1 00:00:00,280 --> 00:00:02,400 Hi, and welcome to the Neil Ashton Podcast. 2 00:00:03,080 --> 00:00:05,920 In each episode, we explain some of the fascinating ways 3 00:00:05,920 --> 00:00:09,080 that science and engineering are changing the world around us. 4 00:00:09,800 --> 00:00:12,760 We talk to leading engineers from elite level sports like 5 00:00:12,840 --> 00:00:16,880 cycling and Formula One to some of the world's top academics to 6 00:00:16,880 --> 00:00:20,450 understand how fluid dynamics, machine learning and 7 00:00:20,450 --> 00:00:23,000 supercomputing are bringing in a new era of discovery. 8 00:00:23,960 --> 00:00:27,080 We also hear some of their life stories, their career advice, 9 00:00:27,640 --> 00:00:30,080 the lessons they've learned on the way that I hope will be 10 00:00:30,080 --> 00:00:33,800 helpful to you too. So sit back and enjoy this 11 00:00:33,800 --> 00:00:41,000 episode. Hi, and welcome back to the Neil 12 00:00:41,000 --> 00:00:45,160 Ashton Podcast. So today I wanted to give 5 tips 13 00:00:45,320 --> 00:00:49,320 for CAE engineers in the era of AI. 14 00:00:50,120 --> 00:00:54,640 Five things that I think will be useful for you from a career 15 00:00:54,640 --> 00:00:58,185 point of view and from hopefully making the most of what I 16 00:00:58,185 --> 00:01:01,240 think is quite an exciting new technology. 17 00:01:02,120 --> 00:01:05,640 You may disagree with some of these things, and if you do or 18 00:01:05,640 --> 00:01:10,240 have other tips that you think people should should adopt, then 19 00:01:10,520 --> 00:01:13,560 you know, feel free to leave a comment if it's on YouTube or 20 00:01:13,560 --> 00:01:15,920 LinkedIn or wherever you're listening to this. 21 00:01:16,600 --> 00:01:21,880 OK, so let's get started. First of all, I'd say an an open 22 00:01:21,880 --> 00:01:26,800 mind would be tip one. It's very easy to come from a 23 00:01:27,040 --> 00:01:31,040 negative judgmental viewpoint when it comes to AI and it's 24 00:01:31,040 --> 00:01:34,080 natural. We as humans often, you know, 25 00:01:34,480 --> 00:01:38,440 see things that are pushed at us and sometimes we have a 26 00:01:38,440 --> 00:01:42,200 temptation to, you know, be skeptical of some of this. 27 00:01:42,800 --> 00:01:47,785 But I think that is probably the the worst attitude you can 28 00:01:47,785 --> 00:01:50,648 have. I think having an open mind and 29 00:01:50,648 --> 00:01:55,404 being willing to look and listen and read is important. 30 00:01:55,412 --> 00:01:59,308 At the end you may still make the same judgement as you did at 31 00:01:59,308 --> 00:02:02,502 the beginning. But I think most people who 32 00:02:02,510 --> 00:02:07,520 actually look into this do end up changing their mind and and 33 00:02:07,528 --> 00:02:11,979 they understand better where some of maybe the marketing or 34 00:02:11,987 --> 00:02:16,576 hype is over egged and but where there are actual benefits. 35 00:02:16,584 --> 00:02:20,582 It is very tempting, of course, if you have 30 years experience, 36 00:02:20,582 --> 00:02:26,078 you know, like some people have in in CFD or FEA, to almost 37 00:02:26,078 --> 00:02:31,464 be insulted by some young, you know, 20 year olds doing ML 38 00:02:31,472 --> 00:02:34,326 research and suggesting that their method, you know, can be 39 00:02:34,334 --> 00:02:39,216 better. But sometimes that's from a good 40 00:02:39,216 --> 00:02:43,222 intent place, they're excited and maybe they don't know or 41 00:02:43,230 --> 00:02:47,112 they're not from the CAE domain. And, and so having an open mind, 42 00:02:47,112 --> 00:02:51,464 and this is true on sort of both sides on the CAE side and 43 00:02:51,464 --> 00:02:56,728 the ML community is important. The second one, which you sort 44 00:02:56,736 --> 00:03:01,520 of need the first one to get to the second one is educating 45 00:03:01,528 --> 00:03:05,256 yourself. You know, a lot of companies 46 00:03:05,264 --> 00:03:06,760 call this continual professional development. 47 00:03:06,768 --> 00:03:12,022 And I think this is very important in the era of AI. 48 00:03:12,030 --> 00:03:18,020 The problem and I which I think creates sometimes this push back 49 00:03:18,020 --> 00:03:23,055 is because the, the ML community and the CAE community 50 00:03:23,055 --> 00:03:24,520 come from very different backgrounds. 