
Large physics models boost the performance of passenger cars
Aerodynamics expert David Wheater is our podcast guest
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Aerodynamics expert David Wheater is our podcast guest
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
- Hello, and welcome to the Physics World weekly - podcast. I'm Hamish Johnston. - My guest is David Wheater, - who is vice president - automotive - at UK and US based PhysicsX. - An aeronautical - engineer by training, Dave spent much of his - career using aerodynamics - to improve the performance - of formula one racing cars for several different - f one constructors.
- Now his focus is on using - AI based large physics models - to design better passenger cars and other vehicles. - This involves combining - computational - fluid dynamics - simulations - with wind tunnel experiments and real world testing. - As Dave explains in this wide ranging conversation, - which also looks at career opportunities - for physicists - in this multidisciplinary
- field. - Hi, Dave. Welcome to the podcast. - Hi, Hamish. Thank you very much for having - me. - So Dave, can we start out? Can you - tell us a bit about yourself? - Who who are you? What's your background? And, - what's your experience in the automotive industry? - Sure. - So I am technical director for automotive at
- PhysX. - That means I I lead automotive activities and - also competition engineering, - based on my background. - So what does that involve? - Working with customers, - on the ground. - You know, we we co engineer - our our solutions with with customers in order - to kind of quickly, - build value. We have - experienced engineering teams, cross functional teams across - engineering, machine learning, data science, and software. And, - you know, living
- side by side with our customers on-site, - you know, we we we really learn about - the real world of our customers and, - and and what we need to do in - the hands of engineers. - So I I work there, and I also - coordinate across the other key pillars of our - company - for automotive. - We have a research group. We'll talk later - about their work on the large physics model. - But, yeah, that's a really world class group, - contributing to academia. For example, a spotlight paper - at NeurIPS, - one of the leading machine learning conferences.
- Also, you know, keeping us at the cutting - edge of research for the benefit of our - customers. - Recent example there, - well, very recent, in fact, Transolver three, which - is a a kind of machine learning architecture - similar to a, - sort of transformer - used in large languages but architected for physics. - That was, - published in academia late February, and we're already - shipping that onto our platform. - And then finally, you know, we've got an - amazing, - AI and and and product
- group who who build a platform that that - everything is based on and that allows us - to deploy our end to end AI workflows - into our customers' - most most difficult physics problems. - And and and there, you know, we need - to make sure that we're integrating with the - way our customers already work. So a lot - of work around, - kind of hook hooks into existing simulation and - CAD tools - and the right level accessibility for engineers. - So, you know, my my work,
- the customers and understanding the customer problems is - is really useful for for working with those - groups. - I should say I've I've been at PhysicsX - for nine months, - and and and it's been, you know, an - amazing experience, - working with that kind of diverse group. - And and prior to PhysicsX, I I I - spent quite a few years, twenty five years - in Formula One, - starting my life as an aerodynamicist - at at Benetton. - And, - by the mid two thousands, I was leading
- the aerodynamics - development - across half the car for Renault f one. - You know, that was still in the days - of fairly small aerodynamics departments, but - we were an innovative - group and and launched - 2005 - with with some pretty novel ideas. - One of which, coincidentally, was developed by our - cofounder, - Robin Talui, which was a a mass damper. - And I worked on a - a a novel rear suspension system, which we - introduced that year to replicate the error effect - of of some of the floor that had
- been removed due to regulation changes. - And and that really kind of cemented, - the championship - win in 2005 - and 2006, - and that was at a time when you - could win championships with with novel ideas. - If I jump forward to my later life - in Formula one, - 2018, I became aerodynamics director at Williams and - then, - aerodynamics technical director at Alpine. - Much larger departments by then, real machines - of of development - where winning takes just relentless pursuit of performance
- over time. - And and that's where I I met PhysicsX, - and and they stepped into the picture with - me as a customer. - We did some exciting work together. I can't - talk about the details, but, yeah, the the - impact - was what caught my attention, - not just that we we could do things - much faster, - but but the real transformation came from what - we could do differently with that speed. - In our case, reacting to results, - over a race weekend in a in a - way that just wasn't possible before.
