 I'm really excited to be here with Roman, it's my first time at Slush, it's also my first time in Helsinki and it's amazing to see all the energy and lights out here, it seems like we're celebrating something, ironically it's also a really special day for OpenAI, it's the one-year anniversary of ChatGPT which it's crazy to think about that ChatGPT has been one year into the product despite it being well over 100 million weekly active and kind of taking the world by storm. I figured we'd start by maybe sharing some of OpenAI's metrics for Europe in particular since we're over here and I think folks maybe are less familiar with that data on the company. In Europe, ChatGPT is at 34 million users, in Finland it's over 600,000 active users which I think is something like 10 or 11% of the population of Finland which is mind-blowing, pretty amazing. Companies all across Europe are using the API, some really big marquee customers for OpenAI in Europe are folks like Spotify, Delivery Heroes, Zalando, Klarna and so on and OpenAI is starting to build a team in Europe. I think you guys open your first office in London and Dublin too, something like 20 or 25 employees here and ambitions to grow the team a ton and so that's really exciting and so maybe to share a little bit about Roman's background, you know Roman is a founder where he started his journey, you're an engineer as well by background. You know the last eight years you've run a developer platform at some of the largest platforms out there folks like Twitter are now X and Stripe which I think is widely regarded as one of the better developer platforms and recently joined OpenAI to lead developer experience so welcome and thank you for taking time with us. Yeah thank you Vince and good afternoon everyone really really excited to be here it's quite amazing to see all the energy here and as you can hear from my accent I'm sure I feel at home here in Europe but really I want to say like I'm super proud like to see like all the progress the European ecosystem has made when I see all this crowd and I'm really getting a kick out of being here you know when I started my journey as a founder in 2008 in Paris there were very very few startups let alone access to capital or even talent in engineering and product and design that had seen like the scale of the big tech companies and now when you look at this room with so many founders here it's pretty remarkable to see the progress we've made in 15 years I can even say I think it's the first time in my career where like Europe is at the forefront of a platform shift right when we went through cloud and we went through mobile Europe was somewhat lagging behind but now with this AI wave it's amazing to see that like it's happening worldwide and especially here in Europe so really really excited to be here today it's pretty awesome we're excited to have you so maybe before we get into product which I know is what our core focus is today I think everyone here wants to know the elephant in the room question the last two weeks have been even three weeks ago three weeks ago was deaf day no I think I think you know what we're talking about I mean the last two weeks have been putting a lot of people on edge on Twitter reading the news I know we're not going to talk about specifics today but I do think it's important for everyone here there's a lot of investors a lot of startups who are customers of yours maybe what's your reflections on the last two weeks and how it's impacted you know potential customers or startups and maybe you at the company yeah well look it's been a crazy time and for the crowd here I've only been at open AI for three months so you can imagine this has been quite the onboarding with deaf day which was extremely exciting and some of the recent events in the past two weeks but really like as you may have seen this morning the announcement like we officially back Sam and Greg are back at the company Miraz back in her role as a CTO and and we're just super super excited to be back to building and continuing to the progress on the mission of like building safe and AGI for for all of humanity so you know personally during that time what has been amazing to see for me was not just you know the resilience and the team spirit at open AI internally but also externally the outpouring of support we've seen from customers from partners from developers all around the world it was really heartwarming to see I've personally received hundreds of messages from from founders who are kind of rooting for us still committed to building with the open AI platform and I'm really proud that during that time we lost zero customers zero partners they were really like you know behind us and and and that's really been like amazing to see so very excited to be back to to building and and you know during that time in fact the focus for us remained on the customers we uh we you know we had on call rotations for the platform we kept uh we went back to where we did best which was shipping uh we shipped chap GPT voice um for all three users in fact that's that's pretty uh that's pretty awesome maybe we can even try it right now what do you think a live demo it's risky I mean why not let me try something all right I'm with Vince on stage here at flush can you greet the audience huh it's pretty cool well to get into product yeah dev day was three weeks ago I think it was your first developer day ever as a company you were a keynote speaker you guys announced a lot of products maybe we can just start for the folks that weren't at dev day or didn't see it what were some of the highlights yeah I mean the first and primary announcement I'd say was the launch of GPT4 turbo so for us there was an update to our most advanced uh uh language model and uh honestly it came with like a lot of um you know feedback that we addressed from all of you customers developers founders and so a few things to highlight