51 00:03:24,520 --> 00:03:27,360 They often have studied different courses at university. 52 00:03:27,360 --> 00:03:30,600 They, they have a different preference in terms of software, 53 00:03:30,600 --> 00:03:33,280 in terms of programming. They, they're, they're two 54 00:03:33,280 --> 00:03:38,040 different communities and the challenge for CAE to look at AI 55 00:03:38,040 --> 00:03:41,120 is that it all seems quite foreign and complex. 56 00:03:42,080 --> 00:03:45,280 And This is why I think educating yourself is important. 57 00:03:45,640 --> 00:03:49,120 Going straight into a journal paper describing an ML 58 00:03:49,120 --> 00:03:51,280 architecture may be a bit overwhelming. 59 00:03:51,280 --> 00:03:54,040 It certainly was for me at the beginning, but there's a lot of 60 00:03:54,040 --> 00:03:58,080 content now that helps you. So Coursera is one of those 61 00:03:58,080 --> 00:04:02,520 platforms where they have some great courses by lots and lots 62 00:04:02,520 --> 00:04:05,480 of people from different, you know, walks of life and 63 00:04:05,480 --> 00:04:08,680 different backgrounds that can really help you. 64 00:04:08,680 --> 00:04:11,680 Your company may already have a subscription or if you don't, 65 00:04:11,720 --> 00:04:14,840 you know, like I did it in the past, I think paying it yourself 66 00:04:14,840 --> 00:04:18,680 is a worthwhile investment. But beyond these sort of 67 00:04:18,680 --> 00:04:22,480 certifications like AWS does them, other tech companies do 68 00:04:22,480 --> 00:04:26,120 them. A lot of conferences, especially 69 00:04:26,120 --> 00:04:29,520 with AI topics, have a real push around transparency. 70 00:04:29,880 --> 00:04:32,520 And so I found that many of those conferences actually 71 00:04:32,520 --> 00:04:36,480 published their entire talks online, which is fantastic 72 00:04:36,480 --> 00:04:39,920 because you can actually watch on YouTube or whatever platform 73 00:04:39,920 --> 00:04:43,040 you prefer Many of these talks, you can speed them up. 74 00:04:43,040 --> 00:04:47,640 You know, if you'd want to get through quickly, you can also 75 00:04:47,720 --> 00:04:50,800 use AI. Remember to summarize papers. 76 00:04:50,800 --> 00:04:56,080 This is something I do a lot. So for example, if you have a, 77 00:04:56,320 --> 00:04:59,600 you know, a paper like attention is all you need or on 78 00:04:59,600 --> 00:05:03,480 Transformers or maybe something specifically on CAE related AI. 79 00:05:04,160 --> 00:05:07,120 Now, most of these AI large language models can do a pretty 80 00:05:07,120 --> 00:05:11,520 good job of explaining things. And if you say, if you upload 81 00:05:11,520 --> 00:05:14,200 the paper and say, please summarize the this to a lay 82 00:05:14,200 --> 00:05:19,400 audience or explain to me why how this bit works. 83 00:05:19,400 --> 00:05:22,480 Or can you can you give me more examples of this? 84 00:05:22,720 --> 00:05:28,000 It's amazing how these AI can actually help teach you to learn 85 00:05:28,480 --> 00:05:33,120 AI, but you can't get away from sometimes speaking to people. 86 00:05:33,120 --> 00:05:36,480 And that's one reason I would definitely try and broaden 87 00:05:36,480 --> 00:05:40,656 yourself and go maybe to even AI-specific conferences like 88 00:05:40,656 --> 00:05:44,280 NeurIPS and others. Come and go to some of the 89 00:05:44,280 --> 00:05:46,560 workshops, listen in. You might not understand 90 00:05:46,560 --> 00:05:50,400 everything, but you'll slowly get and build the network. 91 00:05:50,400 --> 00:05:52,800 You'll start to speak to people one-on-one and ask stupid 92 00:05:52,800 --> 00:05:55,080 questions. People who know me, I often I'm 93 00:05:55,080 --> 00:05:57,680 in meetings and I'll say, I'm sorry to ask a stupid question, 94 00:05:57,680 --> 00:06:00,760 but and often times it maybe not. 95 00:06:01,040 --> 00:06:02,400 It's a completely stupid question. 96 00:06:02,400 --> 00:06:06,080 And regardless, they usually explain something in a easy to 97 00:06:06,080 --> 00:06:09,120 understand manner that it would be difficult for me to get. 98 00:06:09,520 --> 00:06:12,720 And sometimes all you needed is a one or two key concepts. 99 00:06:12,720 --> 00:06:15,720 And once you get the concept, you're like, OK, I see it now 100 00:06:16,080 --> 00:06:20,120 where at the beginning it it, yeah, seems a bit foreign. 