00:01.39
interesting
Hello and welcome to lexicon I’m Christopher Mcfadden a contributing writer for interesting engineering today. We are pleased pleased to introduce. Ah Yaoswa Jappsky Monomo’s chief technology officer. Ah, he leads the company’s technological development oversees the hardware and software development pipelines and directs machine learning research monomo the company explains is using a unique set of data and machine learning techniques to build what it calls are the world’s first large engineering models or elm for short. Before joining Monomo Jaroslov was an integral part of Codemaster’s team behind the racing video games grid and dirt 2 he has also worked as a software engineer at Siemens and held senior roles at Microsoft research and arm Today. We’ll discuss what monomo’s em is how it works. And how it could revolutionize engineering research and development now and in the future yaraswaff. Thanks for joining me.
00:57.99
Jarek
Hello and nice to meet you and I’m very happy to be here.
01:01.74
interesting
Fantastic And let’s get the ball rolling then would you like to tell us about your background and what drew you to work for Monomo and what you guys are up to over there.
01:13.47
Jarek
Yeah, definitely? um so a little bit about myself I’m a physicist by education a very long time ago I did my masters in theoretical physics and then a ph d in astrophysics and but I left academia more than twenty years ago and joined and industry as you mentioned basically started in the video game industry which is which is quite interesting after doing a pgd in astrophysics as you can imagine but it was a lot of fun and and I learned a lot about you know, software development and basically working in in in this industry. And but even more importantly I got ah a good sense for what it means to deliver a product and what it means to see a product being used by other people because you know if you work on a video game and then you go to a shop. And you see that video game on the shelf and you know that you worked on it and you see other people play it and enjoy it It gives you a certain ah you know feeling of pride and I really really enjoyed that I really enjoyed seeing other people use the sort of output of my work and. Moved around ah different software engineering companies as you mentioned and and and most recently was at Microsoft Research working on machine learning there and then from there joined monumento and so as you can see I covered everything from sort of pure science and theoretical physics.
02:38.44
Jarek
Ah, through video games through software engineering machine learning and and and Data science.
02:44.30
interesting
Interesting Journey Sure you’ve had a ref of filming career. So forth. Yeah.
02:46.32
Jarek
I enjoyed it I Really I Really love the variety right? I Really like when you sort of apply different expertise from different domains into new problems. It’s fantastic I Think that’s where that’s where innovation happens.
02:55.94
interesting
Absolutely have you found much overlap as you’re moving through changing different disciplines.
03:05.11
Jarek
Um, a little bit I mean obviously 1 thing that um is definitely prevailing in all of those is problem solving. So obviously you know in physics you need to solve problems in software development. You need to solve problems in engineering you need to solve problems.
03:12.81
interesting
Thrice.
03:20.48
Jarek
So obviously problem solving is definitely one of those things. Um, but you know if there was a lot of overlap then there would also be very little learning of new things right? because the more overlap there is the less you learn right? because you can use all your previous knowledge. So I think it’s good if there is a little bit of an overlap but not too much because that.
03:29.74
interesting
The first.
03:39.62
Jarek
Stimulates you to learn new things.
03:41.90
interesting
Of course of course and also notice me Linkedin profile that you might dabble in martial arts and which which discipline um and do you find it helps you at all in your professional life.
03:53.89
Jarek
Yeah, So it’s a little bit similar to my actual professional journey so I did actually try quite a few different disciplines because as I was moving around in my life. You know I was also changing clubs and obviously then you also change the styles a little bit. Um. I did quite a few different ones I would say that the favorite one I I prefer is kickboxing I’ve done that for probably the longest out of all of them and whether it helps Yes, it does I think it definitely helps with focus. It helps with clearing your mind sometimes.
04:11.75
interesting
Okay.
04:24.10
interesting
Yeah, ah.
04:26.17
Jarek
I do believe that there is a connection between sort of physical and sort of mental health and that we have so finding a good balance is very important and you know I often find it that if I go to a training session and it’s a hard one and the next day your arms and legs hurt a bit.
04:31.30
interesting
That.
04:45.53
Jarek
Um, even though they you know you feel a little bit physically exhausted like mental. You’re actually quite invigorated. You actually have more energy that day to do something that if you would skip that training the previous day. So yes I would say that it does help.
04:50.40
interesting
Um, yeah.
04:58.95
interesting
Fair enough? Yeah yeah, I used to show to can back in the day and oh there we go yeah taught me yeah, many things about discipline and self-respec. Yeah yeah, yeah.