within GPT4 turbo obviously like a lower price now it's like three x cheaper for input tokens and two x cheaper for output tokens and so for founders out here building you know it's really like a big deal uh to have these models that are more capable but also more accessible to integrate with products and ship them and bunch of like things that the that we've that we've been seeing on the market as requests like for instance a larger context window 128k and you can now fit an entire book worth of content in in the context json mode better capability to follow instructions for the models higher rate limits and and also the team is now focusing on speed uh next for for these models so tons of uh great updates for people building with these models um and then we also launched new modalities in the api so as you've just heard like many of you i'm sure are using chat gpt uh but um you know what's interesting is now as as developers and builders you can access these modalities from the api whether it's valley 3 whether it's text to speech or vision all of these modalities are now in the api and and we also launched uh our kind of first steps towards like you know uh agents experiences with the gpt's within chat gpt and and the assistance api yeah i mean i was at the dev day and i'd say um the aura of the crowd felt very w w d c s people were really excited a lot of the developers i talked to weren't even talking about the big stuff they were talking about the json little toggle to get that to come out of the the model and so it's impressive to see you guys ship across both the small details and the large features um you know my understanding of the assistance api and gpt's are at least it's kind of your nomenclature for agents uh and certainly we've spent a lot of time with startups some in the room and many of them uh are trying to build with agents but there's a very wide array of how people are thinking about it uh and the spectrum is quite broad so i think it would be helpful for folks here or certainly myself to understand from open ai's perspective or yours how do you think about what is an agent and certainly over the next year what will agent like products look like yeah sure actually before we dive in maybe a quick like show of hands like in the room is actually building with open ai right now wow wow that's that's half the room it's pretty good yeah it's pretty awesome um yeah so back on on on agents it's an it's an interesting term right because it's been used with very um wide ranges of definition for us like when we look back on like agents uh i think at the end of the day it's about like having uh an ai that's more capable uh that's like more personalized more um customizable and ultimately the idea is that you'll be able to ask a computer uh to do some tasks and the computer will be able to execute those tasks on your behalf so even though the definition of agents goes back decades uh you know with uh the robotics and distributed computing and the ai field i think it's all about having these like systems that are more powerful and able to take action on on behalf of of the users so back to what we launched at dev day to be clear we don't quite see gpt's or the assistance api as agents just yet i think for us it's all about like this first step towards this agentic uh software uh that we want to move towards and it's really going back to the way we think at open ai um about like deploying safe uh ai out there it's all about this iterative deployment and we really want to make sure that our technology is constantly in contact with reality constantly in contact with like real use cases and it's really about laying this white blanket of on earth so all builders and founders can really appreciate and integrate that technology into the apps and services i mean i'm not even a develop i'm entirely financial by background and i was back with my family in the us to celebrate thanksgiving and my brother works in marketing at a law firm and we built a little marketing analyst gpt that he shared with other people in the firm to do stuff so it's fun whereas in korea and we built a korean gpt where it everything was text to speech out in the korean language which well seems small but i could program that all myself which is pretty cool yeah and i think for for those of you who have not gonna try to even build a gpt yet what i find quite fascinating is like you don't need to be an engineer to build a gpt right it's all about like making chat gpt work even better for you and uh ultimately you give like access to more knowledge with some files for instance and some actions that it can take and uh yeah you can really customize the whole chat gpt experience for for your own use case whether it's just for you or for friends family and co-workers too and so yeah it's really all these early stages also for gpt's we're very excited about like rolling out the gpt store also in the coming weeks uh for for founders how do you know when folks should be using a gpt versus the assistance api yeah i think that's that's a great question that i see them as very complementary if you will you know as we just said if you're if you're a builder or if you're a creator it's pretty awesome to gonna create something rapidly and deploy it if you're a founder that may just be getting started and you know tinkering with some ideas or trying some hypotheses like it's quite awesome also to test those things rapidly and distribute them immediately to 100 plus million of weekly active users on chat gpt now if you already have a product and let's say you have like very specific workflows or you're uh you know optimizing for a specific persona of of customers well it's amazing to also bring that technology within your product and that's where the assistance api comes in right and for me as a developer i'm very excited about the assistance api too because you you can literally like uh take all of those