HH54-15-10-21_mixdown.mp3-from OneDrive

Harpreet: [00:00:08] What's up, everybody, welcome, welcome to the artist Data size happy hour it is Friday, October 15th, 2020. What? I'm your host. Harpreet Sahota Super excited to have all the guys here. Hopefully you've had an amazing week. I, for one, have been sick the entire week. I probably shouldn't be drinking beer just because of the sheer volume of acetaminophen that I have ingested. But I'm gonna do it anyways because, you know, I'm feeling a little bit better today. I feel a little bit better. And for that, I celebrate because my health is getting better. Confucius once said that that a healthy man wants a thousand things. The sick man just wants one thing, and that is absolutely true. Her words have never been said. I hope you guys got a chance to tune into the podcast episode that I

Harpreet: [00:00:55] Released

Harpreet: [00:00:56] Just on Friday with Eric. So this this episode was actually the very first conversations episode that I ever recorded actually recorded this episode well over a year ago. It was about September 20 20 when I recorded this episode. So I'm interested to go and listen back at it to see how much I've changed since then. I wonder what kind of shit I was saying in that episode, so I'm excited to have all you guys here. A lot of cool stuff happened this week. You know, I was on

Harpreet: [00:01:28] The Data and

Harpreet: [00:01:29] Impact podcast hosted by community

Harpreet: [00:01:31] Member Christian

Harpreet: [00:01:33] Steinert, so thank you very much for having me on the show. Kristen had a cool episode release with that I saw on the Continuous Neighbor podcast with both Kanji and Marc Freeman. I thought the super cool to see that happen, then epic collaboration. There are a lot of other cool stuff too. Man happened this

Harpreet: [00:01:51] Week

Harpreet: [00:01:52] Was did two presentations this week, one at the ML

Harpreet: [00:01:56] Conf that was just

Harpreet: [00:01:57] Earlier

Harpreet: [00:01:58] Today and did a in-house [00:02:00] kind of presentation

Harpreet: [00:02:01] For our house list at Comet Ml. And so that was. That's a lot of fun. I hope the guys get a chance to check that out. If not, don't worry, I'll be giving presentations for a for quite some time. Shout out to everybody in the room. What's up? Ben Taylor, Kenji Russel Willis, Eric Sims and he's in the background

Harpreet: [00:02:18] Goes in the building.

Harpreet: [00:02:20] So there's Alexandra. Super excited to see all you guys here. So, man, I got a question for you guys. Let's open with this. I'm wondering what? What. What does it mean for a data scientist, be creative and is creativity a necessary skill for success as a data scientist? I'd love to get your thoughts on this. Let's go to Ben Taylor first, and for Ben Taylor, go to Kenji, then Eric Simpson. Anybody else wants to chime in? Please do let me know.

Speaker3: [00:02:52] Um, so I think there's a lot of people in the Data science community that are not creative. And and I'll give an example, so if I said elderly slip and fall, let's go fix this. We're going to go do this initiative to fix elderly slip and fall with I. Most people would say we're going to do video cameras in the hallway. We're going to look at you. You're going to follow. We're going to build an A.I. model. And I would say the creativity with that is horse blinders like they're navel gazing. And I think the distinction I'd say, don't think like, I don't think like some narrow scope perspective, think like a human. So what would a human do? Well, a human would have. You have lots of cameras, but more importantly, a human would be reactive to. Oh, it's eight a.m. You didn't wake up for breakfast. I have a problem with that. Normally you're here for breakfast or you went to the bathroom and you've been there for 30 minutes. I'm getting concerned two hours. I'm very concerned. And so I think it's two times too many times in data science. It's extremely narrow. And which I don't know why that is, because Data scientists are typically really smart. But I see [00:04:00] a lot of data

Harpreet: [00:04:00] Scientists that

Speaker3: [00:04:01] They see the world like this rather than seeing the world like this. So I'm curious how other people react to that. That is is a lack of creativity, a problem in the data science space? And why do so many people take the approach of Hot Dog, not hot dog? I'm going to do a cancer classifier rather than taking the approach of like. What would a human be are going to take all the Data matters. They'll be quiet to that.

Harpreet: [00:04:28] That's a good

Harpreet: [00:04:29] Addition to that question. Let's go to let's go to Kenji in Africa and you can go to Eric and then Russell and anybody else wants to jump in on this one. Please do. The opening question

Harpreet: [00:04:39] Is it's

Harpreet: [00:04:40] About creativity and Data science. What does it mean to be creative in Data science?

Harpreet: [00:04:44] And there's

Harpreet: [00:04:45] Creativity, but requires skill for success and Data

Harpreet: [00:04:48] Science also shout

Harpreet: [00:04:49] Out to everybody else that just joined in the room Gina and truck so you guys can go for it.

