S7E3 'Can AI ever go Green?', with Robert Keus of GreenPT 🤖

James Gill:

Hi, everyone. Welcome to another episode of the EcoSend podcast. I'm thrilled to be back. I'm your host, James, and today's episode is going to be all about a topic that I know a lot of us are very interested in, perhaps can't even escape right now, and that is AI. AI and the world of sustainability.

James Gill:

And I am joined by none other than Robert Koos of GreenPT, who is building basically a sustainable or a more sustainable alternative to OpenAI's ChatGPT. Robert is the founder of multiple businesses including Brothers, GreenPT and Accessibility Desk. And he's basically focused on and committed to revolutionising the way technology intersects with society. Robert, I am thrilled to be talking to you today. How's your day going so far?

Robert Keus:

Thank you for having me, James. Yeah, it was good. We talked about it a little bit too many meetings, but I'm really enjoying being here.

James Gill:

Amazing, it's a pleasure to have you here. Whereabouts are you joining us from? Where in the world are you today?

Robert Keus:

We are based in The Netherlands and then the city Utrecht, I think it's half hour from Amsterdam.

James Gill:

Amazing, big fan of the This is great. So you are clearly a very busy man Robert with multiple businesses, multiple hats to wear. Tell us a little bit more about what you're working on and yourself.

Robert Keus:

Yeah, so a little backstory. Like from a family where my father's entrepreneur and I always wanted to have my own business. And I basically founded my first company when I was 14. It was a company, we were repairing computers in a neighborhood. It was not like there was no social aspect of this proposition.

Robert Keus:

We did it for the money because I wanted to buy Xbox. That was basically the incentive of having this. But later on, if we do it fast forward to after my student time, we founded this agency and we called it Grodders. Was the same friend I founded the first company when I was 14. We were best friends from The Hague in Netherlands and we lived in the same street.

Robert Keus:

And so if you have an agency or a company, it's always really, it gives a certain satisfaction. If you provide services, somebody else is using. So if I build a product, a digital product for somebody, it can be a small website or app or whatever, and it's live and somebody is using it. That's super cool. That gave us like a really satisfied feeling because you build it yourself.

Robert Keus:

You explored the process of building this product. But like, I don't know how we got to that, but like an organization from The Netherlands, they contacted us and they are in the research in drugs addictions and smoke addictions. And they ask us to build a product for them. And then after that, we had like even a better feeling. And we were like saying to each other, Hey, we really enjoyed this product.

Robert Keus:

And also there was a really good fit with the client and the culture we had in like Sebastian, my former business partner, we had a really good connection with them. So then we decided to say, okay, we want to work for companies and products with a social element because you have like a sort of a double satisfaction feeling of the result. And that was really interesting because first of all, somebody's using it, but because of using it, it also solves a real problem. It makes impact basically. That was the moment we decided as brothers take the digital agency to focus on products with a social element.

Robert Keus:

And from there, because we were in the social space, accessibility came around, which is still a really hot topic in Europe and also in The Netherlands. And I really saw like a demand at the governmental organization. Then we thought, Hey, we provide the services from the agency, but it was too off topic because the services were totally different than building software. So I decided to to found a new entity, set up a new entity and put the services, the accessibility services in there. And that's, both companies right now we are with 35 people.

Robert Keus:

Okay, yeah. Yeah, it was two weeks ago, 34, new hire 35. Oh wow, yeah. Yeah, we spoke about two weeks ago and I said to 34. And obviously like as an agency in this time, when you operate in this, I didn't say period in the world, this timeframe in the world, AI is a hot topic, but for us it was really a struggle.

Robert Keus:

How are we going to implement this in our businesses? Because we also want to work for social companies with social products and AI has a lot of value, but also has a lot of like gets a lot of challenges. Let's say, that's very much like that. Next to privacy, security, ethical challenges, hallucinations, but also it has like a large footprint in the environment talking about sustainability. So a lot of our clients, they basically didn't really ask about a sustainable solution, but they ask about a more privacy friendly solution.