modalities you know whether it's vision whether it's like text-to-speech and bring that into your own workflows and experiences so i think regardless i think we have in very early stages for for these modalities and and i think what matters is for all founders i would i would i would hear to be able to create the very best customer experiences ultimately whether they're in chat gpt or they want products yeah it's i mean it's fascinating to think about from the investment side or from our conversations with many founders you know this concept of is that a gpt wrapper has come up a lot and opening eyes gonna maybe you know eliminate the need for my business and so you know how do we think about if this is a gpt wrapper or not and where will a long-term product value accrue and so forth has been a question that a lot of folks are talking about i guess you know we have a lot of founders in the room if you were trying to help them discern from open ai's perspective given how big your ambitions are what is a wrapper versus what do you think is a product worth investing in over the long term you know do you have a perspective on that yeah i think ultimately we're all learning right it's all early days for for ai and i think like there's still a ton of room for people to build here um you know the definition of wrapper is somewhat like nebulous right but i think ultimately if you have as a founder like a unique insight or if you have you know a particular workflow you want to go after you should absolutely do so right i think the way i think about this is that there are a couple questions to to keep in mind right the first one is like let's say like if i if i have access now to these ease of use of ai what's my unique differentiator as i'm building a product and the second one is like if a large company is going after something similar how can i compete and so i think in the short term the way i see it is that most of the large companies are go out are going to build after the kind of in the macro sense if you will they're going to go after like use cases that are pretty generic um but if you have let's say like a particular persona in mind um you know we are super excited for instance about customers fine tuning models for health care for medical use cases that's really awesome uh we've seen harvey for instance they uh if you're not familiar with harvey they're like a gen ai um product for law firm and uh that's really awesome too to see that those startups go after this particular workflows like we're certainly not the legal experts here um and you know like the term the term platform is sometimes a little cliche but i think in the case of open ai it's really true like we want to build this platform and we're not going to try to build the like ai salesperson the ai lawyer but conversely we want the very best founders out there to go craft amazing experiences for for these customers right i think that's what really matters and and when you think medium to long term like let's say five to ten years from now i think what's interesting is to think about like what are the the points of friction what's the kind of pain points but also what are the new experiences you'll be able to enable uh for for your own customers and um ultimately it's all about like taking steps towards that like both to set you up for success but also set your customers up for success in this new kind of agentic future yeah but but you know like this is just me rambling but what about you as an investor like where do you see value accrue and and which kind of founders are you excited to back yeah i mean it's easier said than done i think um you know i think it's easy from the investment perspective it's easy to be dismissive of these things that people call rappers because they seem just like abstraction layers on the api and so forth and cute little products start off and they get a lot of momentum and over time they turn into big businesses and so we've tried to think about when is it just a clone of chat gbt versus when is there true software that you're laying hooks in the workflow or you're building for verticals like law or bio or health care we do a lot of investing in and in that in those cases you have domain expertise and you build little integrations that i don't think you guys plan to do anytime soon at least and so from that perspective you know we think things that start off as small could turn into big companies but most of what we're focused on is how do you go find the big big ideas that start as kind of these odd quirky things you know if you and i this was 2012 and you and i were sitting here talking about the app store on the iphone and what is the future of applications it was probably unlikely we were going to pick ride sharing and food delivery or like airbnb and so today when i think about the future of these products it still feels like we're pretty early because we're not we have not seen as many of these kind of quirky new experiences yet which is i think a lot of the excitement that we're hoping for yeah i agree and i think um what i find personally fascinating too is the way we build software is changing right we're going from a place where software used to be quite static built from static data with very much like constrained workflows and we're moving to a world where software is more assistive you can do more for your users you can help them recoup some of their time improve their creativity their productivity also and i think that's where you can build amazing experiences that you could not do before and for me as a developer for instance i'm super excited about function calling as a feature of our models and if you are not familiar with function calling it's this idea that you can give context to the model like gpt4 turbo about the features and the functions are very unique to your product you can describe which features and functions you have which parameters they take