlemen, please welcome OpenAI CTO Mira Merati. Mira, thank you so much for being here with us this year, for this year's launch. Thank you for having me. I want to start, obviously, for anyone who's been living under a rock for the past twelve months, you're the CTO of OpenAI and you lead the team that built ChatGPT. This, for a lot of people, I think, feels like an absolutely generational shift, one of those very big shifts in technology.

I wonder if you could just take me back. Were you surprised at all by the reaction to the launch of ChatGPT? Well, when we launched ChatGPT, we were actually all very much focused on GPT-4. That was the most advanced model that we had at the time, and the entire team was focused on developing it, figuring out how we would make it safe and actually deploy it. And we were just discovering it, and we were so excited about GPT-4, so we weren't really thinking so much about ChatGPT at the time.

We were looking for a way to gather data from researchers and get feedback and figure out how to make GPT-4 more reliable, safer, more aligned. So we thought it would be great to release ChaGPT as a low-key research preview. And of course, it was anything but low key. And we found out just the next morning that we had 100,000 people that had tried GBT.

And so at that point we realized, okay, this thing went viral and we ran out of GPU capacity that we had planned for, so we had to rethink what we were doing. And alongside us, everyone else in the industry obviously realized that something totally different was happening and something had shifted. And I think, yeah, at that moment we realized, okay, this is probably the inflection point for how we're interacting with these AI systems, because more than anything, it showed us that people just loved to interact with these AI systems.

When we put it in front of people in this different form factor, because the capabilities were kind of already there and we had already showed them through the API. But this dialogue system, this interface, that's very natural and easy to use, you know, how the two of us are talking right now made it very easy for people to grasp the capabilities and see what AI is about. And, I mean, it's been quite incredible, I think.

Congratulations on building what I believe is now the fastest-growing consumer Internet application of all time. For the people who may not be familiar with your story, you obviously live and work here in the US from these beautiful offices here in San Francisco, but, of course, you're born and raised in Albania, in Europe. Can you tell me a little bit about your upbringing and how that shaped your interest in technology and perhaps specifically designing and building things?

Yeah. So I grew up in Albania in the nineties, and obviously there was kind of a messy time in the Balkans. But, you know, when reality seems messy, there is also, at least in my case, there was also this inner sense of possibility and having this inner sense of freedom. And there was, for me, the way that showed up was very much in education and pursuing knowledge.

And I was very interested in math and eventually in physics as well. But, yeah, my interest in math and physics and science in general started from a very early age, and then eventually I went on to pursue engineering, studied engineering and math later in college, and went to work in aerospace and automotive. So I really just loved the idea of making things and building things and not stopping at the theory of things, but really figuring out how to bring the science into the real world and how to bring that knowledge into the real world.

I wonder if I could read a little quote, something that you said. Actually, you said, we're at an inflection point when we're redefining how we interact with digital information, to the point where we're now collaborating with AI and actually many different types of AI with different competencies. What most excites you about the potential of generative AI?

Often people will ask me about the capabilities of the models, what are the most exciting capabilities? And while these models are quite capable, and we've seen it in many ways, people use ChatGPT in their work. They use it for fun, for coding, for creative things, and now especially that we've integrated all to tools like Dali and so on, we see people use it for a lot of things.

But I think perhaps the most special thing about ChatGPT is that it actually is teaching people how to collaborate with AI. It's really teaching us how to use it. And I think that's quite powerful because it's really bringing AI into the public consciousness, and we're seeing the effect that that is having more broadly, you know, in bringing it top of mind to governments and civil society and everyone really in society that needs to be thinking about this and figuring out, okay, how do we ensure that we have the public and technical infrastructure for these advancements to actually be good and beneficial for us?

But I think the actual concept of AI systems as collaborators or as tools that help us and enhance us is not particularly new, because we as humans have always had this concept of having human helpers and enhancing our capabilities, our innate capabilities. And so I think AI system just happened to be the next frontier of that, you know, enhancing our curiosity and creativity, intelligence, and really pushing our collective intelligence. And what we use that for is kind of, you know, up to us. It's really unbounded.

I am particularly excited for how we apply this to education, because you can imagine that, you know, if we have the ability to sort of learn in ways that are custom to us, it can really push the bounds of our imagination and intelligence and what we can do, the impact that we can have in the real world. I'm also very excited for how this is going to be used in medicine now.

We're just starting to see very early signs of how it's going to be used in medicine through perhaps diagnostics. But I think really there is immense potential here, and I'm really looking forward to how it's going to advance science and medicine, thinking about things like new materials and education as well.

