Rise of AI Conference 2024 · Berlin

Rising Against the Giants: State of LLMs/Foundation Models in Europe

Vanessa Cann, Peter Sarlin, Larissa Holzki, Nicolas Flores-Herr

At Rise of AI 2024 in Berlin, a panel moderated by Larissa Holzki (Handelsblatt) with Vanessa Cann (KI Bundesverband), Peter Sarlin (CEO and co-founder, Silo AI) and Nicolas Flores-Herr (Fraunhofer IAIS) discusses whether Europe can build competitive large language models against OpenAI and Microsoft. The panelists agree OpenAI dominates industry and political mindshare with Microsoft's backing, but argue the best model is not necessarily the biggest, and that Europe should focus on smaller, specialized, multilingual and industrial models rather than a general-purpose chatbot. Sarlin presents Silo AI's newly released Viking model for Nordic languages, arguing GPT-4 speaks Finnish or Swedish like a translated American model. Flores-Herr describes Fraunhofer's openGPT-X project and a new EuroHPC-funded initiative training models in 45 languages. Cann warns Europe risks becoming merely a deployment market for US models unless companies rethink their business models, not just bolt AI onto existing products, and shares experience collaborating with a Shanghai AI team. The panel closes on cautious hope, tempered by fear that Microsoft becomes Europe's default answer.

Key takeaways

  • Peter Sarlin argues GPT-4 effectively speaks Finnish and Swedish 'like an American' because its reasoning is built in English and translated rather than native to the language.
  • Vanessa Cann says GDPR has discouraged European companies from using industrial machine data even though it is usually not personal data.
  • Fraunhofer's new EuroHPC-funded project, running on the MareNostrum 5 supercomputer, trains models in 45 languages including Bavarian, ending in May 2025.
  • Silo AI released its Viking model for Nordic languages at the panel, following an earlier Finnish-English-and-code model, without sacrificing English performance.
  • Vanessa Cann warns that Germany's roughly 400-member AI association network and 600 AI startups risk Europe becoming a deployment market for foreign models, with Microsoft as the default answer, rather than a genuine 'deployment champion'.

Detailed notes

Reacting to OpenAI's latest model

Asked about a new, more conversational GPT-4 release, Cann says OpenAI and Microsoft together dominate the conversation with corporates and politicians she meets, and that Europe is eager for its own players even as OpenAI also excels at marketing. Flores-Herr says ChatGPT proved what was possible and made AI a household topic, prompting even his parents to ask what Germany and Europe are doing in response.

Is AI progress accelerating or gradual?

Sarlin argues that scientific progress in AI has been gradual over the past 20 to 40 years, while what is accelerating is the engineering and funding applied to that research; he says current transformer-based, token-prediction models are not a path to AGI, whatever people say about it. Flores-Herr recalls that when Fraunhofer proposed the openGPT-X project to train German foundation models from scratch, the team never anticipated a moment like ChatGPT's arrival, and expects another such leap could still come.

Why smaller, specialized European models

Cann argues the best LLM is not necessarily the biggest one: her own work on industrial models found that large general-purpose models often do not fit industry needs, which instead require multimodal data such as tabular data and knowledge graphs, an area where smaller, focused European models can compete. She also says GDPR has wrongly discouraged use of machine data that is not personal data, and that boards now take AI seriously but often lack the change management to overcome employee resistance. Sarlin says GPT-4 can answer in Finnish or Swedish, but effectively speaks it like an American because its reasoning is built in English and translated, not native to how Finns communicate; Silo AI's newly released Viking model, following an earlier Finnish-English-code model, targets Nordic languages without sacrificing English performance, with a pan-European model in training, because value should stay with companies that control performant regional models rather than flow to closed foreign providers.

Building sovereign models: openGPT-X and beyond

Flores-Herr describes a new EuroHPC-funded project, run on the MareNostrum 5 supercomputer in Barcelona, training both large and small models covering 45 languages including Bavarian, designed so countries like Malta or Portugal can continue training the models on their own data; the project runs until May 2025. He argues Europe cannot build an economy on an API hosted elsewhere and must own the full value chain, comparing a country that only supplies the car radio while the car itself comes from abroad. Sarlin describes Silo AI's model-as-service platform, built on open-source models, giving companies ownership and control of their own models with production tooling, guardrails and retrieval-augmented generation.