101 00:06:20,600 --> 00:06:23,670 And the final way is podcasts, obviously I'm biased, I make my 102 00:06:23,670 --> 00:06:25,560 own, but there's a lot of people creating them now, 103 00:06:25,560 --> 00:06:29,600 interviewing people and those again could be a a great thing. 104 00:06:30,040 --> 00:06:32,400 Finally, of course, depending on, you know, where you are in 105 00:06:32,400 --> 00:06:35,040 your career, going back to university or taking the 106 00:06:35,040 --> 00:06:37,840 university course could also be a worthwhile investment, 107 00:06:37,840 --> 00:06:40,720 particularly if you're thinking of a, a job change. 108 00:06:40,720 --> 00:06:44,520 You know, if you actually want to go into AI and say the sort 109 00:06:44,520 --> 00:06:48,520 of two roles, you can either be the sort of CAE domain specific 110 00:06:48,520 --> 00:06:51,760 person, or you might want to be more hardcore into the AI 111 00:06:51,760 --> 00:06:54,400 itself. And probably for that you would 112 00:06:55,120 --> 00:06:56,840 benefit from some of these courses. 113 00:06:57,040 --> 00:07:00,720 Although many universities now do offer these online courses as 114 00:07:00,720 --> 00:07:02,160 well, like I think Stanford does them. 115 00:07:02,920 --> 00:07:05,120 So you know, there's a lot of options now. 116 00:07:05,120 --> 00:07:07,720 And I don't think there's any excuse not to learn. 117 00:07:08,160 --> 00:07:10,280 You know, it's not like 40 years ago where you have to go to a 118 00:07:10,280 --> 00:07:12,440 library or read a physical thing. 119 00:07:12,440 --> 00:07:14,880 The Internet gives you so many opportunities now. 120 00:07:16,160 --> 00:07:19,640 So if you have an open mind, which is 1 and you've, you know, 121 00:07:19,680 --> 00:07:22,800 educated yourself, which is number 2, and of course they're 122 00:07:22,800 --> 00:07:28,000 continual processes, then let's talk about the AI physics. 123 00:07:28,240 --> 00:07:32,840 And this is probably the most common use case or one of the 124 00:07:32,840 --> 00:07:36,040 most common use cases for AI in the context of of CAE and 125 00:07:36,040 --> 00:07:39,560 engineering AI physics, you know, referring to the use of AI 126 00:07:39,560 --> 00:07:44,640 to create typically some sort of surrogate model where you take 127 00:07:44,640 --> 00:07:47,480 data to train a model. And then once the model's 128 00:07:47,480 --> 00:07:50,000 trained that inference, you can give it a new condition, a 129 00:07:50,000 --> 00:07:52,120 geometry, a boundary condition, and they'll go and predict it, 130 00:07:52,360 --> 00:07:55,262 typically in close to real time. 131 00:07:55,270 --> 00:07:59,940 There are other use cases people sometimes look at developing 132 00:07:59,948 --> 00:08:03,960 better models, you know, transition model, the turbulence 133 00:08:03,960 --> 00:08:07,080 model using AI. But I'd say probably the 134 00:08:07,088 --> 00:08:10,860 predominant one that you may come across is more of the 135 00:08:10,868 --> 00:08:12,710 surrogate modelling. And yes, some people say, oh, 136 00:08:12,718 --> 00:08:14,974 we've been doing that for decades, we've produced 137 00:08:14,974 --> 00:08:18,980 reduced-order models, but typically those did not have the 138 00:08:18,980 --> 00:08:23,772 flexibility like modern day ones in terms of the non 139 00:08:23,772 --> 00:08:27,480 parametric ability as you can just bring any arbitrary 140 00:08:27,480 --> 00:08:30,520 geometry or flow condition in, if you've trained across it, you 141 00:08:30,520 --> 00:08:33,600 can get the results out. So things are different today 142 00:08:34,120 --> 00:08:36,840 than what they used to be. And that common argument of, 143 00:08:36,840 --> 00:08:39,240 well, we've just been doing ML for 30 years and they've just 144 00:08:39,240 --> 00:08:43,448 changed it to something else is probably going back to #1 and 145 00:08:43,448 --> 00:08:46,320 #2 and open minded educating yourself to realize, no, no, 146 00:08:46,480 --> 00:08:50,240 things have changed. Although of course they're based 147 00:08:50,240 --> 00:08:52,360 on foundations from from earlier. 148 00:08:52,640 --> 00:08:54,760 So how do you prepare yourself for the AI physics? 