and what that means is that when your users are starting to interact with your products in natural language the model can know when and how to call these functions and i think that's pretty powerful in in in how we reimagine software in the future yeah i mean one of the founders we work with josh miller who is building a product called arc which is the next generation of what a browser looks like i saw his talk earlier today and he can be sometimes has a flare for the dramatic on the lines but he said the most underappreciated feature in all of ai right now is function calling i would agree uh and it's where they're seeing the most progress of kind of customer utility so maybe an interesting thing for folks to go look into for sure maybe shifting gears a little bit um you know your core function is developer experience you know i think you spent the last five years really at stripe which i think folks kind of would agree is one of the if not the leading kind of developer api at least in payments and does a remarkable job there in documentation a little details you're now at open ai doing the same thing um maybe for everybody here can you talk a little bit about what it means to build developer experience for open ai and a little bit of how you think about the challenge and what makes it different than what you did at stripe sure yeah um i think what i find very fascinating is like we're just talking earlier like about like how do we like you know build the very best frontier models for founders to use and incorporate in the apps but like really they're only valuable if they're easy to access and easy to use and i think that's where that's what developer experience comes into play and so at stripe when we were building and optimizing for the discerning developer as our persona it was all about like building the most elegant api design abstractions developer tools and really helping developers um you know run their businesses on stripe you can think of open ai in the same way it's like all about like elevating the developer experience removing friction and offering the easiest path for founders and builders to bring ai into their products and what i find fascinating is that i wish we i could talk personally to all founders and all developers building on the platform but we can't there's so many yeah and so um that's where it's amazing to have like a great self-serve experience so you know if you are a hacker and if you have an idea on a saturday you can just sign up get your api key get up and running and that's the that's the beauty of having an amazing dx for for everyone to be empowered to build quite rapidly um and you know at stripe and and both at stripe and open ai i think we we spend a lot of time talking to customers and some and to users to see like what are the pain points what are the points of friction we could remove for them and uh you know at stripe what was fascinating is that i felt users were quite prescriptive about like what they were expecting from us right because they had a business to run they wanted to expand to new geographies new payment methods and so it was never about like you know wondering what the roadmap should be but rather about sequencing it and i think at open ai there's quite a little bit of that but also we're all so early in the ai space that we're also just discovering what the best use cases will be and uh you know sky is really limited in terms of what you can accomplish with ai so there's a lot of back and forth that's really fascinating for us uh with our users and developers and and of course it's going to be quite amazing to look back on this talk in a year or two from now with the ai breakthroughs we're going to have uh until then that's an interesting point i mean are there specific points of feedback that would be interesting for you to surface here for folks or like what are the common things that people are asking you for right now yeah of course i mean like we we see things like of course cost was a big one so uh we're super excited to have reduced the cost of those models especially with gpt4 turbo uh rate limits also was a big point of feedback of course uh we doubled all of these rate limits uh a couple weeks ago and we want to lift them even more in the future and uh we're also having a ton of interesting conversations around like how do you bring new modalities into your products and how do you blend things like vision text to speech all together to enable experiences that we have not seen before so yeah a ton of amazing discussions and trying to bring the best of these feedback into the product it's interesting that you bring up cost uh several times i mean we hear it a lot from startups obviously resource constrained startups care about cost you know to bring up a different topic that is definitely part of the central debate right now with AI open source models versus you know at least gpt4 is closed sourced yeah um you know it's a big discussion i think there's definitely philosophies on both sides you know i think our we've been trying to be grounded with companies of what are the real things that they care about or they raise in conversations for why they use open source versus they might use your api and the vast majority are starting with the api but when we hear about open source usage usage a lot of it is cost driven or they think there is more control and ability to fine tune and flexibility of where they run it uh and then maybe that's more of a thing with gdpr in europe too um and so it's interesting to think about you know cost for a startup you know uh at least if your resource constrained you can't afford ml engineers to run all the dev ops i think i saw an interesting report put out there that as you've shipped the new models you've also taken down cost and supported the older models and so three and a half turbo i think i saw uh someone put an article is almost at price parity with