05:03.47
Jarek
Oh yeah, yeah I That was the first Karate style I did. Yeah, yeah, yeah, yeah, yeah, and you know like sort of getting on with it going over difficulties. You know not giving up all of those things you can learn from um from martial arts.
05:18.47
interesting
Absolutely and I’d to take a hit. Let’s be very useful. Yeah I think never back down anyway. Um, so um, what made you ah take the plunge the second employee at Monomo um, was it a big risk.
05:21.26
Jarek
Yeah, yeah, take a hit you know take a hit but then get up again right? That’s that’s important. Yeah.
05:37.17
interesting
The time to to do that for you.
05:39.80
Jarek
Um, there is always some risk involved when you ah when you change right? but you know no risk. No fun right? as they say Um, so yes I think there was a bit of a risk involved. But um as I said I calculated that risk and I thought you know that’s that’s okay.
05:44.44
interesting
Um, yeah.
05:57.30
Jarek
I think that the main reason I mean there are a few reasons why I really liked what what monmo was or is is doing right? So first of all, it’s it’s difficult right? Monument is applying deep tag to optimize engineering design right? That’s like.
06:09.93
interesting
But.
06:15.81
Jarek
1 sentence headline here and I’m sure that everybody in the engineering community who is listening to this podcast will appreciate how difficult it is to actually find the right balance of different components of a complex engineering system if you want to design it right? It’s it’s a difficult problem. So I like that I like that challenge I like difficult problems and applying deep tech to engineering also automatically makes it a multidisciptionary problem because obviously you have to combine you know the latest developments in computer science algorithms optimization math physics. Obviously.
06:42.56
interesting
Can.
06:54.15
Jarek
With Domain expertise in engineering you need to put all of that together and to solve the problem and again I really like multidisciplinary problems I think that that where you actually sort of cross-pollinate you know knowledge from different branches of science is where magic happens. So I really like that and.
07:09.50
interesting
Hey.
07:14.40
Jarek
Um, in a way finally, ah the focus of monument is electric motors and and electric motors are really important for you know green transition of the society they consume 50% of electricity.
07:17.99
interesting
5
07:32.20
Jarek
And so any kind of improvements that we can make to electric motors will have huge positive impact on you know, ecology and our you know movement of the society to towards a more and more green way of of life and I really like that I mean I you know i. Through all my life I was actually quite quite keen on ecology and you know I like nature very much so that sort of resonated with me very well so you know difficult problem multidisciplinal solution and the solution. Once it’s found has definite positive impact on. Um. On on on the society and those 3 things combined made me you know, basically take the plunge and it’s been a great journey for the past two and a half years I yeah as I said about two and a half years ago yeah yeah
08:14.93
interesting
Passing that front time when did you join monomo sorry I forgotten two a half years ago right okay and so can you tell us? Ah, how to answer.
08:30.54
Jarek
And serious. Yeah.
08:32.39
interesting
And and ah large engineering model differs from other popular Ai software like chat gbt or is it is it more like a a dali or a mid journey kind of graphical output. And yeah, basically.
08:43.20
Jarek
Um, ah yeah.
08:48.75
Jarek
So so very good question. So maybe first um, a little bit of clarification. So answer is not the large engineering model answer is ah what we have right now which is our tool for.
08:55.54
interesting
Right.
09:05.29
Jarek
Optimizing and designing electric motors Now there is a very close connection between answer and what large engineering models will be in the future and that connection is that we use answer to generate data that is needed to train machine learning models.
09:08.80
interesting
Frightness.
09:13.10
interesting
K.
09:20.76
interesting
Right.
09:22.96
Jarek
Ah, with the goal of growing those models up and up and up and at some point they’re going to become you know what we’ll call large engineering models now. It’s a journey right? It’s we inside of answer we already use some machine learning. We use that machine learning. To explore more designs and generate more data and the more data we generate the more sophisticated machine learning models. We can train that allows us to again, explore even more design space generate more data more complicated machine learning models and in the end we will. Get to a point where our machine learning models are so sophisticated that they really encapsulate and in a way quote unquote understand enough physics and enough ah engineering to be able to basically propose. Ah. New novel solutions to electric motors and later in the future other engineering applications as well. So answer is basically our current technology that we use to optimize electric motors right? now it is using machine learning under the hood for some things but the primary reason here is to first of all.
10:23.70
interesting
Okay.
10:32.18
interesting
Writes.
10:37.93
Jarek
Already delivered valuable designs to potential customers. So we can already optimize some of the electric motors. But more importantly, it generates data that will need to train those large engineering models in the future.