149 00:08:54,760 --> 00:08:58,549 Before we talk about it in more detail, I said the first one 150 00:08:58,549 --> 00:09:04,480 is, is data Today, the biggest difference between your probably 151 00:09:04,720 --> 00:09:08,920 common use of AI with Google Gemini or ChatGPT or whatever is 152 00:09:08,920 --> 00:09:11,080 they're already trained. You're essentially just doing a 153 00:09:11,080 --> 00:09:13,080 text prompt. You're doing inference on a pre 154 00:09:13,080 --> 00:09:15,480 trained model. The big thing with AI in the 155 00:09:15,480 --> 00:09:19,200 context of CAE and CFD and FEA is you're probably going to have 156 00:09:19,200 --> 00:09:22,880 to train the model yourself. That's most likely at least in 157 00:09:22,880 --> 00:09:27,760 the short term. So you need data and if you have 158 00:09:27,760 --> 00:09:30,720 existing data, it's what formats it in. 159 00:09:30,720 --> 00:09:33,160 Where is it at? Can you get it to the model? 160 00:09:33,160 --> 00:09:35,040 Have you scrutinized it? How? 161 00:09:35,760 --> 00:09:38,400 How do you describe it? The model needs to know what it 162 00:09:38,400 --> 00:09:41,960 was generated with. Not all data is identical, and 163 00:09:41,960 --> 00:09:43,320 that is often the biggest challenge. 164 00:09:43,840 --> 00:09:46,640 Did you run two different geometries, but one, you change 165 00:09:46,640 --> 00:09:49,720 the CFD settings or did you change the material properties 166 00:09:49,720 --> 00:09:53,120 for that crash test? So labeling the data, 167 00:09:53,120 --> 00:09:57,520 classifying the data in a format that a model could read is 168 00:09:57,520 --> 00:10:00,160 important. So for example, you know, coming 169 00:10:00,160 --> 00:10:03,000 up with some sort of schema, some JSON format where you can 170 00:10:03,000 --> 00:10:07,480 say, OK, this data was created on this day using these settings 171 00:10:07,760 --> 00:10:10,720 by this person. You may even add security into 172 00:10:10,720 --> 00:10:13,440 it. I want this to be able to be 173 00:10:13,440 --> 00:10:16,000 trained by this model, not to be trained, you know, if you have 174 00:10:16,000 --> 00:10:19,360 different sorts of data that you want the model to know about, 175 00:10:19,680 --> 00:10:21,920 you should try and describe as well as possible. 176 00:10:22,360 --> 00:10:25,640 The other one, and this is probably more translating now to 177 00:10:25,640 --> 00:10:30,280 new data, is the data that you would traditionally keep or 178 00:10:30,280 --> 00:10:33,440 destroy should be reconsidered in the era of AI. 179 00:10:34,040 --> 00:10:36,280 The example I would give is some CFD. 180 00:10:36,840 --> 00:10:41,240 You may say all I need out of it is the drag and the lift and 181 00:10:41,240 --> 00:10:44,400 some pictures of the flow. And after a certain point, why 182 00:10:44,400 --> 00:10:47,600 do I need the full 3D volume that's 50 gigabytes. 183 00:10:47,600 --> 00:10:51,520 I'm just going to delete that. But by deleting it, you've 184 00:10:51,520 --> 00:10:54,400 probably lost a lot of information that would be needed 185 00:10:54,400 --> 00:10:57,520 by the AI model to go and train the 3D volume solution. 186 00:10:58,240 --> 00:11:01,200 It needs it to be able to learn 3D field. 187 00:11:01,960 --> 00:11:04,840 And if you deleted that, well, you are going to have to 188 00:11:04,840 --> 00:11:08,800 regenerate it. But then how do you regenerate 189 00:11:08,800 --> 00:11:10,920 it with that version of the software that was done and the, 190 00:11:11,160 --> 00:11:13,880 of course, the cost. So traditionally this was a 191 00:11:13,880 --> 00:11:17,400 balance of storage cost versus need. 192 00:11:17,880 --> 00:11:22,880 I would argue now, because data is the key part of AI, it's the 193 00:11:22,880 --> 00:11:26,200 biggest cost of AI when it comes to CAE, you need to keep the 194 00:11:26,200 --> 00:11:27,920 data. So I would argue it's best 195 00:11:27,920 --> 00:11:31,360 investing in paying more for storage to have the data. 196 00:11:31,360 --> 00:11:35,220 Even if you're not today training AI models, you will do, 197 00:11:35,220 --> 00:11:39,960 I'm sure at some point. So generating data, thinking 198 00:11:39,960 --> 00:11:43,451 about how you keep it and maybe outputting more information 199 00:11:43,451 --> 00:11:48,600 than you think you need because the AI model may need it. 200 00:11:48,600 --> 00:11:51,280 So if normally you would just save the pressure and the 201 00:11:51,280 --> 00:11:54,000 velocity, you might think, well, what about all the other 202 00:11:54,200 --> 00:11:58,920 variables that I might need beyond just what I'm 203 00:11:58,920 --> 00:12:02,120 traditionally getting out? You know, instead of being just 204 00:12:02,120 --> 00:12:04,360 the streamwise velocity, maybe I need all the components of the 205 00:12:04,360 --> 00:12:07,240 velocity, little things like that that you would say, well, I 206 00:12:07,240 --> 00:12:09,280 don't need it because I don't need to explore it. 207 00:12:09,760 --> 00:12:11,920 You might need to train a model later. 