- And and, David, I I think it's easy - for a physicist to imagine how aerodynamics - can be very important in Formula one. But - what about the design of of of modern - cars, you know, the sort of car that - I would drive to the supermarket? - Can significant improvements in fuel efficiency, handling, and - safety - still be gained by improving aerodynamics? - Yeah. Absolutely. And and and as you say, - Hamish, you know, in in in Formula one,
- aerodynamics and downforce is is is the currency. - You know, 11% - of downforce is a tenth of a second, - and and that can be a a championship - changer for for a team in a tight - championship. - And and I should say in, you know, - in, - Formula One, it's not just about the downforce. - It's also about the drivability. You know, drivers - but they don't drive to the peak. They - drive to the trough if you've got variations - in downforce over over car conditions. - And so - in automotive,
- kind of passenger vehicles, - aerodynamics plays just as critical a role in - in in slightly - different ways, - and gets into just as many systems on - the car. The obvious one being, of course, - the external shape, the focus there more on - drag than than downforce from Formula one. - You know, above 30 miles an hour, - most of most of your energy consumption is - is going into overcoming air resistance. So - good fuel efficiency - absolutely needs great aerodynamics, whether that's kind of - the obvious details that you see on the
- car, the spoilers, etcetera, - or or the lot less obvious like the - tire shapes, which can can be real really - important. - But also in the thermal management, - systems like powertrains, - NVH, which is the noise vibration harshness, and - that's about controlling the turbulence of the flow - around the car, - and, of course, passenger comfort, HVAC, which is - heating, ventilation, and aircon. - Those are really important attributes for for, you - know, OEMs developing their cars, - and and, you know, pitching them,
- at the top of the market. - And so today, how how are vehicle - aerodynamics - simulated? - Is it, is it sort of a I'm - guessing you're you're using some pretty powerful, - computer models, - to simulate how, you know, for example, how - air flows around a car, how it interacts - with the tires, that sort of thing? - Absolutely. - So, - you know,
- traditionally, - over the last few decades, - we've been using these numerical simulations - for for fluid dynamics or computational fluid dynamics. - And and these, you know, these simulations - take - the the the flow that governs their dynamics, - the Navios Stokes equations generally, - and and applies them to - a discretized - car and and environment so that a computer - can solve that numerically. - And and as you suggest, you know, that - that that takes a lot of compute horsepower.
- You know, the the the simulations can run, - if if if you're talking about, unsteady simulations, - they they they can run for days. - And and, you know, to get even to - get to those simulations, you obviously need to - draw a car and that's likely to be - in CAD and take some time. But these - numerical simulations are are pretty fickle. You know, - to get those numerics to work, - you need a perfectly closed volume to to - to solve the calculations, and that actually often - takes multiple days of work to go from
- something that looks, you know, like a kind - of finished car on the screen to something - that you can get through a simulation. And - and and there's a hidden effort there, which - aerodynamic - aerodynamicists - like myself know only too well. - So so when we start talking about these - AI tools, you know, an important part of - speed up is is also in the the - the kind of the preparation - effort that goes into to - a a current simulation. - There there's also the post processing side. So, - you know, once you run that that simulation,
- you'll you'll want to extract the the things - you really care about. It might be the - lift. It might be the downforce. - So so so that you generally set up - some kind of process around that. - And and also maybe to understand the flow - field, you you might take a few slices, - through through the airflow, - which you'd sort of save down, - because that that simulation file is normally - pretty huge, - and impractical to save. - And you said these are big calculations.