AGI, priorities, and the ingredients for competitiveness

Asked whether reaching AGI just needs more compute or new architectures, Flores-Herr says both are needed, and urges Europe to act boldly now on multimodal models, hallucination detection and tool learning, which he expects to show returns within two years. Sarlin frames AI competitiveness around three ingredients: data, compute and talent, where he says Europe is strong, but argues the bigger gap is in scalable software products and distribution that can turn models into real value, warning that good European models will otherwise also benefit better-positioned North American software companies.

Europe's position, deployment risk, and China

Cann outlines a European AI landscape with high openness toward AI, a manufacturing-heavy industry, Germany's AI association network of about 400 companies and 600 AI startups, and a European AI Forum of national partners, but warns most companies are only adding AI to existing processes rather than rethinking their business models, unlike an example she gives of Amazon considering becoming a predictive shipping rather than shopping company. She fears Europe is gambling its future on deployment of foreign models rather than becoming a genuine deployment champion, and voices worry that Microsoft has become the default answer many companies reach for. Asked about her earlier startup's collaboration with a team in Shanghai, she says she still sees potential in EU-China cooperation despite IP and data security risks that required careful safeguards, though her own venture ultimately failed due to internal team commitment issues rather than the China partnership itself.

Quotes

GPT-4 is speaking Swedish and Finnish like an American.

Peter Sarlin

We cannot make an economy that is based on an API that is somewhere in other countries.

Nicolas Flores-Herr

I really do hope that the answer is not Microsoft, and that's my biggest anxiety right now.

Vanessa Cann

Questions answered in this talk

Why does Peter Sarlin say Europe needs its own language models instead of just using GPT-4?

According to Peter Sarlin at Rise of AI 2024, GPT-4's reasoning is built in English and translated into other languages, so it effectively speaks Finnish or Swedish 'like an American' rather than reflecting how those languages and cultures actually communicate, which is why Silo AI built the Viking model for Nordic languages.

What project is Fraunhofer IAIS working on to build European foundation models?

According to Nicolas Flores-Herr at Rise of AI 2024, after Fraunhofer's openGPT-X project trained German foundation models from scratch, a new EuroHPC-funded project on the MareNostrum 5 supercomputer trains large and small models covering 45 languages, including Bavarian, ending in May 2025.

What is Europe's biggest risk in the AI race according to Vanessa Cann?

According to Vanessa Cann at Rise of AI 2024, Europe's biggest risk is gambling on becoming merely a deployment market for foreign models rather than a true deployment champion, since most companies are only adding AI to existing processes instead of rethinking their business models, with Microsoft often becoming the default answer.

Mentioned

foundation models, openGPT-X, Viking model, AGI, open source LLMs, software-defined enterprise, hallucination detection, deployment champion

Keep exploring

More from Vanessa Cann

More from Rise of AI Conference 2024

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 7,161 words

Show the full transcript

yeah good morning everyone and welcome from my side as well the week has been dominated Again by AI news coming from the US so far and I think so long as no one else is talking about llms in Europe let's do this together uh at a um at rise of AI for the next 45 minutes I'm happy to discuss this state of foundation models in Europe and for that please welcome with me three experts in the field the CEO and co-founder of Silo AI Peter zalin the co-founder of nonionic and former managing director of the German AI Association Vanessa Ken and the team lead for generative Ai and conversational AI at Fraunhofer Eis Nicholas Flores here yeah thank you for joining us please take your mics so Vanessa op may I kicked off this week by presenting CAD GPT 4 um a new version of its language model that Converses more human than any other AI model before what do you think how significant is open my eyes lead so far yeah so I actually