208 00:12:12,760 --> 00:12:15,800 Another angle to look at is monetizing this. 209 00:12:16,280 --> 00:12:19,160 So if you're a business or engineering company or even 210 00:12:19,160 --> 00:12:23,680 individual, you may find that your data is very valuable and 211 00:12:23,680 --> 00:12:26,920 you may consider is the way of making value out of that. 212 00:12:26,920 --> 00:12:30,440 Can you change your business around a little bit to offer 213 00:12:30,440 --> 00:12:33,000 this data, whether it's computational or experimental, 214 00:12:33,720 --> 00:12:35,400 to help people train? If you're a company that 215 00:12:35,400 --> 00:12:39,040 operates a wind tunnel, maybe you start to think about using 216 00:12:39,040 --> 00:12:42,880 that to generate data or to monetize your data. 217 00:12:43,880 --> 00:12:47,270 So there's a lot of opportunities in the era of AI 218 00:12:47,270 --> 00:12:50,730 physics, which is true for AI, for science, that data is key. 219 00:12:50,738 --> 00:12:53,068 And if you have data, it could be valuable. 220 00:12:53,076 --> 00:12:55,870 And you really need to have a good strategy of how to deal 221 00:12:55,878 --> 00:13:00,077 with this. OK, So you have an open mind, 222 00:13:00,085 --> 00:13:04,400 you've educated yourself, you're preparing for the AI physics. 223 00:13:04,408 --> 00:13:08,932 Now what about doing it? And I think the argument or the 224 00:13:08,932 --> 00:13:11,690 thing I would like to discuss now is really about the 225 00:13:11,690 --> 00:13:14,864 build-versus-buy it's a common question I get when I speak to 226 00:13:14,864 --> 00:13:17,972 many people is around, you know, is there an off- the-shelf 227 00:13:17,972 --> 00:13:19,480 solution. Should I be building this 228 00:13:19,480 --> 00:13:22,960 myself? Well, obviously linked to #2 you 229 00:13:22,960 --> 00:13:27,600 need to educate yourself. And I should say sorry on #3 230 00:13:27,600 --> 00:13:30,440 actually, one thing I forgot to say was staff. 231 00:13:31,560 --> 00:13:34,560 You can prepare yourself from a technology point of view in 232 00:13:34,560 --> 00:13:37,970 terms of getting the data ready, processes ready, but you need 233 00:13:37,970 --> 00:13:39,760 to prepare yourself in terms of hiring. 234 00:13:40,040 --> 00:13:43,040 Do you have somebody in your business who is an AI expert? 235 00:13:43,200 --> 00:13:46,840 If you're a manager, if you're an individual, the the preparing 236 00:13:46,840 --> 00:13:49,120 is to prepare yourself, you know, by doing the education. 237 00:13:49,600 --> 00:13:53,120 But if you're a company, you know, don't expect that people 238 00:13:53,120 --> 00:13:55,600 who have been doing CFD for for 20 years are going to be the 239 00:13:55,600 --> 00:13:58,000 best people to do AI and help you develop. 240 00:13:58,000 --> 00:14:01,920 AI is probably going to be, you know, a bunch of graduates who 241 00:14:01,920 --> 00:14:06,040 are AI-native and are very familiar with those programming 242 00:14:06,040 --> 00:14:08,800 styles with that, maybe they've studied at university. 243 00:14:10,480 --> 00:14:14,560 There's just a reality that staffing is important. 244 00:14:14,560 --> 00:14:18,000 And you may also need to look beyond your typical recruiting 245 00:14:18,000 --> 00:14:22,040 grounds to get those staff. Those people may not come out of 246 00:14:22,040 --> 00:14:25,120 the universities that you would normally go to, to recruit 247 00:14:25,120 --> 00:14:27,400 engineers. And I think this is really 248 00:14:27,400 --> 00:14:30,760 interesting blend. And it's a career opportunity 249 00:14:30,760 --> 00:14:34,240 that AI is needed for engineering, but also engineers 250 00:14:34,240 --> 00:14:36,800 are needing it are needed at AI companies. 