- Are can can you do them in house? - Or do you have to rent time on - a supercomputer? - Or do you do a bit of both? - Yeah. I mean, - most, - kind of most automotive players will will will - have their HPC resource. - Oft often that is in house, and, you - know, these are these are big big investments, - but but increasingly, - kind of cloud provision, - and and as so it's moving towards GPUs,
- on on GPU clusters, - are pretty typical. - And what about, - you know, so I I suppose proving - the the the worth of these simulations. - How do you do do you have to - compare - experimental data and, - simulations - in order to to know that you're on - the right track? I mean, do do you - have to actually build a model - of the car and and stick it in - a wind tunnel? I'm guessing that's probably what
- you're trying to avoid because that would be - very expensive. So how how do you confirm - that your that your simulations - are working? - Yeah. - Absolutely, - experiments - are an important part of the picture. And, - if I kind of think about my formula - one days, - then, you know, we we we were using - the wind tunnel - very heavily to kind of correlate and tune
- our our CFD. In the end, you know, - the CFD is a is a model of - a a model of your reality. - And and so, you know, having that experimental - data, - albeit it's, you know, it's it's time consuming - and it's and it's expensive - is is an important part of validation - for sure. - I see. But, - and and and so now I mean I - mean, maybe this is a good time to - to start talking about - AI. - So, David, before the advent of AI, how - were experimental
- and simulated - aerodynamics - data used to design cars? - Great question. Thanks, Hamish. - So during my f one experience, although I - think the answer is similar for for sort - of passenger cars as well. - You know, when I started at Benetton in - 2000, it was really all about wind tunnel - testing, - and just doing that as much as possible - and as fast as possible. - But it was expensive. And, you know, even - with, you know, f one speed, it was
- relatively slow weeks between loops - and limited in terms of the feedback that - you could get, - from those tests. You you'd get force from - the balance and maybe a few pressure sensors. - Then on the CFD side, we had, at - that stage, just one engineer, and their entire - focus actually was on the the airbox, which - is is the sort of feed behind the - driver's head going to the engine. - Otherwise, it was all wind tunnel. And there - was a lot of skepticism at the time - about that simulation with engineers
- saying they didn't believe the CFD and and - not using it. - And then through the two thousands, - you know, we we saw that the the - teams that really thrived were the ones that - saw CFD as something complimentary. - You know, it was much, much faster than - than wind tunnel testing. It provided flow field - insight that that you couldn't get from the - wind tunnel. - So with results in a day rather than - weeks, that massively, you know, accelerated how you - could develop the car, and that insight that - you got really helped inform next steps. So
- then if if we roll through to sort - of 2020, - then - your CFD was absolutely a given. - And and and, actually, - the the majority of team principals in f - one voted in an f one meeting - to turn off wind tunnels completely for for - cost saving. But, actually, I think that that - kinda misses the point, to one of your - previous questions. - You know, the these these tools are are - best when you use them together. And I - think we're living through the same moment with
- with AI. You know, there's absolutely, - some skepticism, - and and and some some engineers see this - as a swap in for CFD. But I - think the most successful users I see are - are those that are are using AI as - another layer, - giving that massive acceleration that it offers, but - working alongside the the tools and expertise already - in place. - I see. And and may I suppose moving - on to AI, PhysicsX develops
- large physics models for aerodynamics - and other applications. - What is a a large physics model? And - how does it differ from - conventional - computational approaches to aerodynamics? - Yeah. So so if I I kind of - start with the differences. - So - machine learning physics model's different in a in - a couple of different ways. - Unlike traditional computational - approaches, - machine learning model, - it it doesn't need to be told explicitly
- what the laws of physics - are. - It learns it learns that from the data - it sees and and, you know, we we - use architectures that are suited to specific physics - problems. - And and and in doing that, it becomes - much more flexible than a traditional CFD tool. - It it doesn't require the same geometry preparation - we talked about earlier, - and and it can learn not just from - simulation data, but also from experimental data. So - it can actually bridge the gap. - It it can bridge the - gap, from from simulation world to kind of
- ground truth. - It it it's got some obvious benefits in - terms of speed. Obviously, that that that's kind - of well known, and and and those speeds - can be up to, you know, 10,000 or - more times. So so really game changing. - And and that allows you to, as I - mentioned - in the example before, that really transforms what - you can do, - particularly when you get into solving multi physics - systems. So, - in in a traditional, - maybe in an OEM, you you you might - be solving problems
- that that span across aerodynamics. - You've made it be in an engine design - where you've got aerodynamics involved, but you've also - got structures involved. - And and solving those different physics goes across - many groups and many groups of experts. - But with with these models, - then then you can solve those physics all - at the same time. And because the models - are more accessible than, - than than sort of simulation - tools, - then you can put them direct in the - hands of a designer.