I didn't really have time but I had to make time to look at this I looked at the demos uh I think at at 11:00 uh lying in bed and saying okay I have to get up tomorrow for for a keynote I I met Z yesterday I think he's he's missing on the stage also of course a great guy in in Foundation models um and and had to look at these de demos I haven't tried it out myself and I'm very cautious these days with with the entire marketing going on I think oh my eyes also very great when it comes to to marketing uh but I I do think this is a a Hu huge new new stage that we're going into and OPI is uh when I'm talking to a lot of corporates out there when I'm talking to politicians definitely the the number one uh name that that rings a bell and also I'm I'm on stages a lot right now uh and and nearly all the time I think this is

the Once in in in quite a long time that I have met anyone from mic yet but Microsoft is is buzzing around and and selling OPI products so I think the the the dominance of of OPI in the tandem of of Microsoft is really gaining speed when it comes to the industry application of AI uh but at the same time I I think here in Europe there's a a huge uh sensitivity towards being being careful towards us players I think us compan or european companies they really want to have European players here they very open towards this so uh I think we just need the right answers to this and I do still think we can deliver them how do you see it Nicole yes so is it far away from all of us I think open may I have shown that uh what is everything is possible and they made the fastest growing up that's a huge achievement and I have to say before ChatGPT I was talking to my parents and they were

like nodding off and and things like that and now my parents discuss with me oh what is happening with AI in Europe with AI in Germany can't you do something like that yeah so um uh of course not only me but I I I think they they they kind of hold me representative for that so there there is that they show what potential there is in topic of AI but they also show that because I believe IQ is distributed among humans equally so they also show what we can do here in Germany and in Europe yeah so that's definitely possible yeah Peter that was only one in a long series of presentations by open AI over the past one and a half year earlier presentations like the one of the GPT store have squashed many startups can you talk a little about how it feels for you and other Founders watching this presentations are you excited to see what's next or afraid of being left behind yeah that's that's a broad question so I think

first and foremost uh well Silo was was founded uh just a few years after open AI uh we're of course a little bit smaller we're 350 people today um but have a fairly similar profile so majority phds strong researchers in the field but we don't see us in head-to-head competition with with with open Ai and I I I'm not sort of from our perspective I'm not concerned about sort of competition coming from open Ai and and I think ultimately I mean they have a brilliant use case in say gp4 and ChatGPT right a general purpose chat bot but that's just one out of so many use cases out there and most use cases are not sort best solved with a general purpose large language model so ultimately sort of from a startup perspective or sort of any digital native be it a startup scaleup or even a more traditional software company I I ultimately think that you know open AI is not in head-to-head competition and uh you can do uh much more if you

rely on open source and if you specialize because in most use cases you don't need General purpose technology you actually need very performant models for specific use cases that are then trustworthy and reliable for that and open AI is not in head-to-head competition for that it's often said that the or the development of AI is speeding up exponentially and Sam Altman also talked about the new jet GPT function as something magical maybe it is because I hate voice messages so much but I'm not as impressed uh do you think that we still see acceleration in AI development or is it decelerating I I well I think in a sense yes and no um because we've seen we've seen a lot of progress um I think there are many perspectives to progress like if you look at it from a scientific perspective what hased happened hasn't really been sort of this super accelerated development that would have been in sort of huge jump shifts it's actually a gradual appro during the past 20 30 40 years

right um now what is accelerating is all of the engineering work to be able to tap into these sort of scientific developments that have been you know done during the past many years and continue to be done during the coming years we have a bit more funds into R&D to to these activities so there will be sort of a bit faster progress on the scientific side but it's not a jump shift and I mean uh there are sort of obvious reasons why certain people talk about AGI and risks and you know I mean I think the reality is that no one knows what technology is going to take us to AGI MH it there might be sort of from philosophical viewpoints there might be good reasons to sort of spend time on on on that topic and and thinking about what the world could look like but I think the reality is that the sort of machine learning we have today which is basically in terms of Transformer architectures and these GPT models is doing

token predictions like token by token um that's not taking us to AGI so you know I it's still creating a lot of value and it can create a lot of both in general purpose chatbot use cases and also in very specific use cases but but that's a totally sort of different question to AGI yes let's dig a Little Deeper um a little later um Nico how do you see it the question of acceleration um so um when we wrote the the proposal for openg gptx the the first project in in Germany uh uh where we said we train uh Foundation models from scratch we said oh yeah it's going NLP Technologies we called it back then they are just like exhilerating like this and we never ever because you have to kind of say oh in 3 years the market will look like this we never ever expected a ChatGPT moment that we witnessed almost two years ago so and as Peter has already outlined the technology is kind of age that's beneath it so