10:51.33
interesting
Okay, so the data it’s producing presumably ah is only related to electrical motor motors the engineering in that. Yeah.
10:57.31
Jarek
At the moment correctly. Yes, at the moment we are focusing on electric motors and you know in a way we we made this decision very consciously because you don’t really want to try to boil the ocean as they say you know if you try to do too Much. You risk that you’re going to basically fail. And in a way if we can’t show that machine learning and sophisticated you know generative models can bring value to electric Motors. You know as ah as a focal point then. There wouldn’t be much point in trying to do that for a larger volume of engineering We believe that first you know, prove that what you’re doing actually brings value and I don’t necessarily mean only monetary value right? like also sort of intellectual or sort of.
11:35.61
interesting
Yeah, yeah.
11:49.18
Jarek
You know, conceptual value but you need to show that it does bring value and once you do show that then obviously you can keep on growing it outwards and so we did make a decision to focus on electric motors. You know, um specifically. But.
11:50.84
interesting
Yeah.
11:59.41
interesting
I see I see.
12:03.78
Jarek
The techniques that we’re applying and the simulation that we build is a multiphysics simulation so it could be applied to other branches of engineering we are indeed. Yes, we are laying the foundations we are building the simulation that is flexible enough.
12:10.90
interesting
So you lay the foundations like a framework scale that you could build on.
12:20.50
Jarek
And that it can deal with you know, Multiphysics problems. Obviously after electric motors the most immediate step will be generators because those are basically you know the same the same device more or less just operating in ah the other direction. But after that we will probably. You know, explore other more different types of engineering as well.
12:43.90
interesting
Okay, fantastic. And okay, then and so you wouldn’t really call answer an artificial intelligence then Thursday the l e m.
12:52.47
Jarek
I would not call answer an artificial intelligence. No I would call I would call answer as sort of simulation and optimization engine. Yeah.
13:02.15
interesting
Right? Okay, so it doesn’t suffer from like the hallucination problems we see with chapter 18
13:10.81
Jarek
Um, yeah, so ah so that’s a very good question Actually so um, you know I wouldn’t call answering artificial intelligence. But as I said we use answer to generate data that we will use the data to train.
13:26.84
Jarek
What people nowadays do call artificial intelligence models. Um now I’m not a huge fan of the whole artificial intelligence term for the reason that you know we struggled really to define What normal intelligence is so so you know it’s very difficult to talk about artificial one.
13:39.82
interesting
Um.
13:43.26
Jarek
But if we sort of say that artificial intelligence now is what people consider chat gdp and you know large language models right? which is sort of a good proxy for what people understand as artificial intelligence right now and then yes we do want to train the same types of sort of neural network-based models. For engineering um and in a way I can give you an example. Um the way it works is let’s say that you have a certain type of an electric motor. There are different families of electric motors. But let’s say you focus on one of them. Um. And that Motoro can have many different variations to it right? and all those variations have different tradeoffs between I don’t know efficiency and the maximum torque it can generate and you know cost and all of those things now if you explore and you simulate enough of different variations of those designs.
14:27.97
interesting
I have.
14:41.48
Jarek
Then you can train a neural network-based model that you know some people might already try to call machine learning Ai -based model but I wouldn’t go that far but you can train a neural network-based model that then you can say hey machine learning Model. Could you generate me a design of a motor that will fulfill the requirement of torque and cost that you give it and the model will give you a proposed design that the model believes fulfills the torque and the cost requirements that you gave it. It’s a generative model that you ask. For certain properties and it will generate you a design in a very similar way to let’s say dalu works or you know other models work where you say I would like a picture of a cat sitting on a palm tree and the model goes and generates you a picture of a cat sitting on a palm tree that’s because it’s seen.
15:30.90
interesting
Yes. Yes I.
15:39.30
Jarek
A lot of pictures of cats and a lot of pictures of palm trees and it understand that. Ah the sort of structure of the language enough that it can then generate you the right image similarly the engineering models they will see a lot of different designs of electric motors. They will see a lot of different properties of the electric motors like torque and and efficiency and then based on that data if you ask for a certain property. It will do its best to generate you a design that fulfills those requirements now. Obviously if you want to do that Across. All the families of electric motors like across all the properties of electric motors then you need a lot of data right? The same way that to train a dali or or charge Gdp. You need a lot of data. That’s why data is so critical and that’s why we started with answer being simulation optimization. Engine.
16:19.23
interesting
Ah.