- And and that really starts to, again, change - the way you're working rather than just doing - things a bit faster. You you really start - to rethink how you're working. - And and then finally, another aspect of these - models is, - that that they've they've got this concept of - uncertainty - quantification. - So they they know how confident they are, - and - and and that's not just important from an - engineer's point of view because we need to - be building trust - in in, - in in the kind of use of of - the results.
- But also from from the model point of - view, we can use that uncertainty to drive - the the simulations that the model needs to - be accurate in the places that it needs - to be. So we can have these, what - we call, active learning loops, - that that automatically trigger new set new simulations - when when a model becomes unsure. - And David, you, - and your colleagues at PhysicsX - have released a study called scaling physics AI - for automotive
- aerodynamics. - And it argues that publicly available datasets - of automotive - simulations - are not sufficient to train - large physics models. - Why is this? And what are the consequences - for those using - AI to design cars? - Yeah. So so I should maybe give one - last piece of detail about the the these - types of models and and just to sort - of transition to the automotive large physics model - that we've,
- developed at PhysicsX because I think that gives - some context. - And, you know, this is a pretrained model, - so it's already seen a huge amount of - data. And and that means it's something that - we can take to a customer that's that's - sort of ready to go out of the - box and and can predict aerodynamics - of previously unseen vehicles. - And and and a a really powerful aspect - of that is when we start fine tuning - that model. So we take the model that - we've developed in the house. We we fine - tune it for a given customer's needs or,
- specifics of their car type and simulation approach. - But underpinning all of that is is is - the data, the high quality data that we - need to build that model. And and that - really comes to your question - about - the the need for, - the the the need for a dataset, - and - our, - our own our own dataset. So there's a - there's a couple of well publicized,
- publicly available datasets, driver net and luminary shift - SUV. - And those were, - both derived from sort of parametric - morphing - of of a small number of base configurations. - And that that morphing is is a kind - of where you take a geometric mesh and - you kind of push and pull, - to to give geometric variation. - And, you know, these, - the these these variations, - obviously give give some diversity, but what we,
- you know, what we saw when we we - started digging into these models - was that they performed well on on when - we tested them on geometries, which were very - similar to those, - datasets. - But but really, they started to fail quite - badly when we looked at realistic - customer geometries, which which were of a different - design. - So to to address this shortcoming that that - you've identified in your study, - PhysicsX has created the p x NETCAR
- proprietary - dataset. - How did you develop this, and how does - it compare to public datasets? - Yeah. So, - we we kind of having having drawn that - conclusion that that, you know, we we needed - this - more diversity, so we built, - what we call, - the the Physics x Data Factory, and that's - really an automated pipeline - that can run many, many, - simulations. - In this case, STAR CCM plus runs,
- Reynolds' average average Navios Stokes simulations. - So, you know, we we got a hold - of 450 - quite distinct car shapes, - from large scale three d model libraries. - And then we - produced our own kinda algorithmic - morphing approach so that we could create - around a 100 variants for for each of - these baselines using a a tool called Blender, - which is an open source three d modeling - tool. - And and we used a Latin hypertube
- sampling approach so that we knew we were - kind of creating a a statistically, - good coverage of of these geometries. - And then we we rigorously validated, - those results in terms of the kind of - convergence and the correctness of the physics knowing - that, you know, - the the the quality of that data was - critical to the performance of the model at - the end. - So that that allowed us to build PXnetcar, - which now comprises - over 30,000 simulations
- across those 450 - distinct, - geometries. And, you know, - comparing back to the driver net and luminary - shift, which is where we started with, you - know, between - those two, that they they gave us 10, - kind of 10 distinct geometry start points. So - it's it's really a complete game changer - scaling up that much. - And and, you know, we wanted to be - quite quantitative - about the, the diversity we were generating. So - we use something called pairwise chamfer distance,
- which basically measures between different geometries - the the the kind of distance between all - the points, and and it gives a a - way of quantifying - that diversity. And we could see as we - build up that dataset that we were, - that we were really building a a very - diverse, - set of classes, not just sort of minor - morphs. - And then when we started applying that, - a large physics model in in practice, then - that very clearly translated into much greater generalization - performance. So, you you know, when we started
- showing it new geometries, new cars, - that then the performance was much much higher - than than than the models that we'd, - trained on those narrower, luminary, - and, and and driver net datasets. - And and we've been talking about cars, David, - but PhysicsX has a range of interests outside - of vehicle design. Can you talk about some - of the other sectors that you're developing - AI tools for? - Yeah. Absolutely. Yeah.
