Hello, Lutz. Good morning. How’s life? It’s pretty cool.

Those who see the video, I’m pretty tense. So we have summer finally in Germany. How’s California doing? The blue sky is today.

It is still very chilly, which is always a good indication that summer is coming. But otherwise, it’s good. So I can assume you made yourself a coffee, a warm coffee. So how’s your coffee this morning?

You saw that I did some recording out of Vancouver. So I actually drove those 16 hours from San Francisco up to Vancouver. You come through like three amazing cities which have an extremely good local roaster community. So I bought so many bags of local good roasted coffee.

Light roasts are the right roasts here. Then I actually had to buy a fridge to actually pack up 360 shots of coffee into 17.1 gram entities. And now I can taste all different kinds of coffees and talk to you. Isn’t that amazing?

That’s pretty amazing. I think you reinvented Nespresso or some similar company, obviously. That’s, yeah. Thank you.

Other people would say, I have a Eura machine. Why? Why do you do all this? But you get a good taste.

I can tell you, like, once you start bringing coffee with me, you will never, ever go to one of those traditional coffee shops again. Yeah, should really come by California and taste it. Okay, so Vancouver is a good hint here. Also California, and I’m based out of Europe.

We have new format. Vancouver is not California, right? Vancouver is Canada. Yeah, I know, I know.

No, talking about what we want to do now, which is talking about Starbucks. So, we have a lot of startups in those areas. So, Canada, the US, Europe, because a lot has happened in the last 30 days plus. And we felt like, hey, we should maybe just talk about companies and be less technical.

Maybe we do another technical podcast next, but now let’s talk about some companies. And we have a list of companies that we want to deep dive still briefly so that for you guys out there, you get some good info, but not too much. You can read the rest. And then we found two companies that we think we should watch out for.

But we’re not going to go into that. But they are still early. Yes, totally. And we’re looking at three nice deep dives, Cujillo Runaway and Hippocratic.

Do you want to kick it off with Cujillo? What are they doing? Yeah, Cujillo. So, the company is from Toronto, not Vancouver.

I think that’s interesting. They were founded 2019. So, you will see now through our podcast that all of these companies are actually older than when the frenzy started, let’s say this year or last year. And the interesting thing there is the founders, if you look at the background, one of the founders, Aydan Gomez, actually came from the Google team, the Google Brain team, who wrote the paper, Attention is All You Need.

And when you read about the background, he saw the application of transformers. We spoke with Retu from Ultimate AI about it. And he felt like he should build a company that helps other companies, not just Google, to apply this in practice. And they’ve been building for quite some time and now raised a pretty sizable round, 270 million funding round at a 2 billion valuation.

We don’t know the revenues. I’m not so sure how much there is. I don’t think the normal multiples that we see on the stock market, I think it’s 6.5x annualized revenue, and that’s forward-looking, can be applied to this round. And it still really shows that there is something going on.

But maybe, Lutz, you can tell us about the product. Because it sounds from… Yeah. Everything you read, it’s another large language model. But that wasn’t actually the idea of Cohere, because the founder wants to help people to really use this in practice.

So what are they doing there? So, actually, I didn’t try out Cohere so far. But I think they are very much on the idea of using large language models for writing. So they fall into the same range as just for HyperWrite and the others.

And I think the discussion, which we had, was, who will have the ability to train and run large language models? And we discussed this before, right? How likely is it that you build out a business case on one side? And how likely is it that you have a good use case and a good UX for this one, right?

So for me, a typical company, which I started in Canada as well, because they had the knowledge, was Element AI. Yeah. Long time ago, right? And Element AI essentially said, we know AI and we are going to do beautiful things.

And we got founded in 2016. And one of the main brains on AI was part of the journey. However, what they were missing is the applicability. And what I think we have here in Cohere is very clearly focused on text generation and very clearly focused on building out the own model.

And I think what’s interesting there is, when it comes to the application, so we know that Jasper AI was one of the early adopters of OpenAI, GPT-3, I think, at first. And they got bit into trouble when this started costing money and they couldn’t really monetize. So they seem to be moving away there. But Cohere wants to focus on more of the explainability part.