there might be another one stay tuned yeah I'm very optimistic about that actually Vanessa the development and training of large language models requires a lot of money and compute companies like Microsoft soft can invest billions in their research does it still make sense for small European companies to compete for the best llm yeah because I think the best llm is necessarily the biggest llm so I think it it doesn't correlate as much as it would make sense that size actually makes it a better llm um I I focused uh with the team at nonionic we focus very much at building industrial models and for industrial models um actually L large Lou language models often don't make sense because they're very generalistic and you have to look at actually which data which um also specialized information is important for this industry and then it also becomes very multimodal because language is only one thing you you're looking at tabular data for example knowledge graphs and and all all these kinds of modalities also maybe looking forward

uh into the field of of even sensor technology so the the data points are are just very very very different and this is where where actually smaller models come into play and I think this is where Europe can really uh get gain some speed uh you talked about it earlier also I think Silo is is focusing very much on on kind of building smaller models that are very focused on uh the issues that actually are at play the the problems and use cases that industry is working in and I think this is where we as Europe should concentrate on we have this huge data set of industrial data that we're actually not really putting into into use at the moment and uh I think um gdpr has really screwed up our minds in that sense even though gdpr doesn't actually have anything to say here because it's usually not personal data we're talking about it's machine data so we have to loosen up when it comes to our data strategy and I think this entire

I I would call it also a hype around generative AI it's often only the the Terry on the top but what's missing right now in Europe is more the under Ling problems when it comes to data strategy I think the boards have now um understood that AI is a topic it's it's mostly really a topic of the board now but it doesn't really go hand inand with a good change management and and so we have employees that are actually boycotting each uh AI development uh inside a company bit related to a product or to uh to the processes within the company and this is really uh something we have to change and I think that goes very much handin hand in delivering models that are suitable and and um have the the best results for for industry use cases yeah Peter you guys at Silo tried to differentiates from op Ai and similar companies um with focusing on smaller languages and you release a new model today so tell us what's special about you what

sets you apart from those companies yes um so indeed and and I think um to to to answer your your someone could say a little bit provocative question on whether uh European companies should be building models and and trying to compete in this space I I think I think it's it's it's extremely important for Europe that we are building models and and I think we can compete uh not not necessarily to build a general purpose chatbot um I think that will require a lot of capital and and there might some players in Europe that can also do such things like mstr um so I think Europe is a small tight community and and I think there are good activities um and then depending on what you're trying to achieve I think you know you do it differently the the world I believe in is um a sort of world where eventually uh language models will be embedded into products and they will be focusing on very specific different use cases and I think it's uh

your European language models is sort of it's a values and culture and language perspective which I think sort of is important for society overall that we have capable models that are capturing uh the European sort of region specific values uh culture and and languages um and and and that's not happening currently neither in closed Source models nor open source models um but but then uh I think it's also sort of a say from a a perspective of ensuring that the value that companies in Europe create stays in Europe uh because now if you tap into close Source models it flows to the companies that own the models uh and and ultimately if there are no performant models that cover European languages then it basically implies that that value will flow elsewhere so yes today we're releasing a model called Viking which is the best model open source model for all Nordic languages um without compromising performance in English and that's the perspective we've taken throughout we released a model earlier this year that covered Finnish

and English and programming languages now expanded to the Nordic languages and we're training at the moment a model covering All European languages with the same logic and I think that we can do in Europe uh I mean we have strong supercomputers we have the data we've been spending sort of efforts on ensuring that there are high quality data sets for European languages um but I think The crucial element is do we have the companies that will ultimately benefit from these large language models because they need to be AI is not creating value sort of on its own in itself they need to be embedded into software products and platforms and I think that's where Europe is to a large extent lag lagging but to your model so um as far as I know uh gp4 for example is also capable of speaking Swedish and Finnish so why is it so it it is so important to really train the model in in other languages so gp4 is is is May maybe to be a bit