- I mean, we work on some incredibly exciting - projects. - And, you know, many of these we can't - talk about because they're they're customer projects. - I happen to be working on one which - is, has been publicly announced, so that that's - quite helpful. - But starting more generally, you know, our our - models add most value in in in sectors - where - complex physics simulation - is right at the heart of engineering decision - making or operation. And, generally, you know, the - time that we spoke about earlier to solve - those simulations - can be a real bottleneck for for engineers
- or product development. - Commonly, we see this across aerospace and semiconductors, - energy, and materials. So, you know, this is - where we we're doing a lot of work. - And and, also, it's not just about design, - but because these models can solve more or - less real time, they become ideal for manufacturing - and and operation processes. - You - know, just to give a small sense of - the complexity, a recent project - we developed was a a plasma reactor for - for converting industrial waste gases into synthetic fuels.
- So basically taking, gases including c o two - and then converting that into carbon monoxide, - which which then went on into these, - synthetic hydrocarbons. - And, you know, we had to model all - the chemical reactions, the mixing of flow through - the reactor, - and and the thermal modeling, of course, to - to make sure that kind of critical components - didn't overheat. - And with with this technology, we can do - all of that simultaneously. - So it's - massive, massive speed up.
- And, you know, we we we reduced the - r and d process by a couple of - years, and and in the end, had a - machine that was giving a tenfold increase in - react productivity, which is, you know, an absolute - game changer. - Moving moving on to the project that I'm - involved in, - so so this is working with GB one, - which is The UK's entrant to the twenty - seven America's Cup. - And it's I think it's an interesting example - because, again, it's it's a very advanced engineering
- workflow. And and and for some context for - those maybe not aware, - America's cup boats are 75 foot hydrofoiling - sailing yachts, - often better described as flying machines. - You know, they they they go pretty crazy - speeds in excess of four and a half - times wind speed, which as an engineer takes - a bit of wrapping your head around, - which means they go they go above a - 100 kilometers an hour. - And I was lucky to join the team, - for some testing in Sardinia a couple of - weeks ago. And and then you really do
- get a sense of of the speed when - this boat is is moving towards you at - that that kind of speed. - And and, you know, like in f one, - it's very much a competition between engine engineering - departments. - So they've got very, very refined development processes. - But one of the big remaining challenges is - just a massive dimensionality, - of of the design and operating space that - that the engineers have to contend with. And - this is where boats are different from f - one cars. - You know, f one cars, you've got the
- steering and and accelerator brake pedal as input. - But with the boats, you've got multiple controls - across the foils, - the sails, - you know, the the sail shape itself, you've - got many controls, plus you've got the the - wind effect. - So not only are there all these design - parameters, but there's also all these operational ones - as well. And and that's kind of where - we've found, - you know, a a a a really good - fit between GP one and PhysicsX because - our, - kind of
- physics aware architecture - can learn all those complex interactions, aerodynamics, - hydrodynamics, - and and predict near real time - both the flow around the car, but also - the, - the the the loads the loads on the - sorry. Around the boat, I should say. - And also the the the the loads that - that then feed into GB one simulation so - we can get a a a kind of - performance output. - And and and also in, you know, in - the structural domain as well. So
- we we can give a really kinda complete - optimization. - So, you know, that that's, - using our platform, taking the learning from the - research group, the, you know, the the the - LPM - development that we spoke about, we can directly - translate very quickly into something that's been useful - for the GB one team. - And, - you know, in in in doing that, we - we're building very much on their expertise and - their physical insight. - But with with such a vast,