And as we discussed before, controlling the model, giving guardrails. We come later to another company that is doing something similar. But it seems to be such a big problem. And they are trying to control these models.

That at Jasper AI, they wouldn’t build it on their own. Or they even mentioned Salesforce Ventures. Salesforce also has some AI in place, but they also seem not to be able to build this on their own. So it really sounds to me like pickaxes and shovels that Cohere is providing here.

Yes. And that comes down to the discussion which we had before. You need to have some form to run those models. This is just a statistical next best word.

How do you manage that next best word? And it’s actually like already, like I tried out both Jasper and GPT. And ChatGPT is actually more flexible because I can, like the whole discussion on priming. I can prime ChatGPT, not priming, prompting.

I can prompt ChatGPT way better than I can do with Jasper. And therefore my interface is easier. I can actually ask ChatGPT to guide me through my article questions easier. By becoming an expert prompt.

All of this shows that there is a huge unlifted value. And I think Cohere is very much focused on that. And then there was another point and I’d like to hear your thoughts on that. I think it’s the COO.

They call him Martin Kohn. He actually was the CFO of YouTube. So quite a nice hire in 23. And he says that in the future, they see search and retrieval like the next core area for their growth.

And they want to give models or chatbots the ability to expand their knowledge base and search the web for information that is relevant to query. Sounds a bit like what Rachel described with Ultimate GPT. That is also kind of a search functionality. And he says, and he mentioned something like today’s chatbots don’t have access to the world.

They don’t know about what happened 10 minutes ago. So that was the attention problem. Does that make sense from what we’ve discussed so far? Or is it just marketing talk from your opinion?

I actually, so there are two things to it. One is the connectivity to access other areas. And that is the race which everybody tries. And I would actually not put this onto the actual large language model level.

Right? This is like Zapier or if then, then, that. You are trying to have an internet connection. You are trying to have an interface where you have all the connectivities out there and all the plugins.

And those will come. I think what we see here more is how do you frame and fine tune a model and give guardrails so that this stays within the right area of a model. Yeah. And it’s actually interesting because I believe you see more and more companies in that space who pre-trained.

There are even platforms for pre-trained stabilization. And stable diffusion models. And here now you have the ability to pre-train on to fine tune the weights of a text model. The question is, will everybody do it by themselves?

And you just need a distribution platform. Then that investment into Cohere. Again, this is no investment advice. But the investment in Cohere will not pay off because it becomes a distribution plate.

Everybody can train and create guardrails. And all what you want, you say, I want guardrails like X. And you need a platform to give you X. Or I want a pre-trained language model like Y.

And you get that pre-trained model. If it becomes more ingrained in the actual structure so that you cannot so easily change this around, it becomes harder. And the second part he mentioned was you don’t just, you know, it’s not a good idea. It’s not a good idea.

You don’t just give more guardrails, control the model. But you would actually be able to ask the model where this information comes from, which definitely right now is not the case. And if I understand the whole application correctly, this is also easy in an enterprise context where I have a direct access to the data. And I can really see, you know, what’s out there, what’s kind of the ground truth.

Does that make sense to you? Yes. It comes down to the setup and identification. Right.

You see this. This is what Bing at the moment in their collaboration with OpenAI start doing. That they say, okay, I show you where that information came from. In reality, this is, these are two models, right?

This is a search model. And then there is this translation model. It’s only that we perceive it as one. Meaning search is not a simple problem.

Bing and Google have search models. If you want to go into the world and become a Google search model, you can do that. You can go into the world and become your own search model. You saw this with you.com.

It becomes actually pretty hard to get the relevant searches up and running. Yeah. So let’s say it makes sense, but fingers crossed for them to make this happen. We also heard that the round wasn’t easy.

There were some sources saying they tried to raise at 6 billion. Now it’s 2.1 or 2.2. The lead is Inovia Capital. Which is already insanely high.