251 00:14:37,920 --> 00:14:41,680 So OK, let's go to #4 So you've got an open mind, educate 252 00:14:41,680 --> 00:14:43,880 yourself and you're sort of starting to prepare yourself 253 00:14:43,880 --> 00:14:45,756 from a data and staffing point of view. 254 00:14:45,764 --> 00:14:47,503 What about the actual doing it now? 255 00:14:47,511 --> 00:14:49,825 Well, this is the bill versus buy. 256 00:14:49,833 --> 00:14:54,100 And I would put the analogy to something I'd probably know 257 00:14:54,108 --> 00:14:58,440 best, which is CFD. And if you looked in the 70s and 258 00:14:58,440 --> 00:15:01,536 the 80s, there weren't really commercial solutions on the 259 00:15:01,544 --> 00:15:04,714 market that were the de facto solutions. 260 00:15:04,722 --> 00:15:09,720 In fact, before the 1970s, people, there was no commercial 261 00:15:09,720 --> 00:15:11,898 solutions. And so you had to develop the 262 00:15:11,898 --> 00:15:14,095 code yourself. And that was a big thing of 263 00:15:14,103 --> 00:15:17,028 NASA, but not just NASA, but almost all aerospace companies 264 00:15:17,036 --> 00:15:21,695 and car companies, and to some extent manufacturing, they would 265 00:15:21,695 --> 00:15:25,260 develop their own codes out of necessity if they needed to. 266 00:15:25,268 --> 00:15:27,616 There wasn't, you know, there wasn't a commercial solution 267 00:15:27,624 --> 00:15:31,152 available. It was only in the 70s and the 268 00:15:31,160 --> 00:15:34,960 80s and certainly going into the 90s and the 2000s where 269 00:15:34,968 --> 00:15:37,800 commercial codes became far more mature. 270 00:15:37,808 --> 00:15:43,800 Interestingly, if you look today and this is maybe contentious 271 00:15:43,808 --> 00:15:48,060 opinion, lots of these companies that did traditionally only use 272 00:15:48,060 --> 00:15:52,276 internal codes are starting to use commercial codes and that 273 00:15:52,284 --> 00:15:55,140 shift is accelerating because frankly, the commercial codes 274 00:15:55,148 --> 00:15:58,436 are getting so good. They've hired so many people, 275 00:15:58,444 --> 00:16:02,340 they've, you know, acquired start-ups that is your code 276 00:16:02,348 --> 00:16:08,140 really as good as theirs. I think it's a interesting 277 00:16:08,148 --> 00:16:11,920 debate. So debate of do you see it as 278 00:16:11,928 --> 00:16:15,242 core to your mission to have your own code or actually is 279 00:16:15,250 --> 00:16:18,320 your main business to go and build something and the code 280 00:16:18,328 --> 00:16:20,240 itself should be done by somebody else. 281 00:16:20,248 --> 00:16:23,495 Most people are starting to shift towards the latter, that 282 00:16:23,503 --> 00:16:27,060 actually it's better just to buy in the code and have less people 283 00:16:27,060 --> 00:16:29,880 developing their own code. Now again, I'm not saying that 284 00:16:29,888 --> 00:16:33,378 is the right approach, I'm just saying that is what I see 285 00:16:33,386 --> 00:16:36,834 happening. So why am I saying that? 286 00:16:36,842 --> 00:16:41,457 If you look at AI physics, it's probably the fact that we are in 287 00:16:41,457 --> 00:16:43,269 the equivalent of the 70s or 80s. 288 00:16:43,277 --> 00:16:48,908 We're in this new phase where there is not the same maturity 289 00:16:48,916 --> 00:16:51,820 of commercial solutions. It's largely start-ups at the 290 00:16:51,828 --> 00:16:55,100 moment. And you could argue that 291 00:16:55,108 --> 00:16:57,680 actually, could I build this myself? 292 00:16:57,688 --> 00:17:00,985 Does that give me a strategic advantage like it was a 293 00:17:00,993 --> 00:17:03,604 strategic advantage building your own CFD code? 294 00:17:03,612 --> 00:17:07,593 I would say I can understand the arguments for both. 295 00:17:07,601 --> 00:17:12,474 I can today I can see why maybe you think if you hire some, you 296 00:17:12,474 --> 00:17:15,240 know, smart engineers and build on open-source frameworks that 297 00:17:15,248 --> 00:17:21,119 you, you could do that, but you have to be prepared to keep up. 298 00:17:21,119 --> 00:17:23,760 And that's my advice or warning 299 00:17:23,760 --> 00:17:30,120 that even if today there is a split on, oh, it's not so 300 00:17:30,120 --> 00:17:34,320 obvious whether to build or buy, you should be flexible and 301 00:17:34,320 --> 00:17:37,400 prepared that maybe in two years, maybe in five years, 302 00:17:37,680 --> 00:17:41,840 maybe at the extreme 10 years, I think it's highly