blunt but gp4 is speaking Swedish and Finnish like an American it is because its capability comes from translation capability the reasoning capability is built into these models because English is so dominating it's built in through English language and then ultimately English language is translated to be German or Finnish or Swedish or other languages so I mean you don't really have a model that is actually representative to the way that a fin would I mean fins are actually sort of quite different in how they communicate so to how a f would communicate SM that that's one part but the other part is really that if you then now you said gp4 that's a close Source model if you're a European company and you want to build Competitive Edge you want to own and control your model uh you wouldn't use gbd4 you would actually look for an open source alternative and then if you look for o look at open source models they don't actually speak Finnish and Swedish so I mean I think we have

a significant Gap and if we want to ensure that European companies create value that stays in Europe then there needs to be performance models that cover European languages yeah okay talking about open source um I think you might get a new competitor with uh NCO and his team Nik tell tell us a little bit more about the urol lingua project that you're kicking off yeah the aurol lingua project we are uh we have um one uh uh huge Grant for compute proposal for in Barcelona at Mara nostrom 5 uh from the Euro HPC so that's European money and we are going to train big and small models as as as Peter and Vanessa have already outlined we need both sizes we need the big ones with have like a more um like profound reasoning capabilities and I can only I mean Peter has already said it all uh uh we need European models we need models that are multilingual that have seen the actual languages in a that coming from the from the member states

and we will make the model such that you can continue to train them like let's say uh uh a country like uh Malta or Portugal suddenly finds texts in huge someone can take them model from oringa and then continue to train it and make it a multi Champion yeah so this is this is something that that that that we will be also we will also be looking into uh making them champions for certain domains if companies would release uh their data so we hope that with this German model we finally can also uh make some uh um enthusiasm create some enthusiasm among company to really bring their data with like what Vanessa has said together with the models so I we need to train models from scratch we cannot make an economy that is based on an API that is somewhere in other countries and then you know you have said it we are going to steamroll uh uh other companies that as Sam Altman has said that right so we have to know what

we have to we have to learn the art to make models we we have to use the models and the and the software and the product that actually employs them we have to know the whole value chain otherwise it's like I'm building a car and you're only putting the radio inside a car that comes from somewhere else that's not a product I mean you can say it's a product but yeah and how long is it going to take to train such a model with 46 languages yeah 45 languages uh and and including Bavarian I don't know if it counts as a separate language but uh it's it it this is just the start I hope really with oinga we will create uh uh U like the spirit and and also also with the Viking models that really can say we in Europe we can really do it and uh uh the project will end in May uh 25 and we are starting already now with the first experiments and um yeah you can wish us

good luck and um yeah so this is and you get some money from Bavaria or why did you choose to yes I I got I got it on my account but don't tell my boss he's sitting here yeah okay Peter you describe yourself as a private research lab too um and working closely with chu University um how do you see the different strength and opportunities between research and Industry in AI so who should focus on what yeah so Silo has been always positioned as a private AI lab when we set up the company a bit more than six years back um uh it was built around my own research group I jump chip from a tenure professorship with a few researchers and a few other co-founders and we were already then sort of with a quite sharp spearhead in terms of sort of AI Talent uh and academic background uh and still a majority of our employees have an academic background so um I I think it it's it is very different though uh to

do sort of academic research and then basically build AI driven products uh and what we've I mean we've been positioned in between I guess we've I I would say we've say operated as a fairly good home for world class AI talent that has been shifting from Academia in order to create real world impact and that's basically what we do I mean we work with some of the industry leading companies including German companies but also other European and North American in in building AI that eventually goes into products um and it's being used by the real world and and where you can really see the impact and and those are the sort of the people that eventually want to join um activities where you can see how AI is having having an impact in in the real world uh but I do think it's a sort of quite quite different activity to do academic research and and and build build products okay but training a large language model doesn't seem to be a business model anymore

