welcome everybody thank you very much for having us here at the uh rise of AI my name is Thomas nebert and I'm very very proud to be here as a native German we are Nester and I are visiting from beautiful California Silicon Valley so we have a little bit of an outside in kind of a view and hopefully we can get that across a little bit what I would like to do is I like to give a a quick like three minute overview of what the transatlantic AI exchange is and then we go right into the um AI index report now to start um I want to refer back to a conversation that I had or contact with Fabian I was here last year and it was energetic i l the conference and then a few months ago I called Fabian and I said listen I'm a huge fan of the so-called AI index report and uh I read the whole thing back to back last year kind of a geeky thing but it was
very very formative and there was what 300 plus Pages yeah 300 pages I think maybe the only person other than me who read it front to back ever so so uh so I I I like it it was a very helpful Report with all the statistics and I used it quite a bit so I called Fab and I said listen what about the idea with all the hype around AI to actually bring the editorinchief of this particular report to rise of AI and actually present firsthand uh all the bolts and knowledge that's like behind this report so that's why uh I asked Nester kindly to join us and that's why we are here to share it firsthand and and the reason why I'm even prouder is that this morning of all situations here uh when when Fabian actually introduced the conference he referred actually to some of the slides that came from this gentleman and then shortly after I don't know if Tina is in the room but Tina Clover also presented some stuff and
she also referred so there may be other presentation that is referring to the AI index report and I have to say with all the reports and analyst things that you get on a regular basis in your inbox this thing came out on April 15 something like that a month ago at this point a month ago I have never seen a report that has been reposted and shared and referenced within the first 48 Hours ever that's how I believe how how valuable the report is and that's why I'm proud to have Nestor uh with me and by the way if anybody's here not necessarily from Berlin but from you know noran vest Fen which is my home state where I grew up and was born uh we are also next week in dorf at two events talking about and presenting the AI index report so with that in mind let me just quickly talk about what the transatlantic AI exchanges and I would like to invite everybody here to actually check it out and uh participate
uh because it's a win-win for everybody it's an open platform that I started around three three and a half years AG ago and it's an open platform for free everybody can participate it started with a bunch of Zoom sessions during covid to take the knowledge experts on AI from primarily Silicon Valley North America connect them with the experts from Germany that how it started and helped to use the knowledge from over there to Kickstart the implementation of what I was called uh was called K no um AI made in Germany that's what what this report was from years ago so that's how it started 3 years later I'm doing this now full-time I was at Intel as a GM until a year ago but this hobby turned after the magical ChatGPT moment into a full-time position and once an entrepreneur always an entrepreneur so I I'm doing this full-time so what I'm doing is um I have this platform uh and please go there on LinkedIn or on the website um it's primarily for knowledge
exchange and there are many many experts now over 150 who have presented their knowledge to the audience it's all for free I'm not charging for anything and if you have anything interesting to say that is valuable to the to the ecosystem not a sales pitch don't your company whatever is valuable to the ecosystem I am more than happy to pull you in in my new uh generated podcast but also to other session and conferences so that's number one number two is obviously the networking event and you see on the website who has already put their names and Brands behind it um and and it's also about conducting business the VCS want to know more about entrepreneurial startups that are coming out of Silicon Valley people over there investors particularly want to know what's happening in Europe the startups there are corporations who are looking for research projects there are research people who are trying to get the research projects commercialized uh into the hands of corporates or for example uh startups and all of that
is being nurtured through this particular platform uh here are some uh examples about the hybrid events uh I did an annual European AI Summit uh in London in Munich and last year in dorf um I started a series called remarkable woman in AI together with uh their German entrepreneurship at one time at one point and I'm proud to announce that with the com with the combination of a few entities in Silicon Valley and Stanford there's going to be a conference day about remarkable woman in AI in September at the very famous Xerox Park in Pao Alto so if we have females here who would like to participate uh I may do it hybrid so people can actually dial in as a speaker as a moderator as a participant um please let me know and then there are all the other relationships with universities that I'm working very closely with um please don't take a picture of this slide and I tell you in a second why first of all I started the podcast around two
weeks ago and my first speaker was actually believe me or not Fabian because because of the timing of Rise of AI and then there is a a good friend of ours of many of ours here in this room probably now Olaf Co he's Germany's a professor at Berkeley very established and then the third one is actually going to be Nestor so on Friday on my podcast you can listen to whatever he says today but much much more in a podcast session uh about the um the AI index report and last but not least and this is why you cannot take a picture I'm not allowed to actually distribute that PDF but and I have written a report a story in the Venture Capital magazine here in Germany in preparation to the um Innovation conference that takes place next week in dorf by The nestan Innovation bank and they are putting this out tomorrow online uh but I'm not allowed to chare the PDF it's a link and I guess you have to subscribe or whatever