running llama too and so even from a cost perspective long term it seems like the curve is going to steep in and cost will come down pretty quickly and on control we've been debating with folks you know if you're not going to run it yourself as a startup you have to pick a partner to run it for you and so ultimately you're picking between having open ai run it for you or another company run it for you and so there's still this trust element in another party i guess from your vet from your vantage point what do you think are some of these trade-offs at open source or do you have any kind of things you would insert into the debate for a founder yeah look i think if i put my founder hat back on uh when when you're trying to optimize as a founder you're really optimizing for like building um vp and getting to product market fit as fast as you can right and and when you're building with that lens i think what matters is to really like not think about like infrastructure gpu scarcity and so on but rather thinking about your customer your experience that you want to enable and really be like as innovative as you can and the fastest to market and so to me it's all about like leveraging the most frontier model to be the most innovative as you can and i think that plays quite in favor of like using something like open ai as an api directly from the get go that being said we love open source and i think open source is playing a huge role in in the community and we would probably not have uh you know come close to where we are now without open source ourselves so we also want to contribute and give back to that community you know when i look at the progress of the teams like mistral who spoke earlier very nice and team it's pretty amazing to see the progress um for us like you know things like uh clip or whisper v3 like these are open source models that we also put out there and whisper v3 was even first on hugging face a couple weeks ago um but the reason why we also have this like larger and more capable model behind an api is is for us to kind of control a bit more the misuse and the abuse so from the safety lens i think that's why we are kind of like uh using an api as the kind of the abstraction for for people to use but um yeah ultimately we don't know what the future holds but if our founder today i'd optimize to build with the most advanced and most frontier models from an api as that makes sense you know if you think about that in practice building at the frontier using the latest and greatest especially if it's not that much of a cost premium i think makes sense to me are there companies or products that you look at that you guys have worked with where you say that product can't be built with open source today or this is a really you know interesting use case that you can only do on the frontier yeah i think that to me it also goes back to like this concept of like modalities right like when you take a look for instance as a developer i love versel as a company and when you look at what they've done recently with v0.dev for instance so if you're not familiar with v0.dev is this idea like as a developer you can write a short prompt and it would automatically create the code of like a react component or even a full on app and but what the added just last week is this idea that you can actually upload like a figma file or a template even and that would in turn create the code and that's a pretty magical experience that's combining vision capabilities with like the capability of writing code so that's a pretty unique one i think that you could not really build otherwise it's pretty cool also we're here in europe so Spotify is a great use case right Spotify for instance using our voice technology to take the voice of podcasters and turning that in every language and that's pretty remarkable as an experience too that you can build when you have access to these frontier models it's pretty cool i'm just looking at the time i think we're gonna run short here in a second so maybe last question for you a lot has been shipped in the last year we're only one year into chat gpt it seems so early from your vantage point inside of the company and having kind of purview of all the stuff you're working on what are you most excited about next year man a ton of things i'm just like super excited to see how we'll continue to transform software you know with natural language it's pretty remarkable to look back like when the iphone came out steve jobs was highlighting you know the fact that like keyboards used to be this like clunky pieces of hardware on mobile phones and i feel like we're maybe in the same spot with like you know products you know they've been kind of somewhat constrained with buttons and controls and i think we can drastically change the way we build apps and products by by using more like conversational conversational AI in there and i'm sure at upon AI we have a remarkable talent density so i'm sure we'll have like many more advances to share but i'd say more importantly i'm excited to see the breakthroughs that all of the founders here and worldwide are going to build using our technologies our modalities things are still very very early but what i'm very excited about is to see which ones of you in the room founders today are going to be on this stage next year presenting your killer app based on open AI tech so that's what i'm excited about well thank you roman i think a lot to think about i mean certainly this concept of building at the frontier with the latest and greatest biggest models is kind of an interesting thing to think about i'm excited you're here in europe and you guys are hiring more people and building a lot of momentum with startups in europe and so it'll be interesting to see if everyone does with all the technology here and we really appreciate you making the time and the trip out here so thank you of course thank you so much and thank you everyone