isn't it that that's true but I do think that um it is important that we have companies involved in building uh these performant large language models in ensuring that it is more than just a research activity uh I don't think we sort of we can afford running hundreds of small European large language model initiatives as research initiatives but we should ensure that there is sort of also ultimately a commercial activity at the end of that and that's I think our perspective but as you allude to basically what we offer is a a model as a service platform a model serving platform which allows companies to own and control they build on top of Open Source our models are ap2 open source uh we're happy to work with with Fraunhofer uh or with other models out there we're happy to see other models come out in the market and then basically we provide the infrastructure the tools to to run the model in production to optimize for a specific usage use case with usage feedback and

have the guard rails and rag in place so that's basically what we offer and we rely on open source models because we believe that companies should own and control their models and data yes niik there are two camps among AI experts some say if we want to build an um AGI an AI that is smarter than any human we just need more compute and others say no way with current models architectures um that won't work how or which Camp do you belong to um I I think I'm in the middle yeah I I I I'm very optimistic that we will see something like AGI of course it's difficult to talk what is intelligence and what is general intelligence I mean then it gets really philosophical really quickly but I believe we will see astonishing developments um we will need both that's why I say it's also important that we train big models in Europe uh uh because we would need also large amount of computers but we certainly need new ideas I'm I'm it's not

an either or yeah at the end of the day of course the norstar is the human brain which a few wats does amazing things um uh but yeah that that's a that's probably some way to go yeah and how do you see it should Europe Focus now on building AGI definitely that's not it's not a question why should we have as I said in the beginning we have smart people here and it's it now is the time to be bold I was sitting here at a at another platform leand uh nice German word for uh artificial intelligence round by the way and and uh I said we have like one and a half years to really do something to make a large investment and I mean not only companies might also mean politicians Au German German politicians we need to do we need to be bold the time to be bold is now we can't be like thinking about how to be bold in three years now okay if if Europe can pursue two or

three topics which should we focus on AGI is one the first I think AGI is like what will naturally develop if we continue to train model that some at some point it will happen of course there's a big debate about it and I might lose my job what I say but um I think now we need to focus on the next generation of models in AOL lingua we will train large language models uh um which I also believe is still very very important and and they will have their Market to together uh with with Peter and other initiatives but then need to focus on the next uh generation of models multimodal models is is very important we need to work on H early hallucination detection in the models so the models still hallucinate and um there also the topic of tool learning and so forth so these are topics we need to focus on so that and they will bring the return on investment in the next two years it's not like a future thing

so it's worth it let's take another look at the positive side um and see how we are positioned in Europe Vanessa can you outline the European AI landscape for us so which players from industry and Academia should we now should we know um so so I think maybe just to give a r rough meter sketch first um I think where we're standing right now is we have a high openness uh in our industry towards AI I think everyone thank you open AI with chat chat TPT has on understood this is going to be a major topic it's not a hype it's a it's an evolution or even a revolution in such sense I think still and I I travel around Europe quite a lot right now and and and I'm realizing that uh many have not really understood that they have to transform their business models too so they're looking at how can we use AI in current processes and current products they're not thinking about okay what's the next leap going to be uh

just to give a a rough for example of of Amazon because everyone knows it of course Amazon is for example thinking about okay will we in in in future still be a shopping company or will we actually become a shipping company with AI because we know so much about our customers that we don't even have have them shop on the online website we just ship it to them and then have have a a truck come by and and collect what what the customer doesn't want and uh they they couldn't do that now uh because of course AI is not good enough to predict and it would be way too expensive but in a couple of years time um they they are are thinking and developing technology such as this and and their entire business model will change and I do not see any discussions like this going on in Europe and I'm I'm actually quite quite worried because I I think we have this huge chance of kind of becoming a I call it deployment

Champion so I I know there's there's some and certainly in this room and on the stage kind of developing large language models but it it's still a huge Gam kind of which Which models will be the best and are we going to use open source models kind of coming from the US so I think the the huge gamble we kind of currently taking is the entire deployment side because we have this immense uh industry that's very open towards AI uh but I think it's it's the next couple of years will'll be very decisive of if we actually understand that this is a transformative technology and not just um yeah something you you kind of put into the software uh that you're currently using anyway and I think that that's how a lot of Industry players are currently uh using and and and looking at Ai and that's not big enough we have to think bigger and and I don't see that happening enough yet in in Europe uh so that's maybe from a from a