James Gill:

Okay, interesting, yeah.

Robert Keus:

Which was not hosted in like AI they use was not hosted in The US. Yeah. And so what we basically did was we said, okay, yeah, you can use OpenAI for Azure and they said, okay, that's fine. So that was basically our way to go was saying, yeah, we use OpenAI for Azure. We can put it like the servers Europe and then it's fine.

Robert Keus:

Most of the time they already use Microsoft services. So that was fine, but more and more, I think also because of the political climate in The US right now, they said, Yeah, I don't know this Azure thing. Do you have something else? And there was also the moment the open source models of the open source AI models became more, how do you say, yeah, more as strong and better. And they were challenging like JGPT for and for our mini.

Robert Keus:

And even the reasoning models, if you talk about AI, you have like different types of models and one of them are reasoning models. Also DeepSeek came on the market with their open source model. And this was also challenging the large models. So, and also LAMA from Meta. Meta also has like really large open source, AI open source environment and infrastructure.

Robert Keus:

So we thought, okay, so what we can do, we can go to hosting providers and ask them, Hey, can you help us? And can you, we want to basically have our own AI. So we have like a, we have a proposition to our clients so we can sell that to them. And basically from the agency, already hosted all our stack at a Dutch environment friendly green hosting provider. We got in touch with them and we said, okay, can you help us with And from there, did a lot of research to which kind of models which we could were the best for our clients.

Robert Keus:

Amazing. So basically when we set this up, we had this environment and this API, you need some knowledge to do all this. It's not really easy, but we were able to do it. We went to our clients and said, Hey, we have this API for you. We want to like change your OpenAI API and we have our own.

Robert Keus:

Would you like us to do that? And they said, Woah, that's super cool. It's sustainable. So I would just say like, Yeah, do you like that it's sustainable? They said, yes, it's better than that.

Robert Keus:

Liked that more than the privacy thing. That's I think that's pretty And that was the thing that I pitched this and I showed this like internal API we use for our clients to other clients. Everybody said, Hey, this is super, super cool. So then I thought, okay, so many people are, their reaction is so positive. Let's build a product around it and let's see if, and also open up our API.

Robert Keus:

And I think now, yeah, we released it about six weeks ago. Yeah, that's a little bit the backstory about

James Gill:

it. Yeah, Robert, that's incredible. I feel like in about ten minutes you've given me, at least from the age of 14 to now, whole story, which is incredible.

Robert Keus:

Not my personal life, I don't know.

James Gill:

Yeah, yeah, none of the personal lifestyle, none of that nonsense.

Robert Keus:

Another podcast. What

James Gill:

a story though, there's so much in there that resonates for me as well. Think maybe similar paths we've been on in different countries and starting young and getting this joy out of building things on the web. I share that whole incredibly fulfilling sense of when you've built something and other people find it useful. I think that's one of the wonderful things about the web more so maybe today than ever, where anyone can build a website or a tool and see other people get value of it. And you don't really have to ask for permission to do that, you can just go and do it.

James Gill:

And it's incredible to hear how that's been that spark for you that got you started. It's interesting as well, I hugely respect your focus on accessibility as well. These fields of sustainability, accessibility and ethics, and all of these worlds are like, they're not totally separate from each other. Think they really overlap with each other. And I think it's just so interesting to hear your work across those different disciplines.

James Gill:

And now, I mean, gosh, I don't think there's been quite a hot topic like AI that I've ever seen before. I think we're far beyond this being a buzzword and there's so many small and huge questions about the consequences of AI going forward. But I think one thing I'm sure a lot of our listeners would be thinking of is the sustainability side of AI. I think arguably if there's one good thing about the sustainability space and digital sustainability in AI, it's that it's actually brought to attention for a lot of people that our usage of the web does still have a physical presence and has sustainability consequences. I'd love to hear a little bit more from you about how big of a deal was that in your mind when you were looking at this?