Yeah, it’s still very high, right? But obviously you know that. You always try to capitalize on the frenzy. And Inovia Capital is a well-known growth fund.

And then they had already a couple of exits. 57, I think. So well-known brand. Our friends from Deutsche Telekom Capital Partners also invested.

They’re very strong investors with index ventures. So fingers crossed. Very strong capital base. Raised a lot of money.

And we just wanted to briefly mention also that, and we touched upon it last time, Anthropic obviously also raised a lot of capital, which is very strong. And we have a similar idea with guardrails. And you can maybe touch on that. Spark and Google led the round.

So we have another corporate investor here. Also Spark as a tier one fund, which is pretty good. But the Anthropic idea we touched upon in our Bias podcast quite a lot, I think. What we see is that we have open source models.

And we said open source model community has actually evolved way stronger than all the other closed models. So instead of open AI, I think we should call it now closed AI. Because they don’t allow access to it. And that’s the reason why Google actually had this leaked paper where they said, actually, it is not Google.

It is not open AI. It will be those open source models. The question is, how do you work with those models? There is a distribution.

CWIT AI could be, for example, as a platform for stable diffusion models to fine-tune. For whatever taste you would like to have. You will see similar things happening in the area of a large language model. This is where Cuhir and Anthropic come in and try to say, okay, maybe we can fine-tune, pre-tune, set this up, make this easier.

And then you will obviously have the guardrails. So that the model is easily doing what you want to do. And then it comes down to, is it MLOps or is it more of those models? And it seems to cost a lot of money.

And some corporates happily investing into that. And we should maybe sum this up with the next discussion we have. Why those corporates might be in these models. So these were more infrastructure plays.

Yeah, monetizing on all different kinds of applications. And then we saw two funding rounds happening in the past. One right today, actually. One is Runway, where Google had a 100 million Series D round at 1.5, maybe, billion valuation.

That’s a platform to edit and create videos. And there is another round that happened, which a company called Synthesia. And they’re from London. 90 million round that by Excel and NVIDIA.

So NVIDIA, the chip producer of the amazing graphic cards I used as a young kid. They were always too expensive. And Synthesia is basically doing marketing videos, teaching videos, learning videos, where you have a character, an avatar that gets animated by AI. And you can easily create those videos with the help of the platform.

Now, let’s start with Runway. I think the background is interesting. They are from 2018. There are various podcasts with the founders.

I think I can highly suggest to listen to them because they started by creating. They wanted to change the way creatives would work with technology. So they invest a lot of time, took some time actually for the platform to take off. And then when it finally got traction, Felicis Ventures, they invested 35 million.

And now, sorry, yeah, that was the Series B. And then now we had another round of 50 million happening. And now this is the 100 million that supposedly is from Google, where they would put in credits, but also cash. And they moved away from AWS, which initially supported Runway as a platform.

But I think, Lutz, you should maybe tell us a little bit what the real product is there. Because it banks on the stable diffusion platform, I think. But what also, what is the real product? What also the real, real product is not just creating videos.

So the product is you create videos with a text prompt. I was one of the first users on Runway. I really liked the idea and I tested it out. What you see on their website looks very good.

It is very hard to replicate. So it’s not like Admit Journey where when you type it in, you very quickly get a good positive success. Runway is very hard. Yeah.

It’s very hard to use. And that the reason is video is made up out of many images. And if you trigger something like stable diffusion or any generative model to create an image, they might make this image slightly different every time. Meaning you put many images together that doesn’t create a video.

Or it might create a video, but it’s jumping, right? It’s jumping, it’s not smooth. It doesn’t fit together. Storylining doesn’t exist.

So it’s actually a very hard problem. Runway is suggesting to have solved. You can download the app. It’s no way where you want to have.

I mean, at Snap, I kind of, you know, at Snap times, like, yes, you can figure out parts of the body from somebody and put a unicorn nose on it or horn on it and spit out a rainbow. Those things are feasible. It’s called motion models. I actually have a Forbes article explaining very simplified how to do those motion models.