meter level what I'm I'm looking at of course there's there's a lots of um um names out there there's a huge startup scene developing in the field of AI building on large language models uh from from Europe to lots of people are using also M for example or your model to kind of build on uh and and deploy AI for the industry I think the uh especially the manufacturing industry is is one that uh is very strong in Europe so we don't have to look too far uh the German AI association with Daniel is is here today so uh a huge network of 400 AI companies there's around 600 AI startups in in Germany so it's it's really the biggest AI Network we also built a European AI Forum of of Partners in Europe uh I just came from AI Austria yesterday so there there's these networks kind of that that really have intense knowledge that are building on top of large language models and uh are bringing a lot of really problem solving to

the table and I think that's where we strung in because um the US market is still very consumer heavy and we are really looking at what are the real problems of Industry it has to be very high quality otherwise no one in Europe will buy it so the technology that's being developed here is in my in my eyes really high tech technology high quality and um so so really look out for uh players not only from the US there's there's a lot of really good players here in Europe yeah do you want to add for something yeah short note I think Vanessa and Nico had had had good good comments and and on on this on a sort of on the same note um I I I think it's we talk a lot about AI this conference and this panel and and and my company is about AI but but I think to really succeed it's not really about AI um I I mean I believe in a uh I believe in a world that

that sort of implies that software is eating traditional industry software is eating the world what Mark andren said in Wall Street Journal 2011 um and AI is accelerating that and and and that I think implies many things I mean for AI I think we need 3D ingredients we need data compute and talent and I think we're pretty good in Europe we have compute with your HBC resources supercomputers that are performant um we have talent that is world class ranked universities that are ranked research that is ranked well and and we have data because basically everyone has access to data uh to build these Foundation models now you need proprietary proprietary data uh to create value and you actually need distribution to create value and that distribution is software products that's scalable software products and platforms and I think that's really I mean we can talk about AGI and you know be philosophical about whether we're going to reach it or not and so forth but I think for Europe and for Europe to be competitive

we need to make sure that we have software product that can be elevated by Ai and if we don't do that we can build good models and we can build good models and we do help European companies but we also help North American companies and and they're much further down the road when it comes to their software products and actually sitting at scalable business models and unless we sort of tackle that with our traditional industry not only in Germany but also in the nordics um I think we will have a significant problem independent of how good models we build Vanessa with niic you started your own AI company that fell apart due to internal differences would you say it is still worth getting back into the AI race and what would you focus on next well I can't speak about myself personally but I think it's always worth founding a company I learned immensely few few immensely much over the the course of the last year uh so it was definitely worth the Gamble and

I think we we still left our Mark and imprint on on this industry in Europe so um I I would recommend everyone who who's thinking about starting a com company uh think twice it's not good for your work life balance and everyone is saying you look so much calmer and and uh and and happier right now because my work life balance right now is is quite good um but um but but still I think it's worth a Gamble and we have to reach high and I still believe in that and of course the next leap I'm going into will also be a high reach okay you attempted collaboration between Europe and China us companies are trying to cut every connection to to China meanwhile do you still see potential here or has the model failed I do see potential to be honest um because um China is is doing a lot in the field of AI and they're doing really really great work um also of course there's from a b2c perspective we were focus

focusing very much on B2B um there's um there's of course a high risk at it and uh we we spoke very intensely in the team on how to make it possible that we um we we secure IP security and data security but also the infrastructure here in Europe and that was of course a huge gatekeeper um and and had really good solutions for all of these uh these things um so I believe it it is possible um and um we had an amazing team in Shanghai that worked very intensely with us it was a a good collaboration on on that uh side because of course lots of team player or team members came from uh by or Alibaba or also um the the the universities in Beijing and Shanghai that are focusing on developing large language models so experts really in large language te technology and that's that's actually quite difficult to come by in Europe when you're looking at the the talent structure because you have to hire them back uh from the US