16:30.74
Jarek
Because it’s the way for us to generate data that we need for machine learning unfortunately or maybe fortunately for monymo there is no internet of electric motor designs. The reason why large language models really started with.
16:40.62
interesting
Like.
16:47.69
Jarek
Text and then later images is because you know internet is basically full of text and images. It’s the easiest data source that we had at different the easiest ah sort of ah diverse data source right? because there is a lot of different images and a lot of different text available. So. Was enough data to train those large models. There is no huge database of all the different weird motor designs right? Unfortunately or maybe fortunately and so we need to generate that data and once we have that data then we can indeed try to train those generative models.
17:12.10
interesting
Yeah, yeah, yeah.
17:20.81
interesting
So are you are you sort of reverse engineering existing motor designs then within answer to give you the data.
17:27.90
Jarek
Um, ah and not necessarily reverse engineering you know as I said there are ah certain types of motors that are well-known and ah people use those types of models for different applications.
17:36.82
interesting
Right.
17:43.31
Jarek
Ah, those motors usually have ah certain parameters that designers of Motors change and those are well-documented so we do all of that plus on top of that we add more parameterization and more freedom in the design because if we would constrain our data generation to.
17:45.40
interesting
Right.
18:02.25
Jarek
Only few parameters of the motor then our final large engineering model would only learn to change this one parameter and therefore it would not generate any actual innovation right? because you know if it’s only 1 or 2 parameters that you can change then anybody can change those 2 parameters right.
18:08.57
interesting
Yes.
18:20.98
Jarek
In order to really be innovative that model needs to be able to generate designs in like you know hundreds or thousands of parameters because only then you can actually propose a design that will be innovative and different than what we see right now and to do that. Our answer engine needs to be flexible enough to simulate and generate data for all of those designs that are again quote unquote a bit weird right? There are not standard types of motors because if we would just train on standard types of motors and standard designs of motors. We would never discover anything outside of those standard designs of motors. So answer engine is flexible enough that it can explore motor designs that are outside of the standard templated versions of motors that people use because we need that very ah.
19:00.18
interesting
10
19:14.43
interesting
Touch. Yes.
19:17.97
Jarek
We need that data. We need that strange again quote Unquote strange data to train models that will be flexible enough to propose innovation.
19:25.89
interesting
So in theory could throw off dead end designs or impractical designs kind of playing around.
19:31.42
Jarek
Oh yeah, so obviously so the simulation and the answer simulation engine. It’s a multiphysics engine right? So we simulate electromagnetism. We also simulate mechanical aspects of it. We simulate a lot of different things. So if for example, a given motor design would fall apart.
19:39.47
interesting
So and.
19:51.26
Jarek
The mechanical check will flag it and we will know that this is not a valid design now. Another very important thing about um what we do is that um we incorporate Expertise and expert knowledge of motor designers into what we do because there is you know physics.
19:54.29
interesting
Right.
20:10.87
Jarek
And and mathematical solutions is 1 thing but there is another aspect of of engineering design which is as you mentioned what is manufactable and what is actually possible. You know, practically not theoretically but practically and obviously what we want is the outputs of our you know large engineering models on.
20:18.69
interesting
So yes.
20:30.77
Jarek
In the short term of of answer to be practical designs right? because we obviously want to you know, make them and actually make sure that those are actually you know incorporated into an actual product so we need to incorporate this more sort Of. Esoteric Knowledge of what is actually practical and to do that You need to talk to people who know how to build electric Motors So We at monument we actually have a team of experts of people who are actual motor designers. They know how to build motors and we make sure that what we do. Inside of answer engine and then later in the large engineering models does capture as much as possible of what is practically possible, right? Not only theoretically possible.
21:16.54
interesting
Okay, that makes sense and so to analyze the practicality of of a design and what was the output from answer is it a series of equations or an actual drawing like a schematic.
21:31.30
Jarek
No, ah, it’s an actual yeah ah it’s ah it’s an actual drawing. It’s ah, an actual design of of a motor. It’s not like a complete blueprint that you can just go into manufacturing and make right.
21:43.54
interesting
That’s.
21:48.91
Jarek
There is um, usually there are like you know, multiple stages of a motor design. You sort of first design the like active parts of your motor the electromagnetism and sort of mechanical parts of it and then ah based on that design you figure out. Okay now how do I actually go and make a physical motor out of it right. So what we produce is this active and design of a motorter that can be then taken into manufacturing but it will you know meet some and tweaking also because manufacturing is not necessarily a uniform problem right? different companies have different manufacturing. Approaches. They have different manufacturing you know capabilities and production lines and what is possible for example, for 1 company might not necessarily be possible for the other company or the cost tradeoffs between different manufacturing techniques are different for different companies. So in a way if you really want to deliver a design that is tailored for a specific customer. You need to work close with that customer and sort of incorporate as much of their knowledge of their manufacturing process into the design of the motor.