so but I want to give you a taste of what we have done we used some data from the AI index report and I would say at this point the content the main message in that report is very similar to some of the messages that I'm I'm hearing here at the conference about what Germany can do to actually get ahead of the curve and and be a leader in certain segments and it's not necessarily that we have to compete on eye level and create the next large language model um that game may be a little bit over but there's a lot of opportunities in Germany when it comes to our core values that we are known for worldwide without any question biotechnology Nano healthc care automotives stuff like that and that's one of the main messages of that report of the story individually again as I said the transatlantic is something that's free that's literally a hobby turning into a networking to help everybody uh my business model is I have been in Silicon Valley
now believe me or not 33 years and I have spent half of my life building startups and the other half building business units for Deutsche Taylor for Intel and others uh on the corporate site and I'm now seeking interestingly uh startup companies out of Germany who would like to enter the United States uh or I have uh other companies either out of Europe or the United States who would also like to come to the European market so go to market strategy fundraising is what I do on a daily basis with that said uh let's talk a little bit about um the AI index report and maybe two slides that's normally presented by by Nestor is I had the honor to listen to his presentation twice now in Silicon Valley uh the first time was an evening event with a bunch of consulates and whatever and then at Stanford last week great presentation but what I wanted to get across is the global importance of this particular report I mean this gentleman goes around and he
talks to people in Washington DC in on his Senator level or you know on the TVs and you know everything that you hear there so to have him here on stage for all of us um I think is an honor and uh with that I would like to give it over for now to Esther thank you very much Thomas Thanks as well for the kind introduction and I mean we have 20 minutes or so remaining and I'm going to try to distill this report which is 500 pages um in basically a few key takeaways but don't let my distillation stop you I mean the program's almost over today you guys can get home crack open the report and I expect that all of you will read it front to back in one sitting that's how engaging it is of course I have to say that I wrote most of the reports so I mean if I can't even hype myself up what is there to say a little bit about the report before we actually
dive into some of the takeaways what is the AI index so the AI index was put together in 2017 for the first time and it was created by a diverse Committee of really influential AI thought leaders people like Jack Clark who's the one of the co-founders of anthropic that is arguably currently one of open ai's greatest Rivals Eric Bolson who's arguably one of the world's leading economists and thought leaders on the economics of AI and James Mana who's the senior vice president of research at Google and these individuals came together with the understanding that AI was really going to be this transformative technology and it was important to to create a resource that could really tell a compelling story of where this technology is and that's what we do at the index we talk about what AI could do now that it could necessarily do before we talk about economic Trends which countries are leading investment we talk about Trends in responsible AI really I like to think that anything that you might want to
know about AI you could certainly flip through this report and and find it there to actually tell the story of what's going on in AI we partner with a variety of data vendors all of whom are pictured here and we also collect some of our own data for questions that we find to be particularly insightful and particularly interesting so let's actually get to the takeaways here and as I present this Thomas we'll also be interjecting with some questions about how some of these takeaways relate to the German AI landscape and again my one goal here today is to put you guys in a position that when you are at your next dinner party or just kind of chatting at a bar about AI which everybody talks about these days you guys will have some very informative and very statistically driven takes to impress your other guests with so with that let's begin so the first the first take I don't think is surprising to anybody here but I think when you actually see see the
manifestation of this take and see the evidence for how AI is improving you really get a sense of how much investment there has been in this space and how rapidly the landscape is moving forward so one of the things that we did at the AI index is we went to Mid Journey which is arguably one of the world's leading text to image generators and we asked mid journey to generate a hyperrealistic image of Harry Potter over time and we asked this prompt to the original version of mid Journey V1 which was released in February of 2022 all the way up to the most recent iteration now I think this jux Theos makes it pretty clear I'll be honest I'm probably more impressed by the one on the left I'm personally a fan of uh postmodern art and that looks like a Pablo Picasso is grening of Harry Potter but again we asked for a hyperrealistic image and the one on the right you almost have to squint and look very closely to see that that