Of course. I want to pick up on a couple of things that you said there. You mentioned medicine. One of the companies that we actually interviewed for this is a company in London called Chiron Medical, and their AI is for breast cancer diagnosis. It's been proven to detect 13% more cancers.

And Sarah, who's their chief strategy officer, said something very interesting, which is, she said, the promise of AI is not tomorrow, it's today. Our product, Mia, can save women's lives today. I wonder if you have seen other opportunities like Chiron for Europe to produce meaningful companies in AI. We work with a ton of companies that are using our models through the API.

I think medicine is really very much in the early days, but we're seeing a ton of other applications, like, for example, Spotify. They use the voice technology to make podcasts available in different languages, but still with the same voice of the narrator. So you keep kind of the storytelling that was you guys. And, you know, I think that's wonderful.

And there's another company, Klarna, that uses our API through a bunch of different things, like, you know, shopping recommendations or streamlining customer service to their 120 million customers. And then there is booking.com, which of course, uses AI to help you plan your trip end to end. So sort of like an assistant for planning your trips. There are a ton of applications from companies in Europe building on top of the technology, especially building the product experience on top of the technology.

Amazing. And you mentioned earlier actually as well, of course, another sort of key stakeholder, I suppose, in this big shift that we're seeing -- governments, this has been quite an interesting year. So, I mean, the EU was obviously the first to sort of move into the regulation space. But actually, we're at the end of November, the US obviously really has come out quite strong. There was Biden's sort of executive order, and of course, the UK sitting in the middle is sort of, sort of not quite sure yet what they're going to do.

I mean, what is your view on regulation? Like, what should and shouldn't be regulated in your mind? Because obviously you're sitting right in the middle of all of these things and also having to deal with these stakeholders directly. Yeah, I think this is really a turning point for regulation. A lot of governments have now realized the importance of AI and there's been a lot of activity.

You mentioned the UK Safety summit, which we participated in, and the UN has also kicked off an initiative to work on AI governance. So there is a lot of activity around AI governance and AI regulation. Now. I think regulation is definitely very important, and we need to figure out how and what to regulate in a way that there is still room for startups and innovators to actually move fast and work on building new technologies, advancing the new technologies, and building products.

And when it comes to frontier models, models of a certain capability, but also that come with a lot of downsides, we should be absolutely willing to regulate them. And this is where we focus the conversations both with the US government and I, UK, and EU governments, to figure out how do we collectively regulate the frontier models that have really great capabilities, but with those great capabilities, of course, come downsides, because we're talking about technology with dual use.

Yeah, and I think one of the things that is a big theme every year in the report is talent. And I remember seeing an interview with you where you said that at the point when you joined OpenAI and you decided that you wanted to work on artificial intelligence, there were really only two companies that you would have joined. There was OpenAI or DeepMind, which is obviously in London, where I live.

Interestingly, here we are in San Francisco, and I feel like there's posters, everything's AI on billboards and signs, and it's the capital of AI. But of course, you've just announced two new offices, London and Dublin. What's your take on the state of AI talent in Europe? What kind of work will be done in those offices?

We definitely have some part of our technical team in the UK office, and we're expanding as well, because our mission is to figure out how to build these advanced AI systems that benefit all, and that includes making sure that we can make our products available in a lot of different countries, gather feedback, but also really engage with the local communities and governments and understand the impact, both the impact that the technology can have, but also how we bring that feedback back into the development of the technology and figure out how to guide the deployment of these technologies in different countries.

And obviously, Europe is a very important place for us to operate and figure out how to help local businesses and how to provide more access through different countries. So a few things that we're doing are thinking about making the technology available in all languages and at the same quality as English, thinking about how to make it more accessible in different countries. So one of the most recent announcements was GPT-4 Turbo, which is much cheaper than our default model.

And so that's something that we've really focused on to figure out how to make these models more accessible and cheaper. And obviously, the third part that's very important is, you know, user privacy and more broadly, data privacy, and making sure that people are in control of their data and they're in control in general of their AI, how their AI is being used. So having a presence in the European continent, but also more broadly in other spaces in the world is very important and will probably expand in the coming year.

So we can figure out, you know, start figuring out how to deploy these technologies more broadly and do so in a way that takes into account and consideration input from local businesses and governments. So maybe we're having this chat in a year's time in our office in London.

You know, I mean, I want to pick up on something you just said there, which about ChatGPT Turbo and cost. And I think it's quite an interesting sort of area, obviously, opening ICO, Sam Altman has sort of suggested one of my favorite lines, that OpenAI might end up being one of the most capital-intensive startups in Silicon Valley history. But you've addressed this, as you've just said, and just now with ChatGPT Turbo. But how did you manage that and why is it so important?