James Gill:

Like, how were you using AI yourself, but conscious of the sustainability piece? Was it something you tried to understand by

Robert Keus:

did you

James Gill:

try and speak to, like, any of the providers about their own sustainability credentials. I'm curious how bad is it? Are there any things that sustainability wise that people don't normally think of when it comes to AI?

Robert Keus:

Yeah, think you have like, first of about sustainability in AI to keep it simple. First of all, you have the servers. Where do they run? So hosting AI models, next to it, it's super expensive. It is also using a lot of energy.

Robert Keus:

Because of the models are really large and if you send a prompt or do a request to a server, the server needs to do a lot of compute power to give their answer back. So first of all, have like, so you have to compute power, which is really important. So if you talk about green AI, which currently there is not really, there's no green AI, it's not there yet. So that's also something I would want to say. It's not that we don't say that we are totally green.

Robert Keus:

The reason for that is, so first of all, you have your question or your prompt to the server. You'll get your answer back. So the hosting provider is playing a large role how sustainable your AI is. But next to that, you also have the training of the models. So what we currently do is we use open source models, but the training of these models are not trained, or let's say, maybe rephrase it.

Robert Keus:

We can't prove that it was trained or not trained with with green servers. Even I know that how much compute power you need to train these models. I can say that it wasn't trained on green I know the new meta model that they have trained this on servers with renewable energy, but I still found it a little bit vague how they phrased this. That are basically the two parts which are really important in the energy conception of the AI models. What we try to do is basically next to hosting it is be really like doing a lot of research in what kind of models you want to give to the users.

Robert Keus:

So currently, we can run larger models than we do now. But in our research, in the research we did is that the quality the response was not, didn't like gave the quality of the response was not that much better than the smaller models we use now than using larger models. Interesting. Right now, so what I always say is 95% of the people using AI, they can perform their task with smaller models than when using like large models as a ChatGPT or Gemini or Cloud. So we run smaller models.

Robert Keus:

So let's say they are like twice or a third of like the larger models that directly correlate with how many people we can onboard on one server. So more service people on one server, the more sustainable your product is, but next to that, so the decision of which models you use is important. Your your hosting provider is really important. But next to that, which is a really also large portion of sustainability, and it's also something we already improve, but can do better is things in prompting. So what is really important is if you ask a question and you get like a large answer back, it needs more compute power.

Robert Keus:

So if you ask a question and this question is rephrased in the correct way, and it gives you a shorter answer, but it still gives you the answer you want. You have like a win in sustainability because need less compute power from the server. So what we are trying to do is to find a way where your answer is shorter, but your question is also shorter or like more sustainable.

James Gill:

Right, yeah.

Robert Keus:

But you still get an answer back, which gives you trust or gives you value, because that's a little bit of a balance. We can always say, always answer in one sentence, for example, but then there's no value for the user. Also that's something I think we really can do better in the future is like, this was just about sending a request and getting a request back like a single question, but also the amount of steps you getting to your answer is also really important. So for example, if you ask AI to make a quotation for you or a description of your quotation, what now a lot of time happens if you use ChatGPT or Cloud or other API, API AI services or chats, is that people need really like need a lot of questions to come to the correct answer. Basically people are bad at prompting.

Robert Keus:

Yeah, people are bad at prompting. Ask like really like, yeah, they ask the short questions without context. They don't define the role, etcetera. So that's also the reason you have a lot of like prompt specialists out there. I'm not sure if it's in The UK, but it's like

James Gill:

Yeah, they do exist.

Robert Keus:

Yeah, do they exist? Are a lot of prompt specialists. I think it's the interesting thing is that they, I think they can add a lot of value to sustainability as well because if they learn people to use less prompts to get to their answer, it has like a direct influence on sustainability. That's a really good point. That are the things we are working on right now.

Robert Keus:

I could tell more about that. Amazing. That's really interesting.

James Gill:

I actually wasn't super aware of some of those things you said there, Robert, especially around the length of response. And I still consider myself pretty a beginner on the AI side of things. So I'm learning a lot. I hope our listeners I'm sure are as well. I was having a little bit of a play around with GreenPT earlier and I love the interface.