Speaker4: [00:04:55] So I'm going to make the bold statement that I believe creativity is integral to the success of most fields, particularly Data science, right? But I think the way that we define creativity is something

Harpreet: [00:05:10] That that that varies greatly. Right? Like, there's different levels

Speaker4: [00:05:14] Of creativity, like you can be very creative in the model training process. You can be very creative in the future engineering process. You can be creative in the

Harpreet: [00:05:22] Questions you ask or the ideation

Speaker4: [00:05:24] Process

Harpreet: [00:05:24] Or has been described

Speaker4: [00:05:26] The way that you approach certain problems. And I think that in order to be successful

Harpreet: [00:05:31] In this domain, you have to be

Speaker4: [00:05:32] Creative in at least one of those areas. But most people are not creative and like all of those areas or multiple of those areas.

Harpreet: [00:05:41] And I think that

Speaker4: [00:05:43] Something we should all seek to do better is think about, OK, where is our creative strength right? I personally think that when I'm explaining

Harpreet: [00:05:51] Things, I come up with

Speaker4: [00:05:53] Creative ways to view

Harpreet: [00:05:54] Them

Speaker4: [00:05:55] Analogies

Harpreet: [00:05:56] Or or

Speaker4: [00:05:57] Different ways to present information

Harpreet: [00:05:59] That would help [00:06:00] someone

Speaker4: [00:06:00] Else to digest it right? Like that is a creative strength of mine. On the other side, a lot of the times when I'm when I'm doing like actual data science work, I find myself doing a lot of the same things over and over again because I've worked historically or whatever it might be. I'm not flexing the creative muscles as much as I probably should be doing. They could lead to greater insight, a better understanding of the problem. Or it could potentially even show that the way I've done things in the past was not a good way to do them to begin with. So I think that, you know, I would answer this question in a really annoying way with another question is like how do we define this creativity? And you know, are there better ways for us to practice and exercise creativity in the different compartments of our life or the different compartments of our work?

Harpreet: [00:06:49] All right, Ken, thank you so much. Appreciate that. There are excellent comments and excellent, you know, things to think about. Eric, let's hear from you then than Russell as well than Mexico's no filter response. I want to hear that as well. And by the way, everybody listen on LinkedIn, on YouTube in the chat here.

Harpreet: [00:07:08] Please let me

Harpreet: [00:07:08] Know if you got a question, just

Harpreet: [00:07:10] Just comment or right

Harpreet: [00:07:12] Right here into the chat that you got a question. I'll add you to the queue. Go for it, Eric.

Speaker5: [00:07:17] So I would say that creativity is automatic at the live stream going on on this other monitor. Turn that off. Ok. So creativity to me is. Approaching something in just the way that's different from the people who are in your immediate vicinity, right? And so because you could be a tremendously uncreative, let's say, artist compared to the artists that you maybe spend most of your time around, then you put yourself in a different situation. And all of a sudden you are tremendously creative because you bring a perspective that nobody else in that group brought. And hopefully, that can even unlock [00:08:00] creativity or new thought previously on had thoughts with other people, you know, and it doesn't necessarily mean that you are, I guess, artistic. We usually kind of say it like that, but I think it's just bringing something, bringing your own thing to that environment. And so that's why I think it's really some of the best advice I ever got from previous boss was

Harpreet: [00:08:26] To he just called it dig

Speaker5: [00:08:27] Your own job. And so I was like, Oh, dig your job. Like, enjoy.

Harpreet: [00:08:31] He's like, No, dig your job as in, you're all here

Speaker5: [00:08:34] Being, you know, analysts or Data scientists or whatever you are. But like, find your spot, dig into it and don't let anybody get you out of

Harpreet: [00:08:42] It because you are the best

Speaker5: [00:08:44] At that thing and you bring that to the table and then you can

Harpreet: [00:08:47] Just get get really good at it and be that

Speaker5: [00:08:49] That awesome person that helps lift the team with what you bring. So that's kind of my take on creativity. But at the same time, I also just like coming up with goofy and original things that maybe people haven't thought of before, because it's just fun.

Harpreet: [00:09:03] A lot of these responses, thank you very much, Eric. Ken and Mickey go doing the Aloha Friday. Ken's asking, Where's my shirt? It's it's full on winter here now in Winnipeg. It's like forty eight degrees cold and rainy outside. It is only cashmere season for me.

Speaker4: [00:09:19] Aloha is a mindset that's not not a not a temperature. Come on.

Harpreet: [00:09:25] Yeah, that's true. I got to find some. Rural voters, I bet I could find some of those. There you go. Straight up, Mickey, give us your give us your your hot take

Harpreet: [00:09:35] On on on the

Harpreet: [00:09:36] Question at hand. And remember, folks, we're talking about creativity and Data science. What does that mean? Is that something so important to success? Let's go to Mexico, then Russell and then Gene. I see you have a comment here. I'd love for you to share that with us.