That’s not what Runway is aiming to do. However, what they’re aiming to do at the moment, I would say, is not yet working. But it is definitely an industry, give it another six months, a year, and we will see huge changes. And that will have huge implications.

We talked about the future is indie on the movie industry, right? It will completely change the world. It will completely change that setup. Yeah.

And if the listeners search for four pizza restaurant commercial, they actually created something that gone viral online for FAUX. So it is in practice. And I think we all want this to work because it would make so much sense to make certain video productions much easier. Obviously, for the more complex one, we discussed it.

You need the human touch and also you need to find tuning anyways at the end. But it would really help the industry. But it’s not. It seems to be very expensive, right?

It’s video editing. And that’s actually the interesting part, right? How did Runway get their investment? Not in cash, but in cloud credits.

And it reminds me a little bit on the internet hype before the first bubble, where a lot of media companies started to invest in consumer internet startups. Not by giving them cash. But giving them free space on their media portfolios as banner, as real estate, whatsoever. And suddenly, companies like Axel Springer in Germany became a huge portfolio of internet companies because they just used that.

It’s not for free, but they used that lever that they had. And it seems we are seeing here very similar play. It’s a media factory, right? And also companies like Zalando, which is the Zappos of Europe, they really benefited a lot from this media factory deal that was more from a different TV station.

Yeah, but these investments, I mean, we spoke about Anthropix. Google invested 400 million there. Midjourney, we touched upon various times. Google also invested there.

They also invested in character AI, which is more on the storytelling side, character development side. So there seems to be quite some interest to gain, let’s say, probably more ground. By the way, I have a question. By the way, I heard this one podcast with the Midjourney founder.

And he actually said that a while back. And he said, like, I don’t actually need investment. And then later on, well, he did take the credits, the cloud credits from Google. Oh, yeah, well, that is fine.

Yeah. Yeah. So honestly, I mean, we heard so many podcasts where founders were claiming they’re always profitable from day one. And then when you see the numbers, well, so we’re not talking about Midjourney, obviously.

So that’s runway. And we all hope it will work because it makes so much sense in the creative space to have support there. Also, maybe lower the cost. It’s quite repetitive, probably the boring work.

Now, I think when we talk about a bit more the boring stuff of creative work, Synthesia from the UK is, I think, a good example. Because you have these people doing marketing videos, explaining a product or promoting a product. I just saw recently the famous Steve Ballmer video where he promoted. Windows 1.0.

Everybody should watch it. I don’t know if an AI can make this up or him jumping on stage, obviously. But the idea is, hey, a human being connects more with you. You get more trust.

It’s more comfortable. They explain it nicely. I can use every voice, every type of character. So this makes the whole thing much cheaper and faster to produce because also it’s a marketing video.

I have to iterate. I have to test, right? If I do a TV spot, it’s kind of a one off. I have to hope it will work and then it doesn’t.

And yeah, next one. Yeah. That is the brilliance of Synthesia. I really like Synthesia because Synthesia does not go into the world and saying, I’m going to do generative AI and it’s going to be awesome.

No, they say, well, that is a value created by marketing material that costs money to generate. That gives you a workflow which is complicated. I will simplify this. I create value for you.

And yes, I use AI for that. What I’m very much looking for when I see pitches, people who tell me pitches about AI, shrugging my shoulders and saying, that’s not necessarily what you want to hear. You want to hear how you generate value. Now, in turn, fun fact, when I’m pitching my businesses, I just pitched it recently and the guy said, this is all good.

Can you put a few more, like real story? Can you put a few more times AI into the text? Just for me. Randomly fine.

But aren’t you excited that I actually make money and have a business? But it was like, no, but we need some AI. Give me, this is very similar to Monty Python’s discussion about I need the machine with a ping. Nowadays, I need to pitch with AI.

So, okay. Don’t worry that I make money. Like it has to say the words AI. Otherwise it’s not worthwhile.

Now, Syntesia is actually the antithesis for this, which I really like. I mean, there’s a problem. We solve this problem. We solve this problem better than anybody else.