- you know, vast space to explore, - the these tools are a complete game changer. - And, - you know, - one one of one of the points we - touched on earlier was, you know, bound the - the confidence of these models. And when we - want to start exploring such, - fast spaces, - we can, again, we can use that uncertainty - to kind of drive where we collect data. - So we we we're starting relatively small with - the dataset, - and and then as we start optimizing and
- the optimizer is exploring the design space, we - can collect more and more data. - So, - you know, what we're seeing is is all - of the all of the, - lessons that we've learned in the, - large physics model are are giving us this - leg up with with our GPU one work, - and that's that's really exciting. - When when you're doing the, - the the work, you know, with the with - the boat, - are you are you just looking at - design, - or are you also
- looking at - the the way in which the boat can - be sailed? - You know, looking for, - I don't know, maybe different ways of doing - things or the best way of of sailing - the boat in addition - to, - I suppose, adjust, - for lack of a better term, the the - fluid dynamics - of the boat? Or is that is is - that something different or maybe something that you - could look at in the future? - Well, yeah. Great question. And and maybe there's
- a bit of both there because, - you know, even at the design stage, if - we wanna optimize the design, we need to - consider how the boat's gonna be sailed. - You know, we we can't do it in - isolation if we want to really kind of - get to a optimum performance. So, you know, - we are considering, - many aspects of how the boat will sail, - you know, as we're as we're going through - that design iteration. - But to your, you know, to your point, - we we can also use these types of - models in operations. So we're we're we're absolutely
- thinking about, you know, how we might be - able to, kind of take the next step - and, - deploy these models, whether that's sort of, - give it giving kind of an almost advice - to the sailors on on on optimum, - sailing strategies or or or in in in - control. - And finally, here at Physics World, we're always - really interested in exploring - careers - for physicists. - And, I'm sure that some of our listeners
- would be interested in a career in the - commercial development of large physics models. - What sort of employment - prospects are there in the sector? - And could could you maybe give a a - bit of career advice for someone who who'd - like to to get a job, - developing these models? - Yeah. Absolutely. And, you know, my own journey - to physics x is is is maybe not - a typical one. - But but maybe maybe there's a little bit,
- you know, in in that journey that I - I still sort of take away. - You know, - when when I started out in f one, - it was definitely not a well trodden path. - And, you know, I was writing a lot - of letters, and I was looking for any - break I could get. - And and and, actually, my break came from - a rejection from what was then British American - Racing. You know, they wanted to employ some - of the experience, but they, - they they were on a notice period. So - there was, there there was a a a - few months where there was a vacant seat. - And so I - said, okay. Can I can I just come
- and, you know, sit there and and learn - for a bit? And and and, actually, I - had the chance while I was there to - go and do some, testing in Italy. - We we were, - basically kind of putting wires across the the - race car to understand deflection effects. And then - I learned a huge amount. It was a - a great education. And, you know, it was - a a lesson definitely for me in, you - know, looking for those opportunities even when they - come in in unexpected ways. - And, you know, that - gave me my career in f one and
- and then the the opportunity, - and the connection with PhysicsX also came with - some serendipity. - You know, it wasn't an opportunity I anticipated, - but, you know, after my experience as a - customer, not just the the impact - that I saw in f one, but the - scalability of this of a wider industry. - You know, I I knew it would be - an amazing place to be. And so, again, - kind of looking for that opportunity - to to kind of make the connections. - And now, you know, now I am at - PhysicsX, and I'm I'm seeing, you know, the