Speaker6: [00:09:53] Yeah, so I think so what I think creativity and Data science or lack of I I'm going to be honest, I'm thinking adults because [00:10:00] kids and teens are definitely well before its speed now of them, they're remarkably creative and intelligent and funny and all that. But I think in terms of like the rest of us, older folks, I feel like a lack of creativity honestly comes down to like empathy and world experiences. And this and you see this a lot. I feel like, for example, on the product side where I don't know, you know, like people either don't dog food, their own product or whatever. And so, you know, their users will complain about like, we can't do this or it's not flexible, it's not user friendly, like whatever. And the engineers will go like, No, no, no, it's perfect. Like it's built like this. It's like, have you, Mr. Mrs. or a person engineer? Have you actually tried like using your software because it's it's not pleasant or things, for example, like? And this can kind of be one of those challenges if you're doing modeling or analysis in a domain that you have no experience with, like, for example, medical or sports. If I mean, if you tried tossing me out of baseball problem, I'm sure I would screw it up and Ken would have lots of things to say about it, like, why didn't you try this? Why didn't you do this? Why don't you

Harpreet: [00:11:16] Do?

Speaker6: [00:11:17] The sports means like you throw a ball at someone, maybe at their face, maybe across a field? Who knows, but you'll run into that. And I think it's it's one of these weird things where like. It feels like the solve would be to just like practice empathy, try out different life experiences, try doing work in a different domain, try doing a Kaggle competition, whatever right. But like literally just expose yourself to something new and take yourself out of your current status quo. And I don't think a lot of people do that because of, like the whole explore and exploit trade off, right? Like, they kind of want to stay in the pocket of what they're really, really comfortable with. But that means that they end up missing out on like the the big value [00:12:00] opportunities, which a lot of times are out of your comfort zone. So I know it's it's not a hot take, it's a lukewarm take.

Harpreet: [00:12:06] But I like it. I like it, though. Thank you so much. Russell Lewis, let's hear from you on this and then Alexandra, if you'd like to jump in here. Sorry, sorry. Go to Russell, then Gina. Then if Alexander wants to jump in, I'd love to hear from you. Also shout out to Dylan. Good to see you here again, my friend Aspinwall jewelries is also in the building. What's up, Doris? Russell go for it. Then we'll go to Gina, then Alexandra. And then, by the way, anybody listening, like I say, we're taking questions. So if you got questions, let me know. I'll add you to the queue.

Speaker5: [00:12:36] Thank you, Aubrey, evening, everybody. So I agree with everybody's statement so far on creativity. I put two comments in the chat. The first was saying creativity isn't essential for anything, but it certainly enhances and can accelerate most tasks or actions or anything that you need to do. But if we take creativity as like a dictionary translation, it means to create something. So if you if you write some code, you're creating it, therefore you're created to a certain extent. So I think Ken's earlier comments about exactly what creativity is were very pertinent. So my my first comment

Harpreet: [00:13:20] Was taken

Speaker5: [00:13:21] Creativity to be, you know, to the to the extent, you know, a really creative person that can come up with some out-of-the-box thinking. Okay? And then my next comment was saying that creativity is generally inversely proportional to orthodoxy. So if you learn one way to do something and you do it that same way every time, that's the status quo. That's the Orthodox approach. You could be deemed to be low creativity if you change it up and you bring learnings from something else into it, which may bring good or bad results. But you're doing something different and can learn from all of the positive aspects from that. Then I [00:14:00] would think that creative is to say, for example, you weren't talking about Data science. We were talking about, you know, cooking your your evening meal in the kitchen. You know, you have the same thing week on week. You not be very creative. You cook different meals each day of the week, you'd be a little more creative. You cook different meals every day for, say, a month, a bit more creative. Then you know, you mix up those meals. Every time we cook the same meal, we throw something different than, you know, a different herb, a different seasoning, a different ingredient. You just change stuff up. You're even more creative. So there's definitely a scale of creativity. And in my opinion, creativity will will help everything, even coding. So as I say, if you if you choose one function or command to get a certain return in some of

Harpreet: [00:14:46] The code and one day

Speaker5: [00:14:47] You shake it up and try something else, you may find that you can come up with the same result in, say, 10 steps rather than 100 steps and find that it's going to, you know, really improve the efficiency and performance of the model. So creativity there, I think, definitely enhances, but the bare bones basic creativity will get you by. It'll just be a bit more boring, shall we say.

Harpreet: [00:15:14] Thank you very much, Russell. Gina, let's hear from you, then, after Gina will go to Alexandra.

Speaker5: [00:15:20] Yeah, hi.