22:57.60
interesting
Kate can.
22:59.47
Jarek
And we do that We very we very happy work with partners right? to to make sure that the designs we produce will fit what what they need.
23:06.44
interesting
So with that with answer and when you’re trying to design a motor with your customer and how does it work So you’ll tell it like you said oh we need this kind of motor for this job here are the constraints and.
23:18.46
Jarek
Oh.
23:24.76
interesting
But that include the customers manufacturing abilities am I making any sense.
23:29.84
Jarek
Ah, no, no, no you you you totally are right? Ah yes, so obviously it depends a little bit on on the circumstance because some of the manufacturing capabilities are a little bit of a secret source for customers and they don’t necessarily want to share it and so basically the things that they can share.
23:39.21
interesting
Of course.
23:47.41
Jarek
They share for I don’t know to give you an example, they might say that the minimum thickness of a lamination cannot be more than you know? ah sorry the minimum that cannot be less than I know half a millimeter because their stamping tool is like that.
23:59.79
interesting
Yes, okay.
24:02.55
Jarek
Then we can obviously incorporate that into our optimization in answer and make sure that the design that answer produces will not have a thickness of eliminatation that is smaller than that right? That’s easy now for things that they can’t really share with us or they can’t or the the things that can’t easily be captured in a. You know, mathematical way as you wish then usually it’s an iterative process right? So we will sort of work on something generate a design we show that design to them then they look at it and they say okay, no, That’s nice, but could you change this and this and this right and then we can obviously redo that show them the design again. And it’s a bit of an iterative process.
24:40.30
interesting
So you can go to an and say right? This design’s good. Keep these bits change this bit at Re redo it by with midjourney. You can.
24:48.70
Jarek
Ah, yes, yeah, yeah, you you can so you know answer at the moment is mostly ah you know a simulation optimization tool right? So different parameters that you simulate and optimize. You can either freeze them or change them or you can sort of change the range of the parameters that answer explores all of that is possible right? so.
24:55.58
interesting
Fast.
25:07.53
Jarek
If I have a design that is good. But for example, let’s say that the customer says that you know what the length of the magnet is a little bit too much could it could you make sure that it is you know half a millimeter ah shorter. Then obviously you can go back to answer. Put the design that you already have as the initial starting point and say hey could you now try to optimize it again but make sure that the length of the magnet is within the range that that the customer asked for and you just do it Again. So yeah, that’s all possible now longer term when we talk about large engineering models and.
25:36.40
interesting
First.
25:44.84
Jarek
The iterative aspect of it. You know one of the beautiful things about Ch Gdp and similar Lms is exactly that iterative part of it right? You can ask judge gdp something. It will give you an answer and then you can say oh that’s interesting. Could you tell me more about.
25:54.11
interesting
Yeah.
26:02.68
interesting
Yeah.
26:04.50
Jarek
Blah right? and it will from the context it will Learn. It will sort of understand from the context what you’re asking for and it will give you more information about what you ask for and you can keep on going. You can have the conversation. That’s one of the most appealing things about Lms and chat Gdp. So. Ah, for the large engineering models and we will need to do the same thing with one difference is that we don’t really care that much about the language aspect of it. You know Engineers are very used to sort of communicating with and software as you wish with numbers right.
26:26.86
interesting
Yeah.
26:35.69
Jarek
So there will be an interactive process to it but the process will not be based on the text chat prompt but it will be based on sort of providing ah new ranges of numbers and sort of saying okay this you know freeze that those parameters and redesign me.
26:36.25
interesting
Um, of course and.
26:39.74
interesting
Bright.
26:53.17
Jarek
Those parameters and things like that. But it will not be based on the language it will be based on numbers but there will be an interactive part of it.
26:57.20
interesting
I see so you wouldn’t be able to Okay so you wouldn’t be able to commune with it in in the natural language. You wouldn’t really and be able to understand that idm in the eventually.
27:05.96
Jarek
And yes, so yes, in in a way. The the reason for that is that we are not. We’re not going to focus on it because it’s in a way. Not important right? What is important for life engineer engineering models is the actual outputs of it. Not necessarily how you communicate with it now.
27:16.11
interesting
Yeah, okay, of course.