likely that 303 00:17:42,000 --> 00:17:45,374 the commercial solutions will be as good or better than what 304 00:17:45,374 --> 00:17:48,440 you could do yourself. And so just as people are now 305 00:17:48,640 --> 00:17:51,480 having to debate whether to bring in commercial solutions 306 00:17:51,480 --> 00:17:54,560 from a non AI point of view, I would say you need to be 307 00:17:54,560 --> 00:17:58,200 flexible enough to be able to adopt when there is a good off 308 00:17:58,200 --> 00:18:01,360 the shelf solution that's maybe better than what you could do 309 00:18:01,760 --> 00:18:03,880 internally. And that's part of the sort of 310 00:18:03,880 --> 00:18:08,960 prepare yourself and planning. Luckily today most solutions are 311 00:18:09,240 --> 00:18:13,840 sort of API driven and therefore if done in the right way, you 312 00:18:13,840 --> 00:18:16,800 can sort of integrate things together a bit like you can use 313 00:18:16,800 --> 00:18:20,360 the API of OpenAI to call that model and call the different 314 00:18:20,360 --> 00:18:23,360 model. I suspect it'll be the case. 315 00:18:23,360 --> 00:18:29,640 So I would 100% advocate people coding and building stuff 316 00:18:29,640 --> 00:18:34,080 themselves, whether they go to production with it or they use a 317 00:18:34,080 --> 00:18:37,160 start up or or buy something. It's very much an individual 318 00:18:37,160 --> 00:18:39,480 choice. But I do have a strong feeling 319 00:18:40,000 --> 00:18:44,880 that given the investment it is likely that they'll be so many 320 00:18:44,880 --> 00:18:47,040 better and good commercial solutions. 321 00:18:47,040 --> 00:18:51,880 Given how big this AI market is for for CAE, that you may need 322 00:18:51,880 --> 00:18:55,560 to be flexible on that choice and reassess it and don't get 323 00:18:55,560 --> 00:18:59,200 locked in to thinking I can't take a commercial because I've, 324 00:18:59,600 --> 00:19:01,120 you know, gone down building myself. 325 00:19:01,120 --> 00:19:02,960 I need to. There's nothing wrong with 326 00:19:03,200 --> 00:19:09,560 having a mixture. And finally, .5, which is 327 00:19:09,800 --> 00:19:13,520 probably the newest, I would argue in the wave of AI for CAE 328 00:19:13,560 --> 00:19:17,160 is agentic AI and a bit like AI physics. 329 00:19:17,160 --> 00:19:19,640 Sometimes people can get people's backs up with the 330 00:19:19,640 --> 00:19:23,536 marketing and the way that it's pushed, that AI can automate and 331 00:19:23,536 --> 00:19:26,800 do everything. Essentially the agentic AI is to 332 00:19:26,800 --> 00:19:30,360 sort of next wave of where LLMs were. 333 00:19:30,920 --> 00:19:35,120 So in the sense what are the frustrations of using a 334 00:19:35,560 --> 00:19:38,480 traditional LLM is it can't go and do things for you. 335 00:19:38,760 --> 00:19:43,160 It's very much a, you know, look at this document and summarize 336 00:19:43,160 --> 00:19:45,800 it for me. But what you would really like 337 00:19:45,800 --> 00:19:49,080 to do, and to be fair, even now, some of the off the shelf that 338 00:19:49,080 --> 00:19:52,720 can do it, you'd say, I would like you to go and do this for 339 00:19:52,720 --> 00:19:54,840 me. Go and research something, go go 340 00:19:54,840 --> 00:19:57,160 on the Internet, go and search for this, then call this, then 341 00:19:57,160 --> 00:19:59,240 do this. So what does that mean in the 342 00:19:59,240 --> 00:20:05,360 context of, of of CAE? That means at it's very simplest 343 00:20:06,080 --> 00:20:09,960 things like a text prompt to set up simulations and run 344 00:20:09,960 --> 00:20:14,000 simulations. So rather than you clicking 345 00:20:14,000 --> 00:20:16,200 buttons and and going and setting up a simulation and 346 00:20:16,200 --> 00:20:17,640 manually running a bash script. 347 00:20:17,648 --> 00:20:21,330 It's through a text prompt. You should be able to instruct 348 00:20:21,338 --> 00:20:23,256 an agent. And usually there's some sort of 349 00:20:23,256 --> 00:20:25,344 master agent that's then controlling other agents and 350 00:20:25,352 --> 00:20:27,662 sending that on. And I will do a dedicated 351 00:20:27,670 --> 00:20:31,740 episode on this because I think it's such a fascinating topic 352 00:20:31,748 --> 00:20:36,900 that essentially allows you then to quote-unquote, have a sort 353 00:20:36,908 --> 00:20:41,886 of AI engineer where that agent would be able to go and call 354 00:20:41,894 --> 00:20:44,245 other agents. And again, the reason I'll do a 355 00:20:44,245 --> 00:20:46,144 dedicated episode on this is there's been some good papers 356 00:20:46,152 --> 00:20:49,089 out there that I'd like to discuss and and talk about. 