Harpreet: [00:15:22] Great to

Speaker5: [00:15:22] Be here again with my second visit

Harpreet: [00:15:25] To Data Science, happy

Speaker5: [00:15:26] Hour. I have a lot of thoughts on creativity. But let me just limit it to a couple and kind of

Harpreet: [00:15:34] Thinking that

Speaker5: [00:15:35] Piggybacking off of Russell's answer. So I think a lot of times like related to what Mexico said and then, you know, also clarified that of course, children are often creative because no one's told them, you can't do this or that until they tell them, Oh, that's stupid, or that's not the right way to do it or, you know, et cetera, et cetera. And so, you know, along [00:16:00] the lines of what has previously been said, I think, you know, aside from all the other influences school, family, et cetera, organizations, the organization, both the organization you're

Harpreet: [00:16:11] In and the

Speaker5: [00:16:12] Organizations, maybe that you've been in in the past, people are going to you're going to respond. Everybody is going to respond to the the the incentives that are presented to them. And so stepping aside from Data just for a minute, I could tell you, I've been in a number of organizations and I've almost kind of.

Harpreet: [00:16:33] You know, you somehow wouldn't be. You want to meet yourself. You're right in the middle of seeing some awesome.

Speaker5: [00:16:43] Okay, there we go. Sorry about that, that was odd. Yeah, yeah. Maybe Siri or something, but yeah, I think it depends on the organization you're in. And what I was saying is stepping away from the data science side just for a moment. I've been in organizations where I feel like you like I've been kind of whipsawed in a way. One organization like almost mandated, you know, aggressive might be the wrong word, but some organizations, you know, almost required that if you don't speak up, you've got nothing to contribute. Other organizations are so much more consensus based, and having come from one organization that was

Harpreet: [00:17:22] Very much the former

Speaker5: [00:17:23] And then going to one, that was very much the latter. I made the mistake early on of maybe, you know, expressing my opinions a little too forcefully, and I don't think that got me off on the right

Harpreet: [00:17:35] Foot with a lot of people. And so I think I

Speaker5: [00:17:38] Just I want to put that out there. Sorry, I got a little background noise to consider that.

Harpreet: [00:17:45] And in your current

Speaker5: [00:17:46] Organizations, please encourage

Harpreet: [00:17:50] People

Speaker5: [00:17:51] To be creative. And if people are encouraged to be creative, then maybe they're not going to be very creative. So it's a combination of, you [00:18:00] know, realizing that sometimes if people don't seem creative, it might be because they were that was stifled for them in the past. And so I guess be mindful of what incentives are put in front of people because I think a lot of people have creativity, but it's a matter of whether or not they were rewarded for expressing it. So I guess what? I'm the long. The short answer is, you know, let's don't assume that people aren't creative just because, you know, maybe they don't initially seem to have the most creative answers, even for what they might post in their GitHub

Harpreet: [00:18:37] Or on

Speaker5: [00:18:38] Their profiles. They might be timid about putting something out there because maybe they believe that

Harpreet: [00:18:47] They need to show like,

Speaker5: [00:18:49] Look, I've got down the basics. I can solve these types of problems in in music. It's kind of the equivalent of man, I've got my

Harpreet: [00:18:57] Scales down, you know, I've

Speaker5: [00:18:59] Got my arpeggios,

Harpreet: [00:19:00] I've got all these basic kind of building

Speaker5: [00:19:03] Blocks that show you, you can at least you can at least do the basics or or even technically do more than the basics. But you're not

Harpreet: [00:19:12] Quite a jazz improv.

Harpreet: [00:19:15] Absolutely love that, Gina, thank you so much. Let's hear from Alexandra on this and then coastal, I see your hand up, is that too to speak on this topic as well? Awesome, so go to Alexandra, then we'll go to Costa, but then Greg has a question that was left over from last week about compilers. So Greg, if you can type out the question, put it in the chat so we can prep people about compilers. Alexander, go for it.

Speaker6: [00:19:39] I was just going to add know I'm coming from the perspective of being a master's student right now, and I think for me, the biggest hindrance in the opportunity to be creative is really just time. I think a lot of times we have the the innate ability to express that creativity. But then all of a sudden, when the professor assigns you five more projects that are due at [00:20:00] the end of the

Harpreet: [00:20:00] Week and a teammate

Speaker6: [00:20:02] Comes in with a Data set that they've used before in a project idea of ready to go. It's really easy to kind of jump on the bat bandwagon and just go with the low hanging fruit of what's easy, especially from from a college perspective. So I think one thing that I'm really trying to prioritize in the creative Data science process that are going through is just kind of almost giving myself time block saying if they present a project idea or a proposal today, I'm not allowed to even answer that question until tomorrow or the next day or the end of the week. And sometimes that's not always feasible. We don't always have the the pleasure of extra time, but when it is, I think, just giving yourself that space, that mental capacity to think about problems, it really helps for me.