And by the way, I tried to get on Syntesia’s, just to be like a trial list very early on. I failed at that time, but they are so awesome in the approach in the UX and how they use it. It makes sense. I really like that investment.

Well done, Axel. Yeah. And then I think all, especially well done, Matt Chirk from FirstMark. A funny guy to follow of those of you who don’t, haven’t done it yet with, if you like memes.

Very good character, French, good humor. And he also writes, and those of you who listened to our previous podcast, he writes about exactly what you described, Lutz. So some part of the AI, if it’s proprietary, does matter because you can influence the product experience for the customer. It’s feedback loop from the users, the data.

And this, I mean, Syntesia was founded way back in 2017, I think. So they had some time to iterate here. And at the time it wasn’t even generative AI. I think they called it kind of creative AI or something.

So there was a strong focus on the output and the workflow. That’s why they built their own AI. And it’s not just AI as you described, Lutz. It’s the workflow.

So yes, I get a character talking, but maybe I want to alter the nose. Maybe I want to alter the voice. Maybe I want to put it into a PowerPoint. Maybe I want to publish it online.

And all these need solutions because else I would have to do it. Yeah. But otherwise, I would have to do it myself with coding by hand, connecting. And this is the key part so far, what I read, why Syntesia is very successful.

Workflow, UX. We said it, I think, in every episode, UX wins. Why is GPT probably better than just AI? Because the workflow is way more easy to be utilized.

If you write an article. Okay. In Just AI, you have to go on the website, click in this box, select here, what do you want, what type. In GPT, you just have your prompt.

And I think that workflow, in what we see for Synthesia, workflow, again, wins. It’s slick. It’s easy. Yeah.

And that’s why, as you described, you’re still not 100% happy with Runway. But they might be getting there because they’re focusing a lot on the workflow to really create something out of that pre-created video, which then is obviously the human touch and all the good ones. So with Synthesia, I mean, there’s a promise, right? Right now, I use Midjourney and I get frustrated after some times because the pictures don’t end up like I like them.

And there’s no real workflow, honestly. Just repeat or reprompt, And here, you have much more control. So I can actually promise to you there are cost savings. You would do it with an agency.

Now with us, much cheaper. And you can see it. We should talk about Midjourney and you getting frustrated. No, I don’t want to talk about your frustration.

I’m not your psychologist. But no, that’s not the point. It has generated something which has a certain style. If you think about the old Midjourney models, they were very easy to identify as, oh, you created that model with Midjourney.

It followed a certain style guide. What we know today is you can train different models differently to follow your personal style. You can train different text models. You can create models to be different in writing style.

You can create models for healthcare who show a lot of empathy. And you can create models for the person who likes shouting. You can fine tune those models. And that fine tuning, whether it is fine tuning or prompt design or prompting or one-shot training, whatever gets you there, creates a different set.

So the future is actually in the ability to put in the right amount of effort. And that has suggested that participants should pick up either a broad range of models, which brings us back to the whole discussion on what KOHIO tries to do, And that has suggested that participants should pick up either a broad range of models, which brings us back to the whole discussion on what KOHIO tries to do, use case to one specific area. And Centesia is doing one specific area. If you want a marketing video, somebody explaining your product, you probably do not want a bloody Japanese cartoon character chopping people’s heads off while screaming loud your marketing brand.

Therefore, they narrowed it by the use case down. So either you have a platform or you have a very narrow, very big use case. Yeah. Yeah.

And I mean, that’s also shown in traction. So far, the traction sounds a bit like, so they’re saying they have user growth of more than 440%. You know, in a year, 12 million videos created, 50,000 businesses are using it. Wow.

Not so sure how many of them are paying, but they say they have already 35% of the Fortune 100! So they’re kind of using it, probably not all the same amount, but those usually pay for what they do. So that’s actually, yeah, pretty impressive traction, I would say. And also the angels, they’re joined now, and this is a later round.

I mean, we have the founder of Scale, we have the founder of Webflow, one of the founders, Miro, Datadog, that’s Figma, only the CFO, but still. So it seems to be something that is taking off in the AI space. Now, that leads us actually to our third deep dive, right? Because also there, we have, we have a very specific niche workflow.