as a startup company it enabled us to to grow quite quickly we we grew to 35 uh by kind of uh by uh nine months and uh that's still very difficult to be honest as a startup company in in Germany even if you look at the contracts lots of people are in non-compete closes um also takes um experts to kind of join the team around six months so it it kind of gave us a couple of chances uh even though it it also made a lot of things uh more more complicated and of course it it needed the the commitment of the entire team and uh that sadly wasn't there Peter how do you see it should we cooperate more within Europe should we rely on Partnerships with the US with China or even look for new partners for example in the Middle East I I think I think it it depends on what we we refer to in terms of work together with and and build Partnerships uh I mean if if we refer

to the sort of three components that I referred to earlier uh data talent and compute uh then that's one thing um and then if you if you if you think about sort of uh eventually distribution uh that's or or basically where eventually AI gets deployed which is sort of products uh that's that's another element uh but I think I mean we have some of the world's most talented people in AI uh in the academic community and we are continuously um sort of producing more talent in Europe we should ensure that we're retaining that Talent um and of course attracting more as as well I I do think we sort of have a challenge in uh ensuring that the most talented people stay uh we I I ultimately when it comes to to all of these problems I I think we should ensure that there are moonshot initiatives it's not enough to have sort of hundreds of small AI startups we need to have a few moonshots that are capable of attracting the best and then

eventually capable of of ensuring that we are competitive what about you your favorite partner still is Peter or um Peter is certainly among my favorite Partners but um uh in oinga we are collaborating also with AI Sweden who also collaborates so it's a it's a it's almost it's a it's a European family and from this European re search and also uh I have to underline in in in my team we also develop Solutions so we we also where um uh um the first ones that had like where you could talk to a car like CET GPT and that was one month before CET GPT came out so we can do bring also things into application and we do that in one team that this is also what my team is is is doing and and so I think this is really and this work of course inspired the guy who train the models and and and and I think Talent is very very important and um yeah before we conclude I want to hear a

short statement um from each of you if we meet here again next year what gives you hope that the Gap to us companies hasn't widened want to start what gives us hope M that the Gap has not widen um today I think what gives us hope is that we are doing very well on the three ingredients we're doing very well in terms of having compute in Europe uh having talent in Europe and having access to data um I think uh a key question is uh are we able to actually ensure that those are channeled in the correct way to ensure that there are uh competitive initiatives probably then fairly large initiatives um and then I think uh I'm I'm gradually seeing that there's more of a discussion on the transformation that needs to happen in European companies around actually being I I usually say the way the way same way that car manufacturers started to talk about software defined Vehicles every European company should talk about about the software defined Enterprise or software defined business

because ultimately you need to be that um and and I I think if we are here in one year uh and we can see that companies are gradually understanding that they need to be software defined um then uh I'm happy yeah what El what about you I think I want to add on on kind of your your last point because that's really important to me um I think uh I I do believe we can become a deployment world Champions here in Europe uh we have everything it needs from the data to the strong industry to the interest and enthusiasm I think in the boards now um but I really do hope that the answer is not Microsoft and that's my my biggest anxiety right now because that's yeah that that's the answer that many companies are actually seeing right now and I think the the fight was uh LED actually when we discussed guix for example in Germany so when a lots of companies decided okay we're going Cloud but the Cloud will be Eder

so our data is there and now Microsoft is kind of on every stage and saying hey my co-pilot here and uh all my our products and we have so many really great startups in in Europe um that have a lot of really good answers for industry out there so it needs conferences like these and ecosystems building uh and I'm very en enthusiastic I think how much we as a European Community have also grown together um so that we can see it as a strength being in Europe and not as something that separates us when it comes to the borders and uh so that's that's my enthusiasm um that I think the the openness is there we just have to make it possible that it's a European answer anything to add no they took away all my punch lines so um no really I I think it's important that companies learn that they have to own their AI otherwise their data is going out and and and this is basically is and for this open source

model from this European family I just described will be very important because they are trained on European languages and of course if they use these models we in the future will make better models and new approaches so so and this has to this m if this Machinery works like a push and uh from us the models yeah and the pull from the industry not only like small uh proof of concept project but really big projects we can do that Silo and and mral and then I will sit here and say yes what I have said a year ago works fine so looking forward to discussing the same topic with you next year again thank you again for joining us on stage thank you audience for listening have a wonderful conference day thank you thank you Larissa

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.