Harpreet: [00:20:43] So important, man, like that time and space thing. Like lately, I just haven't had much time or space, and I just feel like I just haven't had a good

Harpreet: [00:20:51] Ideas like

Harpreet: [00:20:51] My ideas are just. Nothing really say that. Let's go to a

Harpreet: [00:20:59] Coke stub, right,

Harpreet: [00:21:00] I think that's who's next in line coke stub. And then and then we'll go to Greg's question on compilers.

Speaker3: [00:21:07] Yeah, Alexandra, you touched on a really good point there about giving yourself time, excuse me, giving yourself time to be creative. I think that's really important. But there is another side of it when when we complain about teams aren't being creative enough in the workplace or we're not able to be creative enough, a lot of it comes really down to have we structured the team properly with the right variety of skill sets. Now, if it is a team of like four Data scientists, you're dealing with a particular problem in one domain space and all of you have the

Harpreet: [00:21:36] Same background

Speaker3: [00:21:38] Experience. You're not. You're all going to approach the problem

Harpreet: [00:21:40] Typically in a quite similar

Speaker3: [00:21:41] Manner, right? Like where I find some of the

Harpreet: [00:21:44] Benefit is as like because I

Speaker3: [00:21:46] Come from a bit of a weird background that is not more from the robotics and vision side of things. I try to apply a different lens to things that say someone who's not from a robotics background might. Right? So if we're able to bring our experiences from different [00:22:00] areas of life into our problem

Harpreet: [00:22:02] Solving process, what we basically

Speaker3: [00:22:04] Rely on is what they call synthesis level

Harpreet: [00:22:07] Thinking, right?

Speaker3: [00:22:08] And that's

Harpreet: [00:22:09] Something that

Speaker3: [00:22:10] Comes from a combination of time spent

Harpreet: [00:22:12] In a domain and

Speaker3: [00:22:14] Time spent in multiple domains. So in restructuring our teams, if we want to be creative and data science, we've got to structure our team

Harpreet: [00:22:20] So that what I

Speaker3: [00:22:21] Bring to the table is significantly different to what someone else brings into the table. Right. So it really does come down to team structure.

Harpreet: [00:22:29] And then, of

Speaker3: [00:22:29] Course, like Alexander

Harpreet: [00:22:30] Said, giving them

Speaker3: [00:22:31] The space and like Gina said, giving them the leash

Harpreet: [00:22:33] To do it, to be creative.

Harpreet: [00:22:37] Joseph, thank you so much. Thank you, guys for participating the opening question, a lot of great nuggets here. Comment coming in from LinkedIn, from a woman in there saying this, there is definitely a price to pay when it comes to creativity. That is time and effort and with not much guarantee on result. While our current work culture is the result is result oriented, does not support scope for creativity. It was nice to know that Google gives its employees 20 percent

Harpreet: [00:23:03] Of their time

Harpreet: [00:23:04] For creative projects, and many of

Harpreet: [00:23:06] Google's great projects came from these side projects.

Harpreet: [00:23:09] Yep, that's absolutely true. We'll go to Greg's question about compilers, but before we get to Greg's question about compilers, the other often use I want to show you guys this week is I just heard that

Harpreet: [00:23:20] Our community member, Antonio

Harpreet: [00:23:23] And his wife just brought a baby boy into the world. So congratulations, San Antonio. Go for it, Greg.

Speaker5: [00:23:34] I put my questions in the in the chat here. It's about, you know, having someone explain to me what compilers are who in the data science

Harpreet: [00:23:45] Team should know about them.

Speaker5: [00:23:47] Is it the ML engineer? And how can. What are the best practices to make sure that these models are built to be optimized from a sourcing,

Harpreet: [00:23:59] You know, [00:24:00]

Speaker5: [00:24:00] Processing perspective? Or is it applicable only? Or should I worry only about it for edge cases where, you know, I read I read Quinn's article about this, but anyone who has some insights about compilers would be super helpful, especially like things like always going to see companies design their own chips to optimize specific for specific use cases, whether it does deep neural network or things like that. I know

Harpreet: [00:24:33] Michiko, I'm happy

Speaker5: [00:24:35] You're here already ready to rock and roll.

Harpreet: [00:24:37] So can you break it down? And also just said just, you know, explain like I'm five, like what is compilers? What's their relationship to machine learning as well as Greg's question?