It’s not niche, but like a very specific workflow, which has a big enough market to get people excited. Andreessen Horowitz is in it, which is Hippocratic. What’s Hippocratic doing? It’s actually an interesting name, right?

It’s the Hippocratic Oath, I guess they derived it from. So the general pitch, the first pitch you read is like they’re building large language models, but specifically for healthcare tasks. So you would get a lot of people who are like, oh, I’m going to get a bit nervous, I guess. But they use proprietary data to actually do recommendations.

And it sounds a bit like a chatbot. So I have certain questions maybe for my dietary, maybe for my test results somewhere in the database and I can interact with it, but it guarantees that it’s only looking at this particular data, it’s safe and secure and that I get correct answers. So a bit similar to Anthropic, but a more focused use case, I assume. It’s an interesting one.

I get correct answers. The tricky part for me here is you only get correct answers in a very narrow defined area. That’s the guardrails. And you said this is super important.

All of this works. But now that very defined area, like in the discussion we just had about Centesia, is the value creation. So let’s quickly think about what we have seen. We have seen many tools out there, which starting with IBM Watson, Mm-hmm.

Oh, we replace doctors. Guess what? Clinical things are complicated. Guess what?

Doctors have established a long tradition of finding clues in behavior, in information they see, and they react to that. Guess what? It’s not like an easy way to just take a book and saying, this is what you have, period. That’s the reason why in my view, IBM Watson was just too early, but their biggest mistake was they thought about replacing doctor instead of supercharging doctor.

Now, in the industry, we see a lot of things working where you work with a doctor, you support a doctor and Hippocratic is not trying like other models, which we have seen out there to create the answer and saying, this is what you have, period. No, they say, if you have a negative result, where the doctor says, okay, we checked your blood and it’s not it. That’s not the problem. Then that’s a very easy conversation to have, but that you get the bot.

The bot makes that easy conversation. If the doctor says, oh, by the way, you are diabetes, eat better and exercise more. And you want to have questions of why that is. That’s an easy discussion to have.

You get a very defined one, but if you have symptoms and you would like to get a symptom checker, and we have seen many symptom checkers out there, it’s actually not so easy anymore. And you probably want to have a human in the loop, but that’s not what Hippocratic is doing. Hippocratic takes, like Sympathia, a very defined workflow and works only on that in order to ensure that can deliver value and quality. And I also, when I, when we read the information correctly, they also spend quite a lot of time on making or creating, integrating, I think is the best word, those guardrails out there, which are healthcare certifications.

So they claim they have more than OpenAI. And I was pretty surprised by the way that OpenAI has a participants participants participants already has more than 100 on the platform. Hippocratic has more, but still. And I think the interesting part here is, as you mentioned, it’s kind of, if it works, it’s kind of the best doctor you can have, but around the corner, empowering a local person.

And I think that’s, I’m living here in Berlin, I sometimes wish to have, so I don’t care where I’m going. I’m just looking for, you know, the nicest doctor and more closely. But I know they’re always up to speed, up to date when it comes to recent topics. And then please, not getting too creative with their work.

I’m actually at the moment writing a Forbes article about how to like measure clinical outcomes. And when you say the nicest doctor, it’s like all girls are like, what does he mean by nicest? Anyhow. Yeah, I think my wife and I have a different opinion there even.

So yeah, what does it mean? But think about it. Hippocratic is closing the gap between you talk to a doctor or a nurse, or you do a Google. Search, right?

If you do a Google search, you might get all kinds of information and you might not be guided through the right setup. And it takes you forever. If you ask a clinical question, already today to the chat GPT, there are a lot of, there are still false positives on false storylines and still hallucination, all true. But already there, you get very good answers, which you cannot so easily get from Google.

Mm-hmm. Yeah. Create what they call one box. So if you search, let’s say you type in diabetes in Google, then you get this one box at the side.