357 00:20:49,097 --> 00:20:53,278 But at a very high level, it essentially would be the ability 358 00:20:53,278 --> 00:20:56,720 through a text prompt for one agent to call another agent, 359 00:20:56,728 --> 00:21:02,015 which is perhaps a surrogate model to go run a simulation. 360 00:21:02,023 --> 00:21:06,284 But importantly, another agent would then perhaps do an 361 00:21:06,292 --> 00:21:09,830 analysis of that, but in a fully automated way. 362 00:21:09,838 --> 00:21:14,400 And then finally, another agent may decide to go and then do 363 00:21:14,408 --> 00:21:17,260 some optimization of the geometry and pass the 364 00:21:17,268 --> 00:21:20,707 information back to your surrogate model agent. 365 00:21:20,715 --> 00:21:25,519 Then do more analysis, let's say image analysis through an LLM 366 00:21:25,527 --> 00:21:29,600 and they may write you a PDF. And the important bit that's all 367 00:21:29,600 --> 00:21:32,252 been kicked off by one prompt from yourself. 368 00:21:32,260 --> 00:21:35,832 So it's agents that are essentially calling each other. 369 00:21:35,832 --> 00:21:37,560 And there's a lot of complexity, 370 00:21:37,560 --> 00:21:41,720 of course, with this, but it really gets much more than just 371 00:21:41,720 --> 00:21:44,040 the surrogate. The surrogate still relies you 372 00:21:44,120 --> 00:21:48,400 as a human to run the simulation, a bit like a CFD 373 00:21:48,400 --> 00:21:52,040 simulation or FEA simulation. The agentic side really starts 374 00:21:52,040 --> 00:21:55,800 to get more, I would say, into the the vision of AI, where it's 375 00:21:55,800 --> 00:21:58,824 more fully automated. And if you extrapolate this to 376 00:21:58,824 --> 00:22:02,436 its maximum, you can imagine many agents acting almost like 377 00:22:02,436 --> 00:22:07,040 their own engineering company. That's obviously quite far into 378 00:22:07,040 --> 00:22:09,760 the future. But I would start preparing for 379 00:22:09,760 --> 00:22:11,080 that. I would start reading up on 380 00:22:11,080 --> 00:22:13,360 that. This is very much where the the 381 00:22:13,520 --> 00:22:17,560 AI researchers at the moment go and read papers, type in into 382 00:22:17,560 --> 00:22:20,880 Google agentic AI, go into YouTube, watch videos, look for 383 00:22:20,880 --> 00:22:24,360 start-ups, speak to them. This is definitely a new wave 384 00:22:24,360 --> 00:22:27,840 that's coming that AI physics is a crucial part of it. 385 00:22:28,640 --> 00:22:32,880 But the agentic AI is arguably more potentially transformative 386 00:22:32,920 --> 00:22:35,640 and more important to to be aware of. 387 00:22:35,640 --> 00:22:38,880 But there's even, I'd say less solutions on the market now, 388 00:22:38,880 --> 00:22:41,760 which is why it's really good to stay ahead of the curve. 389 00:22:42,480 --> 00:22:47,520 So those are just 5 tips being open minded, educate yourself, 390 00:22:47,760 --> 00:22:50,360 prepare for this sort of AI physics revolution. 391 00:22:51,160 --> 00:22:54,520 Get involved in the AI physics, try stuff out, build, buy, you 392 00:22:54,520 --> 00:22:57,760 know, kick the tires. And then prepare yourself for 393 00:22:57,760 --> 00:23:02,720 the agentic AI move, which could really blow away and be quite 394 00:23:02,720 --> 00:23:06,920 transformative if things live up to what people hope for. 395 00:23:07,640 --> 00:23:09,280 So I hope this has been interesting. 396 00:23:09,280 --> 00:23:11,280 I tried to keep it a bit short and snappy. 397 00:23:11,640 --> 00:23:13,209 I hope you've learned something. 398 00:23:13,217 --> 00:23:16,176 Agree with at least some of the stuff I said. 399 00:23:16,184 --> 00:23:19,882 And what I'm going to do is over the course of the the next 400 00:23:19,890 --> 00:23:24,422 season, we'll, we'll dive into a couple of these topics a little 401 00:23:24,422 --> 00:23:28,704 bit more and hopefully do #2 educate a little bit. 402 00:23:28,712 --> 00:23:32,435 So with that, thanks very much for listening and hope to see 403 00:23:32,443 --> 00:23:33,960 you in the next episode.