Speaker6: [00:24:51] Yeah, so to be honest, I actually might not be the best person to describe compilers versus interpreters, mainly because I feel like. So at a high level like compilers, they take source code and put them into machine code and then interpreters, which then still has to be like repackaged into another stage and then interpreters kind of go from source code to like the output. So they kind of can just run the program. So a book I'd recommend actually that I picked up recently is crafting interpreters by Nystrom. It's actually a really, really delightful read. And for me personally, I'm trying to get back into the fun mills. I guess my question is in a way is like. Here is how to like how deep into the nuts and bolts. Does your team need to be in the sense that so even in my company right now, like we're really kind of trying to figure out where do machine learning engineers like begin and end, especially like in terms of how we navigate? For example, like tests like software testing at scale? That's like a big thing we're trying to figure out right now. Is that something that, like the Data scientists [00:26:00] should be writing? Is that something that, like the engineers should be writing? Or is that something that should be like on the Data side? Like how much of it should be automated versus sort of handcraft? So I mean, in in terms of kind of like who would need to know that know like about compilers, I feel like for me, part of it depends on like what kind of level of tooling you're building.

Speaker6: [00:26:24] So for example, if you're building. If you're building like in-house developer tooling. That needs to be really proprietary. You probably do want engineers who have a really good understanding of things like how to craft compilers, interpreters, what languages should you be using? What are the sort of like trade offs and kind of keeping up with that? Things like environment, environment management, package dependency management, the engineers should definitely kind of know those things. But it kind of just depends on like what level of abstraction your ML engineering team is operating now because some small engineering teams, they tend not. So for example, like my team, I feel like in some ways we're sort of mid to low level where we have to really care about things like dependency, like dependency management environment.

Speaker6: [00:27:18] Setting up the environments for, like the Data scientists because we're essentially providing in-house developer tooling, but we're not necessarily going so far as to like. Um, like totally craft things, craftsman craft things from scratch. We still operate at a fairly high level of abstraction, I would say. But some places, for example, like I imagine if you're like Waymo or Tesla, are you doing anything self-driving that would probably be much more of a concern, especially if you're dealing with like sort of more kind of real time. But I would say, like it, just in general, kind of like the question of who who in [00:28:00] like so in the like the supply chain of ML, like who does what? I think a lot of those boundaries do tend to be a little bit sort of like gray. And it's, I think, a very big discussion at most companies like who does what do you do the engineer or do you like do the engineers take care of the engineering? And when we say take care of the engineering? Are they doing infra or are they doing like in-house developer tooling? Are they doing customer facing stuff? So I do think that's like an active conversation.

Harpreet: [00:28:28] It says, let's hear from you on these compilers, Joe Diwaniya, inside Joe or Ben, any insight on compilers? Up, go for it.

Speaker3: [00:28:37] Yeah, I guess the other thing it kind of depends on is what what's the software stack? You're sorry. What's the hardware stack you're

Harpreet: [00:28:44] Working on, right? Like if you're working

Speaker3: [00:28:46] On on cloud computing devices, then a lot of it does get

Harpreet: [00:28:50] Abstracted away from you

Speaker3: [00:28:51] Because you are relying on things like GCP and OS. But on the other hand, for example, when when I was doing machine learning on a robotic system, we have to be a little bit more mindful of what the actual chipset was that was going on like on the UAV, right? That's slightly different. So you need to start to out a little bit of that more. But I didn't have to understand that as much. It was more the guys creating the chips. For example, in Australia, there's a company called Zealand, and they basically make. They basically leverage some

Harpreet: [00:29:23] Neuromorphic computing chips

Speaker3: [00:29:25] And stuff like

Harpreet: [00:29:25] That. I'm not fully clear on what they do,

Speaker3: [00:29:27] But they make vision specific hardware and that embedded hardware basically takes whatever Tensorflow or PyTorch machine learning models I've created and can basically

Harpreet: [00:29:36] Run that in a more efficient

Speaker3: [00:29:38] Manner. It's the same thing that Nvidia is doing with Tensorflow, RT and the Jetson packages and stuff like that. So the guys creating those the software layers between that specific embedded hardware and what I'm doing with my Python models, they're probably the people that really need to understand the depths of how far the compiler impacts the performance [00:30:00] of your network, right? So we're starting to see that separation

Harpreet: [00:30:03] Of skills like if you

Speaker3: [00:30:04] Rewind two years ago, it was kind of like, Hey, who understands embedded and understands

Harpreet: [00:30:08] A bit about ML? I'm sure you can do the job, but

Speaker3: [00:30:11] We're starting to see people specialize a little bit more where

Harpreet: [00:30:14] People like I'm really good

Speaker3: [00:30:15] At embedded programing for the vision space, designing for the ML pipelines, and you're seeing a couple of companies come up with that. And obviously, like Makiko said, your self-driving cars, the self-driving industry is really driving that in a big way. So we're starting to see that kind of split off as another kind of niche set of skills. But yeah, I mean, I don't know whether most Data

Harpreet: [00:30:37] Scientists or male engineers would

Speaker3: [00:30:39] Deal with it at that depth.