That is actually a written, a human created box where there is editorial content in it from Cleveland Clinic or somebody else, whoever it says in the bottom, who is actually has written this. But you need to update it and you, it might not be 100% the focus of your question. And then you click. So, okay, you can click on symptoms, treatment, and so on in your one box and click through.

Then in today’s large language model is a completely outdated interface. You want to know what is the problem and you want to talk to somebody who understands the same information. And that’s what Hippocratic actually tries to do. I love it.

It is very specific, very clearly controllable, beautiful. Interface. So for me, it is well done. Andres Norowitz and General Catalyst.

Spar. Which are both by the way, tier one funds from the Valley for those who don’t know them. And it’s a seed round. So seed is kind of one of the first rounds you would do and it’s 50 million.

So it’s, it is a big problem. It’s very impressive what they have done. By the way, we spoke about this citation and controlling the model about an anthropic coherence. So it’s always kind of in the same area, but as you just described it, now we have the workflow now with the focus.

And so the output quality is hopefully then better, but let’s see, it’s, it’s very early. So there wasn’t a lot of noise around the founder of Hippocratic actually as well. Like he screwed, he wiped out investors and screwed them, the former CEO. Yeah.

And people were laughing about it and saying, you know, a founder screwing their employees, not paying the employees. Screwing the investors. But then you do something with AI and everything is forgotten and you get another 50 million. And they, they called out on Silicon Valley’s disability to remember.

But I think the question here is, can you create value? And is there, are you on top of the technology, which is on the forefront and creating value and therefore Silicon Valley is probably. Forgiving. I don’t say it’s right.

I don’t, I am not in the details, but I, I just thought we should call it out that this is beyond Hippocratic being a nice Hippocrates discussion. Yeah. It’s a very good point. And those of you who want to read about it, it was yesterday on Forbes, one article, I think it’s always tough if you have a founder ideating the next idea in his previous company, especially if that company doesn’t really work out and has other investors, so there’s a lot of conflict of interest that’s very, very tough to navigate.

Also. We have some of those, are some of the things that we want to make sure we are not putting into diligence. I mean, yes, we we’ve seen this also in, in Europe. It’s a tough one.

Yeah. IP rights, IP rights. That’s the three big ones to, to sum it up. So we had the rounds of Cohere 217 million, 2 billion valuation, Anthropic.

That’s also more than 2 billion. So Cohere and both Anthropic with corporate involvement, for example, Google put both of them big. Runway, Synthesia was more on the application side. corporate involvement, Google, Runway, Synthesia, NVIDIA, smaller rounds, 100 million and 90 million.

But it kind of feels there is, obviously the rounds are large. So people trust that these things are real and something will happen. Now you could argue that same round size we had in crypto, but we think at least it’s different here. Plus on Runway and Synthesia, the applications are very concrete.

There’s traction. People are doing a lot with it. You can monetize on it. The usage is on a corporate level, publicly listed companies.

And then Hippocratic, early seed round, but huge problem, strong team, strong backers. And they’re approaching this very similarly to the larger companies, Cohere and Anthropic, but on a very focused use case and also with focus on the whole workflow around healthcare. So very interesting one. I think for all three deals, it comes down to, we are past this, oh, we, create general AI and it will work.

We are past this stage. All the big and smart investors, they are looking very specific towards use cases that create value and that are defined enough to use this tool set, which we have called AI to create a competitive advantage. Defined use case with a defined flow. Is it guardrails?

Is it distribution? Is it a marketing workflow? Is it a consumer bot workflow? A defined flow and use AI to make value.

And because you mentioned it before we come to our two companies that people should look out for, I’m still lagging a little bit the consumer cases here, and we would love to see more consumer cases. It just makes so much sense to build consumer products around like with AI, with the experience of AI. Hippocratic is a consumer case, right? Yeah.

Yeah. You’re true. Yeah, that’s right. It will be profitable.

It will be in a corporate sales environment, but it’s for the end consumer. You’re totally right. Who are the companies we need to watch out for? Yeah.