is not in fact Daniel Rad and there are a lot of different manifestations of the tremendous progress that has been made in AI I've seen some other presentations today at this conference that have already referenced one of the probably most popular charts in the index report which was the fact that on a lot of benchmarks or tests for how well AI systems perform AI systems are already exceeding human capabilities now there is a question to be asked here about why are these systems getting get better what's you know is it just the case that open AI is starting to pay everybody a lot better are they just serving very good kombucha and Silicon Valley like what's actually going on here well a lot of this is scaling and the the scaling laws so quick show a hands here who's familiar with the concept of scaling laws in AI all right so a couple people not too many though so that's good because it will make my information even more relevant so what you're seeing on
this chart is you're looking at the estimated training cost of a selection of AI models from 2017 to 2023 on the right you have some of the more recent popular Foundation models and this model on the left that was launched in 2017 is the original Transformer model this is the model that basically introduced the architecture or the kind of way of building AI systems that now Powers virtually every single large language model so a lot of these modern models like GPT 4 and they use that original architecture and the key difference I mean Beyond some minor tweaks like rhf is they basically scale the model so it was well realized with the launch of gpt3 which was around 2020 that if you simply feed these models more and more data they tend to perform better on a lot of tasks and that's basically what all of these AI companies are doing now they're pretty much betting that if we feed these systems more data they're going to get increasingly better and that's really what you've
seen But as a result of this some of the state-of-the-art models that are now coming out of silin Valley like gp4 we estimate at the index that they cost $80 million to train and Gemini Google's Flagship model $191 million and really there's only a few players in the world that could afford to build such models and this is really the reason why these models are becoming so expensive these developers know or they're betting that if we train these systems on more data they're going to be better to train a system on more data you need more computational resources which you see on the x-axis and as you scale up computational resources you scale up the cost fairly massively now that's a quick question then if you look at this slide and then the previous slide and obviously being here at Rise of AI in Berlin it begs the question where does Europe stand with in this bigger picture of llms in regards to the cost for training but also the resources available uh and I
would say is a game over for Europe on llms on the big ones no I mean I wouldn't say so I still think there's a lot of value in training these kinds of systems and it's not necessarily about I mean what a lot of these companies are trying to do are train models that are state-of-the-art that are better than any other ones but there's still a lot of value in making models more efficient and ensuring that their latency goes down ensuring that they could do more with less and those kinds of research questions in AI don't necessarily require $200 million to do so yeah it might be difficult for certain European players to compete with some of the big Tech Giants unless they have that kind of money to Splash around but there's obviously so much that could be done in the space of AI so a third Trend that personally I'm very excited about and this was I think one of the most important economic trends that we saw in this year's report which
is that there are now a collection of studies across different fields and different Industries which really show in a very compelling way that AI is associated with more productive work and also faster work and work of a higher quality what you're seeing on the slide is you're seeing a cross study comparison of how fast it took a bunch of different groups of people to do various tasks each of these rows represents a different task in a different study and the 100% Baseline is representative of the speed with which it took people to do this task that did not have ai tools now in these tasks there were certain groups of individuals that did have ai tools and they were able to finish the tasks anywhere between 73 to 26% times quicker which is a fairly massive Improvement in terms of the speed with which the task is done done you also see similar results coming out of the world of Consulting in a similar study where there were one group of Consultants that were not
given AI tools and another group of Consultants that were given AI tools like picture here in which there were three groups there was one group that didn't get any AI to do a task and then there were two other groups which got the same AI but one of those two groups was told that the AI was bad that it could do good things but sometimes was prone to hallucination and the other group was told that the AI was good that it was always reliable got the job done very well now if you compare the ability of these three groups to actually do the work both of the groups groups that had access to AI did the work better than the group that didn't have access to AI but of the two groups it was actually the group that was told they were given bad AI that did a lot better than the group that was given good Ai and the researchers basically speculated that what happens is that if you are told that the AI
is very good A lot of these employees just went on autopilot they just kind of stopped checking the output of a lot of these systems and for anybody that's used AI I mean it's good it gets the job done but it's also prone to making a lot of errors and that's an important I think business consideration if your company is integrating these tools it's still very important it's still very essential that you keep humans in the loop with with how you use AI a point about scientific progress that I think is is very valuable I think when you go to these conferences and when you think about AI everybody talks about ChatGPT Claude Gemini but there there's been tremendous progress that has been made in science as a result of AI so many models that have come out that have done tremendous work and I mean I could sit here for five hours and talk about these models I'll talk about one that I think illustrates this point in the best possible way Alpha Man