Speaker5: [00:30:42] But would you say that your your knowledge of the equipment available, ICU, you take you take cloud out of the way because cloud most likely will take care of that for you? What cloud can also hook you up with a hefty cost because of the processing on their platform, right? So if you could, would you say that knowing understanding the capacity from a hardware perspective will influence what kind of models you would select to build for your use cases and deploy if cost was something that you really worried about and as a constraint? I think the understanding what the hardware does can influence how you deploy what you deploy, right?

Speaker3: [00:31:28] Yeah, look, I mean, I'm sure it will, but at this stage, the maturity that the industry is that it's less a cost constraint. I'm talking about self-driving and autonomous robotics here. Right. That industry is so young and

Harpreet: [00:31:42] So inherently

Speaker3: [00:31:44] Expensive to begin with anyway that it's less of a cost constraint exercise a more of it. Do we even have the capabilities yet kind of exercise like we're still trying to get object

Harpreet: [00:31:52] Detectors that can deal with two

Speaker3: [00:31:54] Hundred and fifty objects in the image in real time at 60 frames per second, right? And still have [00:32:00] the kind of model accuracy that you need to keep people safe. So the first thing is, let's get to that and then we look at cost optimization. So yeah, obviously it will eventually come into a cost

Harpreet: [00:32:09] Question into whether you go

Speaker3: [00:32:10] For like whether you need to know that or whether you rely on abstractions like in GCP. And it's still unclear. The bigger reason that we don't use GCP as often on mobile robotics is, for example, if I put a robot under water, you're not getting into that.

Harpreet: [00:32:26] Plain and simple, you're just not getting

Speaker3: [00:32:27] Into that, so you need to rely on being able to use your local GPU or use a tax to or some other kind of specialized hardware

Harpreet: [00:32:35] For it, right?

Speaker3: [00:32:36] So that's kind of where the decision point is at right now, but that'll change as the industry matures. Right, so. Cool.

Harpreet: [00:32:44] Joe, go for it also, that was the stuff is going way over my head. Definitely got left to

Speaker5: [00:32:51] Compilers are good for people to

Speaker7: [00:32:53] Know if you're touching code. I don't think it hurts, like in Python, for example, right? Understanding how byte code is created and executed. Like that's that doesn't hurt you, right? And like, when is it appropriate to use C to drop down to C for primitives or to use like the JVM, a Java virtual machine? So I think these lower level things are something that Data scientists eventually should know. I don't know that it's necessarily like knowing compilers isn't going to make you like a data scientist out of the gate, but I think it's going to make you a far better one, for sure. And just knowing how your code should execute, especially in constrained environments, kind of talk about. So yeah, I don't know a certain point. I think it's

Harpreet: [00:33:35] Just certain

Speaker7: [00:33:36] Things that you should know in general, just basic computer science and algorithms and how that translates into byte code, for example, right?

Speaker3: [00:33:43] And even diving into just how. How does the

Harpreet: [00:33:45] Assembly work, right?

Speaker7: [00:33:47] You know, the primitive language for machine code. I don't think I've heard you at the end of the Data know how instruction sets are. You know how to perform. You know, you write Python docs out of Python Byte code that it drops [00:34:00] down to the assembly. How does that work? That, to me, is it's it won't hurt you. It's again, it's it's not something going to come across your day to day job, but it's sort of like understanding why do doctors spend, you know, spend so much time understanding anatomy. It's not going to come across stuff every day, but you kind of do.

Speaker5: [00:34:17] So to me, it's kind of like it seems like it's a good case use case for ML engineers, right? So if they're trying to run their processing on the edge edge devices and things like that, so it looks like they have they need to have an understanding of what device to use or what. I'm sorry what hardware to use on that device for compatibility. I think I think that article was very helpful for me, that chip you wrote. So thanks for your responses. I know you got some more I'd like to hear.

Speaker6: [00:34:55] Well, it's just it's funny like it, depending on what teams or organizations you talk to you saying, like the the statement that. Well, look, the question of how what skills should how engineering heavy should scientists be and what skills constitute engineering is an oddly controversial discussion I found in a couple of circles because yeah, it's funny. We're we're kind of coming up against something like in some places, conversation about like testing and testing standards and things like that. Apparently, people have very, very strong opinions as to who that belongs to. So I imagine if in those same groups, if we were like, you know, it'd be helpful if you learn like some more lower level stuff would just be madness for. Blood shed tears.

Speaker3: [00:35:55] Just to button that for a second. How much do you reckon that is more due to the history [00:36:00] of the actual skill sets belonging to people? And how much

Harpreet: [00:36:04] Is it more belonging to

Speaker3: [00:36:05] The way we perceive organizational structure?

Harpreet: [00:36:07] Like we need

Speaker3: [00:36:08] To have an organizational

Harpreet: [00:36:09] Structure where there's the test

Speaker3: [00:36:11] Engineering group Ml guys and the infrastructure guys versus can there be a more fluid understanding of how we operate with different skill sets?