James Gill:

Just for anyone listening who's wondering, is it something new to learn? I think what I found very interesting is it's pretty much it feels just like you're using ChatGPT, but you're doing all of this in the background to reduce the environmental impact, which is wonderful. One of the interesting things I noticed, say you don't do the image generation at the moment. No. And I was curious, that's clearly a choice and how big of a deal is that for us?

James Gill:

Because I know now with ChatGPT, you can kind of just say, give me an image and they've made it very easy to generate images. How big of a deal is it for anyone from the sustainability side thinking whether or not to do that?

Robert Keus:

Yeah, so I think this is also a question. We thought about this carefully. First of all, to set this up, you also need a lot of investments and servers and etcetera. We can run these models. We're able to do this.

Robert Keus:

Not as good as ChatGPT right now because the open source models are not there, which are as like the quality they have, but even if you have the quality of the latest version they just released, you should ask yourself the question. It's not even a sustainability question you should ask yourself, Maybe the second one, but the first one is, do I always want to have a perfect image? Like, does it always like, I basically don't see the value as a company, like using this as a company for marketing purposes. If I see this company using these images and they use it in their marketing like coverages, would say, yeah, okay, but I see it. I see that it's a perfect image.

Robert Keus:

You can see it in the skin of people. You can see it in the background. Maybe not everybody is seeing that, but I see it. That's also from that perspective, we can't challenge them in the quality. So because we are not, even we are not like close to what they now have, because it's pretty, it's pretty good, but still I think it's too perfect.

Robert Keus:

If we want to have this, we will give our users something which is not as good as they have, and it uses a lot of compute power to generate it, a lot, a lot, a lot. So right now I don't see the value of adding this when other people are like so much better in it. And even if we have it, still see it. Maybe you can, I don't know? Maybe you can prompt it.

Robert Keus:

It less real.

James Gill:

Less perfect.

Robert Keus:

That's perfect. Maybe you can, don't know, I didn't try.

James Gill:

I'll find a prompt expert to help.

Robert Keus:

Yeah, yeah.

James Gill:

I appreciate that context there, Robert. I guess I know a lot of what we've been talking about has been the sustainability side of things, which I think is obviously incredibly important to our listeners. I think with AI in particular, there's this whole other question of, at least that I have around ethics and the ethical consequences of AI. I almost feel a little bit old school, even talking about this in a world where seemingly, if you at least read the headlines or pay attention to any kind of business news, everyone's using AI all the time to do everything. But what are your thoughts on AI usage from an ethical standpoint in terms of the content that it is trained on and perhaps maybe leaning into the future of where does this lead us to and what does it mean for certain roles and jobs?

Robert Keus:

Yeah, I think that's an interesting question. So first of all, I think from an ethical way, I think what will be an interesting discussion, maybe it's more on a higher level is as a European country with European values. I say UK.

James Gill:

We're still pretty close.

Robert Keus:

Yeah, pretty close. European countries. Not that you say you're not part of the EU, but you're European

James Gill:

We actually moved The UK a little bit further away.

Robert Keus:

No, as the European countries with European values. We should ask ourselves, how are these models trained? And I'm not talking about the copyright material in there, because that's my second topic, but also about, are they all reflecting our values? And it's not only, so right now there's a large discussion. I was at a governmental organization and they said, yeah, we don't want to use any Chinese models open source.

Robert Keus:

And I said to them, okay, but you do want to use The US models then? Let's see. So because obviously, because in The US models, there's also a lot of like ways of thinking and also which maybe not our values. I'm not judging, but it's more that you, I think that is a discussion we should have. And it's the same with the Chinese models also have like values we don't share with the people from like the Chinese government.