Ence is a model that was released by Google in the last year and its job was to classify inessence mutations so a inessence mutation is a mutation of a human protein now the important point to know here is that scientists estimate that there are roughly 75 million different menance mutations prior to Alpha Man Ence human annotators were able to successfully annotate 0.1% of that 75 million different mutations Alpha inessence comes along and it has so far been able to classify 89% so we've gone from 0.1 to 89% and that's kind of the power of using some of these large language models and AI systems in the domain of Science and that's something that we talk about at Great Lengths in the index a fifth highlight here and this is personally one of my favorite to share just because I think this is a very pretty chart it also took my data analyst team a very long time to actually figure out how to program this on python so you bet that any presentation I do
I'm going to find a way to squeeze it in this is looking at survey data from McKenzie and I mean all jokes aside I actually think this is probably one of the most important charts in the report especially for a business audience and in this survey McKenzie basically asked individuals managers that were integrating AI into their business function what percentage of them reported realizing cost decreases and what percentage of them reported realizing Revenue increases and the the important thing to kind of focus on here is the last row average across all activities 42% of managers using AI reported realizing cost decreases and 59% reported realizing Revenue increases now I'm just a humble academic I write a report and make a lot of fancy charts I don't know anything about running a business but I do know that if you're decreasing cost and increasing Revenue all else being equal you're putting yourself in a position to outshine your competition and I think what this chart is really showing is you could obviously say speculatively that of
course AI is going to lead to business gains but this chart is showing that from a data perspective it does really seem to be the case that businesses that are using AI are putting themselves in a position to race ahead of their competition and then a final Trend here before we maybe leave some time for audience questions so one of the things that we did in the report this year and last year for the first time is started thinking about how the public actually feels about Ai and what you're seeing on this chart is you're looking at the percentage Point change in opinions on AI in a variety of countries from 2022 to 2023 so ipsos did a survey on AI and it did it before ChatGPT in 2022 and after ChatGPT in 20123 and you can compare how people's answers to different questions have changed and this is looking at how much have these answers changed year over-year so you could see a bunch of different questions do you agree that you have a
good understanding of what AI is do you trust companies that use AI the one that I've highlighted here is asking do you agree with the statement that products and services using AI make you nervous and you'll notice that for every country included in the sample there is substantially more nervousness reported now than there was in 2022 and this is actually a very important point and a very important issue because if you if you look at how a lot of this public opinion data is disaggregated nationally it's really the case that a lot of non-western Nations like Indonesia Thailand and Mexico are actually very bullish about AI they're very excited whereas the Nations that actually score the lowest in terms of pessimism I'm sorry to say are Germany the United States and a lot of other Western Nations and that of course has a lot of very important implications when you think about AI technology and how it's going to develop and unfold and for me just to add to that um when we picked the
slide and the topics out of that massive report we kept in mind that we are here on stage with applied AI so we were trying to pick a few things that are related directly to applied AI um but there's so much more to cover there and one is I put that slide in in a very last minute because you had your favorite slide before the one that was last is one of my favorite ones because if you look at every single country has increased the mistrust of AI you have to ask yourself the question what does that mean how do we solve that because if we don't solve it the numbers can only go up and if it goes up it's going to be counterproductive and then it can all kind of things can happen so we need to ask us uh from the government perspective from an Investor's perspective from a corporate perspective how do we educate the public about AI the pros and the cons it's not all perfect but that is something
for me when I saw the slide that made me nervous that if we don't fix that and address it something bad may happen and the second one is uh which we didn't put actually but I just thought about it um because there was a lot of topics this morning about research and I just reminded myself Germans are really good in research we are really good we have a whole structure in the country with universities and Fraunhofer and dfk and what have you we may not be as good as for example the Americans when it comes to taking the research projects and commercialize or productize that's what the Americans are really good at and there is a whole section about research and professors and where do they end up do they go to the industry maybe you want to add a few comments on the research element of your report yeah I mean I think when you look at the research and development Trends yeah Germany actually does tend to stack up fairly well I think
it's really the case and I think this was made in a presentation earlier at this point is that there seems to be a tremendous gap between the investments in a lot of these countries which is suggestive of the fact that there's really a skill not only in doing the research but then actually spinning spinning it off and turning it into something commercializable and that's something that Americans seem to really excel in yeah