27:24.80
Jarek
The communication part is in a way I’m going to say solved problem right? because the large language models in a way solved this problem for us the way of learning language and then translating that language into some kind of a representation. That can be used for something else later is in a way I mean I don’t want to say a solved problem but it’s advanced enough that if we wanted to add a language processing layer on top of our Llm then I’m pretty sure we could do that it just.
27:57.97
interesting
Yeah.
28:00.15
Jarek
I Don’t think that this is necessarily a complexity that we will want to add at the beginning because it doesn’t bring much value in a way but maybe maybe at some point we will hear.
28:07.37
interesting
Of course yeah it Yeah I’ll be more of a non technical person interface wouldn’t it really? Um, yeah yeah.
28:13.29
Jarek
Exactly exactly? Yes, yes and we might do that at some point right? But as I said initially what we really focus on is that the outputs of those lem models are valuable and interesting designs right? Not necessarily how you ask for them.
28:30.45
interesting
Okay, fine. Um, so could it potentially then also be used for training your engineers how motors work or.
28:41.98
Jarek
Ah, that is a very interesting question. Ah yes I think it could I think that once your um sort of large and ah engineering model understands enough of the sort of intrinsic complexities of engineering and and physics behind it then it could, but. For that you would almost certainly need to have this extra layer of Language. Ah, but as I said the layer of language is probably the easier bit and so I think once you do have a lamb and it understands enough about ah engineering and electric motors.
29:03.28
interesting
Um, of course. Yeah.
29:08.80
interesting
Yeah.
29:16.81
Jarek
Then yes, that’s a very good point you could use it for for training engineers or you know even in schools for or or or yeah, yeah, yeah.
29:21.55
interesting
Yeah, last yeah, very much the kind of the word I’m looking for the spinoff application is’s not it primary role. So yeah.
29:31.55
Jarek
Yeah, yes, and you know in a way I think this is a good point. Um, at the moment we’re focusing a lot on language and images like when people think Ai they think you know language models and and image generation and that’s great I mean.
29:42.19
interesting
Yeah.
29:48.00
Jarek
Advances that has been made there in the past few years are just stunning right? like I I think it’s amazing. What what has been done in in the recent years but if we want to really move towards something that is closer to like general intelligence. We need to go beyond that we need to go beyond just language and just images right because the world is more than just language and images. The world is full of physical things physical devices that work in a certain way. The world is you know, full of math and you know physics and.
30:12.24
interesting
Perhaps.
30:22.43
interesting
Yes.
30:24.58
Jarek
Biology and all of that and if we want to have models that can really understand all of that we need more than just text. Um and people are working on it right? Obviously you know there’s alpha fold that is you know you know dedicated to you know protein folding and protein interactions and you know.
30:31.16
interesting
Of course.
30:44.22
Jarek
People are working on models for you know fluid dynamics and things like that so people are working on different things and we are working on engineering and electric motors and all of those parts are very important because if you really want to have. And model that will understand the world you will need to combine all of those things right? you will need to combine all of those things together and and we’re moving in that direction and you know monument is just doing its part in in engineering and model design. But it’s extremely exciting I really think that in the next five ten years
31:07.94
interesting
Strip.
31:19.99
Jarek
Ah, there will be some really very interesting breaksource in in the machine learning and Ai field.
31:26.16
interesting
Has ants already produced any well if you’d like to tell us I suppose any motors that have gone to production or Canngo production.
31:36.20
Jarek
Ah, ah, that has gone to production and no, no, not yet you know the the production cycle of of motors especially in the ev space is is very long. It’s like you know 5 years from the design to being in production or so so since monument is two and a half years old that would be virtually impossible.
31:45.90
interesting
Okay, yeah.
31:53.97
Jarek
Um, but and we we do have ah pretty advanced conversations with some you know large manufacturers of electric motorters. Um, and ah, you know, hopefully at some point some of the designs that monument is is proposing will end in production and in a way. That’s why I remember at the very beginning I mentioned the video games and how nice it is to see a product that you worked on on a shelf. That’s another thing that I really like about monument is that you know there is a potential to at some point you know, look outside of the window and you see a car passing by and you say oh yeah, that car has a motorter that I worked On. And I think that’s a very nice feeling seeing that you sort of worked on something that is then you know used by other people is is a very nice feeling.
32:33.57
interesting
Here.
32:39.88
interesting
Absolutely is and obviously you’ll be and I’m trying to say you with with like the E M or answer there’s never going to be a pinnacle up the ultimate electrical motor design is there or whatever component you’ve designed because they.