And dear listeners, at the end of the podcast, we wanted to quickly highlight two companies we think you should watch out for, and they’re both from Europe. So one is called FlowX AI from Romania. The other is called Accelera.ai, I assume, from Eindhoven in the Netherlands. Let me start with FlowX.ai.

So highly competitive round. It’s a series A, 35 million. Dawn Capital, from London won it. So congratulations to our colleagues there.

What the startup does is it helps organizations to automate digitalization. Wow, that’s a difficult word for German processes. So the idea here is basically that you use AI and with the help of AI, like everything you have as large corporations, as software applications, it’s all very scattered. We even heard from companies that would use for the same kind of use case, but different departments would use different software.

So you manage to actually integrate all of this software on a single platform where you can manage it, where you have an overview and where you get a feeling of control. Now, this sounds crazily complex. So far, the company has generated 1.55 million in revenue last year, 2022. It was a 700 plus percent year on year growth.

So that’s crazy. So something is working there. A lot of the backers, by the way, come from the UiPath people. Which also automates processes more on a script base.

This wants to do much more, much broader. They have large customers. That sounds great. So there’s a European bank, all more on the Eastern European front.

But apparently, I mean, Dawn does very good due diligence. There is a software working. We are a bit critical here because we’ve seen those automation cases over and over again. And what is working, coming to the focus topic that we have said before, UiPath, started or actually grew quite fast when they had a very clear automation case.

So something that was outsourced before was documented as a process. People, so people were doing it and now the script would do it. So it was kind of, you just rip and replace people with a script, very clear case. And everything around it was really tough to build.

Now this is a broad approach again. So probably a lot of enterprise customization work, consulting work, and really figure out what the use cases are. However, the round is big, so I’m pretty sure, So I’m pretty sure, it’s not going to be a big deal. So don’t get too looked at this.

So watch out for FlowX.ai from Romania. And now the next one. The next one is Celera from Eindhoven, beautiful town in the Netherlands. They raised their 23 million as a series A.

What I like about them, so they are essentially doing computational devices at the edge of the network. So you have seen NVIDIA, they are doubling their share price. You have seen all the big corporations fighting for dominance in the space of doing like all the computational power. And we discussed a while back that there will be players who come up with fine tuned hardware, with dedicated hardware.

And here’s one, like Celera is one that actually focuses only on the edge devices. And we all know how this looks. If you can download it, you can download your OpenAI, like ChatGPT model already on your phone, or StableDiffusion actually works on your phone. That is, your phone is an edge device.

Your phone is not very powerful compared to one of the big computers you could get from Amazon, Google, or like anything which has a huge NVIDIA chip in it. Amazon and Google don’t have one, but let’s say any of those three, they kind of focus on small edge computational devices. So pretty, amazing company, pretty amazing niche. And let’s watch what’s happening.

And we heard they already signed some contracts with phone manufacturers, which will probably take some years, but if this really materializes, this is huge for a European company. And it’s, I mean, like think about it. If you have two phones, one is the phone and the other one is way better. You have already that fight on between Apple and Pixel or like iPhone and Pixel that Google says, I make better pictures.

I make better pictures because they actually use AI to make the picture look nice. Well, Apple says, but I have the better quality lens. We said, well, that quality lens doesn’t help you at night, but I just identified that you want to make a photo of the moon and I just printed a photo of the moon. How about that?

Yeah. And I heard also, by the way, the face ID from Apple is with probably a similar chip. So let’s see, let’s see how this works out. Let’s yeah.

I think this new phone, this new format, at least for me, it was a lot of fun. So thanks for your time. Absolutely. I wouldn’t call it new format though.

It’s we should probably do both that we straddle the world. So we get some content and then like, but the investment discussion is it’s fun. Totally. Nice.

And this is also, by the way, it comes from feedback from the users. So not only us are working here. Thanks for your feedback. And yeah, we appreciate to have more of this.

Tell us more what you want to hear. And if you want to get our feedback on you. So you can also send us a message. So we will pitch it and we will talk about you.

I don’t know whether this is a good thing though. Let’s see another experiment for us. Have a, have a good day. Have a good one.