- work that we're doing on on large physics - models. - You know, it's incredibly multidisciplinary. - Obviously, - it needs domain expertise. - Also, you know, we we need simulation engineering - to generate and validate our our datasets. We - talked about how important they are. And the - machine learning research spanning areas like geometric learning, - physics informed methods, - and and a lot of statistics and uncertainty - quantification. So all, you know, really important ingredients. - And, - then familiarity
- with - pretrained models is increasingly relevant, and that could - come from language or image or video. So, - you know, many of these - same ideas and architectures, - kind of transfer into this domain and and - and and so those experiences - are all relevant. - I I guess what I've seen is, - it's not just about - one one skill that sort of makes the - difference for us. It is the collaboration across - physicists and simulation engineers and machine learning researchers,
- infrastructure engineers, - you know. And and - from my own experience, often, you - know, there there's a different language there eve - even for the for the same problem that - you're trying to solve. And so bridging the - gaps between all these different experts is where - a lot of the value is created. - And so, you know, I think, of course, - being an expert in one area is is - absolutely the foundation of your credibility, and there - are, you know, the areas that I've talked - about there. But, also, I would advise you - know, - try and develop that familiarity
- and understanding across the full scope - because the people that can kind of really - thrive are those that can have a meaningful - conversation - both with a a maybe a CFD engineering - engineer in the morning or or a system - architect in the afternoon and and and and - join those dots up. - And - one of one of the things I've loved - about being at PhysicsX is we we hot - desk here. And, you know, every day I - find myself sitting someone, - sitting next to someone with a really deep - expertise - in in something fascinating that I've learned from.
- So I suppose my overall advice there is, - you know, be curious and be open minded. - Yeah. You never know where where the next - opportunity will come from or the next idea. - And and and there's a good chance the - next best idea for you will come from - someone else. So, yeah, stay receptive to people - around you and, - you know, always ask for feedback, - because it's it's really the best gift that - you can get. - So I think that would be that would - be my advice. - Well, it sounds I mean, it sounds like
- a fascinating field - that you're in. - And, you know, especially for, - you know, for for for a physicist or - a mathematician - or, I suppose, even with you, a a - chemist, - that, you you know, there that there are - opportunities. - You know, I suppose AI gets a a - lot of bad press these days, but, - it sounds like it's really opening up, - a a a whole new world when it - comes to, - to design. - Yeah. You you absolutely. And and and, you
- know, I think, - the the AI in the physical space is, - you know, is is quite a different beast, - and it's it's really accelerating. - And as you mentioned, you know, we we - we - have experts in in chemistry as well as - physics, - and, - you know, bringing bringing bringing those experts together - as well as, of course, the the kind - of fundamental - AI behind the under the hood, you know, - that's kind of where the magic happens. - Well, that's great. Thanks, David. Thanks so much
- for, for joining us today and, - talking about your career and your work at, - PhysicsX. It's been a pleasure. Thanks very much, - Hamish. - I'm afraid that's all the time we have - for this week's podcast. - Thanks to Dave Wheater of Physics x for - an enlightening discussion - of how large physics models - are making cars - more efficient.
- I'm Hamish Johnston, and our producer is Fred - Isles. - The music for this episode is called one - three seven, and it was composed and performed - by the physicist - Philip Moriarty. - We'll be back again next week.
Nobody is credited on this episode yet.
Michelle Simmons of Silicon Quantum Computing is our podcast guest
PicoQuant’s Emilio Gutierrez-Partida is our podcast guest
Our podcast guest is founding director of the IBS Center for Quantum Nanoscience in Seoul
Marge Ryan of Johnson Matthey is our podcast guest
Quantum experiment is controlled from thousands of kilometres away