32:56.90
Jarek
Ah.
32:58.70
interesting
They’re different applications presumably. So.
33:00.76
Jarek
Yeah, yes, so there is That’s that’s another very very good Point. So um, there is many things that change so people are for example, developing New So New materials. New manufacturing techniques are being developed. That allow for certain things in the design to be changed so because our sort of understanding and knowledge and um, sort of the general um aspect of what is possible in electric motors is Changing. So is the optimal design right? because obviously if now there is a new type of I don’t know let’s say steel that has different magnetic property properties than half a year ago then the design of the motor will change So Obviously with that new steel you need to now.
33:37.60
interesting
Plan.
33:45.11
interesting
Ah.
33:54.25
Jarek
Redesign your motors to make them optimal again. So you’re completely right? that you know in a way at a given point in time. Yes, there is an optimal design but it’s only at this point in time you know half a year later it will not be optimal anymore and on top of that since we’re working not on just.
34:06.31
interesting
Um, yeah.
34:13.59
Jarek
The motor itself but the whole power train that includes you know the inverter the controller the gearbox potentially all of those things keep on changing and being developed and being improved and if 1 of your components changes if you want to really have a system level optimal design. Then you have to redesign the whole system right? to make sure that your whole system is now optimal when this one component has changed so you know in a way I do not worry that we will just find 1 optimal design and then there will be no work for us I think that it’s in a way. Never-ending story as we progress and our sort of. Ah, advancements in different um parts of the power train or different parts of engineering change. The optimal design will have to be updated all the time.
34:59.41
interesting
Yeah, this yeah sort acts a catalyst and it to speed up the evolution of the technology. They very very interesting and so moving away from motors and I think I saw on the website. There’s talk of heating cooling.
35:04.96
Jarek
Um.
35:17.81
interesting
Improvements with one of those models. So obviously not right now but could it be used for things like nuclear fusion or thinking really long long term ahead.
35:17.84
Jarek
Oh.
35:29.59
Jarek
Yeah, um, so it’s it’s an interesting question. Um I would say that you know conceptually ah what we do right? It’s um. 2 things 1 is the simulation and optimization that is you know, answer at the moment that is used to generate data and then we use that data to train large engineering models that are then good at solving the problem that that data relates to right. So obviously the simulation that you need to do for electric motors is I would say a little bit different than the simulations. You need to do for nuclear fission. However, once you do the simulation and you do have the data. Then this other part of what we do at monument which is using that data to train large engineering models now that applies right? because that is in the way data agnostic right? Ah, this just sort of are models that can take engineering or science type data and train. You know. Models on it. So I would say that our simulation part you know in reality I can’t claim that you know we can use the answer to simulate nuclear fusion right? That’s that that that’s not really possible. Obviously you can.
36:55.92
Jarek
Build the simulator that would be able to simulate nuclear fusion and then use it to generate data then then the other part of what monument is doing which is training the large engineering models on that data could be used but to do that you would need to you know again.
37:06.27
interesting
Yes.
37:12.37
Jarek
Get a team of people who understand nuclear fusion build a simulator for it generate the data, etc etc. So I would say it is possible I’m not sure whether nuclear fusion is something that we would go after electric motors. We probably want to make a little bit of a smaller smaller Step Yes, ah.
37:25.71
interesting
Um, interesting and the hardest thing I can.
37:30.92
Jarek
Plus as you know nuclear fission is always going to be solved in 2 years time is going to be solved right? Um, so I would say that yes we will explore things like that probably more you know heating and cooling and maybe agevac things are more sort of in the proximity of electric motors than nuclear fusion.
37:32.55
interesting
Um, yeah I know.
37:49.15
interesting
Gotcha.
37:50.64
Jarek
But who knows you know I mean never say never right? Um, it might be that Um, by that time we want to do that. There will be a dataset. Maybe you know from other you know companies or research institute that work on nuclear fiion. There will be enough data that we can just get that data and maybe trade the models on that data. So.
37:58.50
interesting
S.
38:09.23
interesting
So so the Ibm would be eventually. What would it be like a general. Yeah I’m trying to say it can solve certain problems like you can do motors and ah generators and something else.
38:09.84
Jarek
We’ll see never say never.
38:29.50
interesting
Or they you have like a specific here’s a motor E M is the generator I M my making sense.
38:33.53
Jarek
Yeah, yeah, yeah, yeah, so that again is ah it’s a fantastic question right? I think that it will be a transition Initially, we’ll probably have sort of smaller lamps that are specialized at different things and but.