Robert Keus:

So our country, maybe the culture. So what I think it's really interesting is because of the development and it will get more easy to train your own models. Obviously you need more and more, you need a lot of budgets to do it, but I think it will get easier and the information to train the models will be there. I think it's also interesting that we also have like our European alternatives, like Bistral, for example, which is also getting better and better, which is more reflecting maybe our values. Maybe, I'm not sure if you can talk about European values, but

James Gill:

still I It's a really good point and I think the start of 2025 has shown just how much the world stage can shift under our feet. Yeah, perhaps some of those conversations about concerns around things being hosted or run by US companies wasn't really in the kind of, at least not in the day to day conversation, if you rewind five or six months ago, now you look at that and I know EcoSen, we see quite a few customers coming to us just simply because we are not an American company. And there's very real world impacts of that and its impact on AI and the biases of any of these models is, yeah, I think it's such a good point to raise. Yeah.

Robert Keus:

And next to that also you have the copyright subject. I read a tweet today, like tweet or I don't have x when I saw it on LinkedIn. So I copy pasted it like as a screenshot.

James Gill:

I'm not here to judge Robert.

Robert Keus:

No, no, no, no, all the regulations and law around IP and copyright laws should get dismissed or something, because then it will be easier to train the models. But on the other hand, I also read something, saw LinkedIn like last week about somebody calculated how much it costs OpenAI just to license the data to train their models. That's the other thing. If you get like a 40,000,000,000 investment and your government is also going to invest 500,000,000,000 in AI, why aren't you just can't pay for it? Right?

James Gill:

That's a good point. You do raise a very good point,

Robert Keus:

Rov. Yeah, I think it's interesting on the other end.

James Gill:

This has been such a fascinating chat. The time has flown by Robert. We've already done half an hour. I didn't know if there was any final thoughts you wanted to share, whether that's thoughts about the future, thoughts about what's next for you, anything you wanted to close on that would be any hope

Robert Keus:

in the future? Yeah, hope there will be. Think the hope is that AI is already there for a long time, but for us, it's like, let's say the people using it in real life, it's there for a couple of years. And I think we still are in the beginning of everything. And I see that also as more as a good thing that regulation and the way we want to interact with it and how things will evolve, they will need to find their own path.

Robert Keus:

Believe that it will be a positive way. I really enjoy being here, but also from a Greenpurity perspective, we try to work hard to get like the most green AI out there. And hopefully we can also prove in the future that our models are also trained with green energy and hopefully, but that depends on how good this goes that we can help or maybe we can also train our own models, but this is something which is not like right now that is that, yeah, you need so much investment for that.

James Gill:

You need some of that 40,000,000,000 coming your way, Robert, right?

Robert Keus:

Yeah. Maybe 1%, I don't know.

James Gill:

Yeah, just 1% would

Robert Keus:

be fine. 1%, one That'll do, that'll do. Yeah, yeah, yeah. This has been

James Gill:

such pleasure to speak with you. For anyone listening who wants to go, maybe go use GreenPT, where should they head? Yeah,

Robert Keus:

you can go to greenptai. And if you would like to get a waiting list invite, you go to greatpt.ai/invite. Amazing. And yeah, we onboard I think the last two days already 200, three hundred people. Amazing.

Robert Keus:

I think you will get a, yeah, in the near future you will get an invite to the account.

James Gill:

Exciting. We'll put those links in the notes and yeah, fantastic to chat with you Robert. Really great work on such an important space. I think AI is gonna affect every single one of us if it hasn't already in some small and some big ways. To see your stance and your focus on doing it right is just incredibly inspiring and heartwarming, honestly.

James Gill:

So thank you for sharing your story today, Robert. I really hope other people will really enjoy this episode. So thank you very much.

Robert Keus:

Yeah. You too. Thank you for having me.

James Gill:

Thank you also for listening. So if you've enjoyed today's show, please let us know. Please tell your friends, family, colleagues, dog, cat, everyone about the episode and, make sure you're you're going and checking out Robert's, great work. So thank you for tuning in and we'll see you next time.

Creators and Guests

Robert Keus
Guest
Robert Keus
Robert is the founder of Brthrs, GreenPT, and Accessibility Desk, he’s committed to revolutionising the way technology intersects with society.
S7E3 'Can AI ever go Green?', with Robert Keus of GreenPT 🤖
Broadcast by