We're constantly figuring out how to optimize these systems, how to make them easy to use, cheaper, faster. So we're, you know, after we developed AI model, and obviously the initial focus is on the capabilities and the reliability and safety alignment of the model. But once we start working on deployments and making it available in a product or integrating it, it's very important to actually make it cheap, easy to use fast, both for the direct users, but also for integration into other products and making it available on the platform and for businesses to use. Because who doesn't want an AI system that's reliable and fast and cheap to integrate into other products?

And how important do you think that step is in the evolution of OpenAI? If you look to the future, I think it's quite critical because on one end, we are developing the most advanced AI systems, but to figure out how to develop them in a way that's aligned with human intentions and deploy them in a way that's safe and it actually benefits people. You can do it in a vacuum, you have to do it in collaboration with people.

And in order to achieve that, you actually have to make the technology accessible. And, you know, we haven't figured out a better answer so far than to actually deploy the systems iteratively and do so in a way that, you know, people can actually interact with them in a way that's, you know, cost-effective and fast. And from an interface perspective, it's also friendly and easy to use.

Yeah, absolutely. You know, it's obviously been an incredible year for you, for the company, but also for the world. We're on the cusp of this huge, big transition. As I sort of said at the opening, what's the most valuable lesson that you've learned this year? And actually, if I can add a bit to that, what's the thing that you're most excited about as well for the year ahead?

Probably the most valuable lesson would be the value of just empirical study and seeing what happens when technology comes in contact with reality, with the actual friction of the real world. And, you know, the fact that we're not so great at predicting the impact of things in the real world.

And so we should try to make this evolution as continuous as possible and bring people along, because the progress of technology in this technology is happening really quickly. And so we want, in order for people to benefit from it, we need to bring them along, and we need to make this a very continuous adaptation.

We couldn't have possibly predicted the success of Chajub and how much people loved it because we were looking at the technology more advanced ourselves. And so it was very important to actually get that contact with reality early on and when stakes were still low with GPT-3.5 before advancing to GPT-4, and see how it affected the people's workflows and, you know, what risks it presented and how we could have built in mitigations before we moved on to more advanced models like GPT-4.

So I think really learning from this contact with reality and figuring out a way to collectively bring the input from society into building our models has been a very valuable lesson. In terms of what I'm most excited about, I would say definitely the progress that we're making with these technologies brings me a lot of optimism about the possibility of dealing with great challenges when it comes to education and climate change or medicine that we talked about earlier, but also the possibility for really changing the way that we're interacting with these technologies and making it as easy as possible.

Easy and inspiring. We saw this shift from GPT-3.5 on the API to GPT-3.5 through ChatGPT interface, and it's a massive shift. And some people have likened it to analogies between telegraph and the telephone, you know, 20 years apart, but telephone easy to use, and it really changed productivity.

And so I am really excited about the potential of changing, continuing to change the interface with digital information and making it easier to work with and truly collaborate with this AI system. And I'm very excited about building the first state of European tech bot that will answer anyone's question anytime when they email me.

Speaking of inspiring, I spoke to a few friends in the industry when I said I was coming here to meet you, and I asked them, is there one question that I should ask you? And this was my favorite. If you could solve one global problem with AI, what would it be and why? I think it would be giving everyone the most capable and free AI tutor and really pushing everyone's ability to learn. That's a great answer.

2023 will probably go down as the year of AI, and I think it's a phenomenally powerful technology. I don't think there's anybody in our industry who doesn't acknowledge this very fundamental shift.

In your mind, what does the future look like? Well, you know, if you look at all the previous revolutions we've had, based on that, we'd have predicted that AI would just automate a lot of basic tasks. But in fact, what we're seeing is kind of like a digital renaissance that's really pushing bounds of creativity and curiosity and self-expression and intelligence.

And if you really push that, you know, what we're seeing right now is that natural language, just how we talk to one another, is really becoming like a new coding language, let's say. And so with that, people have this superpower to really build anything. And a lot of tools are easily accessible today. So with the tools that people have at their hands, they can just do incredible things.

And I think we will see a lot of interesting products, a lot of interesting companies being built on top of these technologies in some of the domains that we just discussed. We can't obviously right now predict everything, but I think we will just see really amazing, amazing companies and products being built on technologies that will become more and more capable. They'll have more, you know, creativity and more reasoning capabilities. They will really expand our way of communicating and interacting in every way.

So we don't know exactly what the future is going to look like, but we know for sure it's not going to be boring. I can't think of a better note to end on. Mira Marati, thank you for being with us for the launch of the 9th annual State of European Tech reward. Thank you.

Technology, Innovation, Leadership, Artificial Intelligence, OpenAI, ChatGPT