38:46.62
interesting
Yes, yes.
38:49.99
Jarek
Then as we get more and more data we will try to combine them together the same way that you know at the moment you have a model that can sort of predict protein protein interactions you know from alpha fold and you have a model that can generate a picture of ah you know in a style of Picasso.
38:51.70
interesting
Right.
39:06.20
interesting
Right.
39:07.99
Jarek
Those are 2 very different models and but in the end at some point we if we want to move towards this more general Intelligence. You know, whatever that really is you will want to have the combination. You have a model that can do both um because in a way the understanding of the whole. World requires you to somehow understand all of this so we will start Small. We’ll start with smaller lambs that can do specific things but then we will try to combine them because there is I believe that there is value in. Ah, model that understands something about you know Plasma physics and very high and um value. You know, magnetic, magnetic fields from Nuclear fusion I think that some of that knowledge could bring some interesting insights into motor design.
40:01.89
interesting
That.
40:03.92
Jarek
We don’t necessarily know what and I don’t know how interesting that would be but you know in the history of you know, human Innovation. It’s often happened that you take you know a nuclear physicist and you take that nuclear physicist hey could you work on protein interactions with us. And then suddenly they break an insight that makes a breakthrough in protein design. It might be a numerical method. It might be you know,? whatever but they just have a different view of the world and that different view allows you to find a new solution to a problem and I think that the same thing will need to happen with machine learning right. Some part of what the model learned from nuclear fusion perhaps will allow it to find a very novel solution to an electric motor. Very yes, Yeah, yeah, yeah, yeah, exactly yes.
40:49.44
interesting
Yeah, that would be exciting very exciting even biology could do bring something interesting those um okay and going into the future of it and with this. The Elium and I trying to say is it is it going to be proprietary for any for Monomo or is it going to be something for general release um to other engineers. What’s the yeah was the idea.
41:15.99
Jarek
Yeah, so and the idea at the moment is to indeed keep the lem and the answer engine proprietor to moneymo and use it to generate designs and then the old designs is what we sort of license to to customers. Um. That’s the ah current current model. Obviously that model might change in 5 years or whatever that depends on on how things go but at the moment. Ah yes, ah, we will keep the answer and the lam models and the data that we have ah proprietary to moneymo.
41:35.54
interesting
Right.
41:45.98
interesting
Got you got you so engineers don’t have to worry about losing their jobs right now. Okay.
41:53.00
Jarek
I I don’t think that Engineers really have to worry about losing their jobs I will always need engineers you know all of this is um, it’s a tool and we’ve seen in the past that each time a new tool is developed in principle.
41:56.82
interesting
Of course.
42:08.66
Jarek
Humankind has an option either I’m going to use this great new tool and do the same thing that I did before but with less humans being involved or I can use this new tool and all the humans that I have to just do more and we always go for Let’s just do more.
42:24.36
interesting
Death.
42:26.41
Jarek
Because that’s in human nature if you can do more you will do more and that’s obviously both human nature and also economy of things because you know if one company doesn’t do as much as they could Then they’re just going to lose in the market. So.
42:38.52
interesting
Exactly.
42:41.51
Jarek
Both the human nature of wanting to do more and the economy will I think in a way assure that this new tool is used to just do more not to do the same as we do now but with less humans being involved. So yeah I think that you know nothing to worry about.
42:59.30
interesting
Fantastic.
43:00.70
Jarek
Be honest I mean some retraining and obviously a little bit of more understanding of how to work with machine learning and with Ai type things would probably be necessary but that’s the same as you know when Cad was introduced right? Engineers had to learn how to use Kat right? I mean that’s that’s fine.
43:11.96
interesting
Game. Yeah brilliant and that’s basically all my questions and thank you for your time and wrap things up and anything else. You would like to add before we wrap things up.
43:28.81
Jarek
Um, no thank you very much for for having me. It’s been ah a real pleasure I think that you know a close thought maybe will be that I really think that we are living in a very interesting times from machine learning and Ai point of view and the next five ten years are going to be really really interesting. And I think that engineering will be a critical part of the ml ai revolution that will happen so really looking forward to it.
43:51.70
interesting
Completely free. Absolutely and yeah, so our audience and I will pretty certainly very interested how this develops and so with that this concludes this episode of lexicon. So thank you all for tuning in and being our guest Today. You can learn more about monomost technology on its website and of course follow our social media channels for the latest in science and technology news goodbye for now and thank you again? Bye Thank you.
44:18.77
Jarek
Bye