Speaker6: [00:36:21] And I'd actually love to hear Joe's take on this, because I feel like in some so years ago, right, we had the like the full stack Data science unicorn. And there is this idea that regardless that like the only real data scientist was one who was like into and had to know like a hundred million things. And and they were the only people who deserve to be Data scientists. Right in that almost was a paradigm like, you know what, like just five, ten years ago. I mean, it wasn't that long ago, right? And I feel like we went from that to like, OK, everything should be like hyper specialized and especially in bigger companies, you just can't get away from, like, hyper specialize. Right. But I feel like there should be something a little bit more in between. And I'm not sure why. In some places, it's just so volatile.

Speaker7: [00:37:17] There's a lot of gatekeeping, though, right? Organizations, people want to protect their turf, and it just kind of how it is always used to make the joke like a data scientist like the crossfitters of Data pretty full stack ones. Or they're just kind of mediocre at everything like. But it's changing. I think to your point, specialization does occur now. But it depends on Data maturity of a company, right? So. But I think, yeah, it doesn't hurt to know, I think the primitives of each area of Data really, I think it only makes you better. So understanding if you're Data engineer, you should understand databases work, for example, at a good level like that. I just think that's necessary. But. Um, it's a tricky one. [00:38:00] I don't know that it's going to be solved any time soon, unfortunately.

Harpreet: [00:38:03] Maybe you'll solve it here tonight, who knows, but kind

Speaker5: [00:38:05] Of doubt it. What's what pushed you to get that book? Michael, like you said, you said in the comment you just purchased a book about compilers. What was the pusher there? Yeah, so,

Speaker6: [00:38:20] You know, because I'm not already just slamming my brain with a bunch of stuff, right, with the GCP studying and trying to get better at Jenkins and Kubernetes, right? Because I don't I don't have enough stuff I'm shoving in my brain. I think part of part of it was like a recognition that. And this is really funny, because at some companies write them on during function is really Data scientists plus. And in other places, it's it's considered engineering, and they're like their engineers that can happen to talk shop IMO. Right? And I think for me, I made a very. I made what I felt like was a very informed decision away from Data science to focus more on the engineering side, but I also recognize that like I have a lot of gaps in my skill set, which come up in like really weird ways. So for example, like we maintain essentially inside inside our developer tooling for our Data sites, right, so I can speak to them very well. But when something comes up in our tooling and I'm troubleshooting and then it's like, Oh, well, there is this dependency issue that then impacted like the environment that we were using in all these other things like I do find that I can follow along. But then I end up having to. I don't quite have like the conceptual underpinnings. And so I'm like, well, you know, my company has a really. Really generous educational stipend, so [00:40:00] and that and also recognizing that I haven't I've never really taken any sort of formal computer science courses. I kind of realized that I really need to reskill or upskill to be on how you look at it. So there's this wonderful website called Teach yourself Cars.com. And they have like a bunch of these recommendations for what are considered canonical, like textbooks and resources. So there's a list of books they recommended computer networking, the algorithm design, manual structure, interpretation, structure and interpretation of computer programs thinking, high level writing, low level crafting interpreters. Computer system, sorry, they're all like this table right here.

Harpreet: [00:40:40] So it was just a

Speaker6: [00:40:41] Recognition that like, OK, like I, I need to have my fundamentals and I also need to understand kind of how sort of technology works right now. In order to continue growing in my career as an engineer, as a data scientist, I feel like I feel like I probably could have gone away with not hitting them all that stuff for at least a few years. But as an engineer, I feel like it will actively kind of hurt my career progression if I don't. That's all.

Speaker3: [00:41:11] How much does that kind of depend on the team that you build around? I mean, this kind of goes back to what I was saying earlier, right? Like for me, for example, I yeah, I'm directing more towards the engineering role because I've

Harpreet: [00:41:23] Got a background with the systems

Speaker3: [00:41:25] Design and thinking of, you know, very weird and complicated systems, right? That's what you get when you deal with robotics. So I'm more comfortable with that than with the deep math of being a pure data scientist. So I'm earning towards that engineering side myself. And like you said, I'm like way behind in all of this stuff, so I'm trying to crash course everything at the same time, but it's part of building the

Harpreet: [00:41:46] Team around us. So right

Speaker3: [00:41:47] Now I'm working in a team of three

Harpreet: [00:41:49] And that's letting me.

Speaker3: [00:41:50] Basically, there's this multiplicative effect, right, where we've got one person who's very deeply into the mathematics of the data science side of it and then combine that with my more systems thinking kind of [00:42:00] approach. Together, we're able to both of us do whatever level of specific model development that we need to do, whatever level of systems design that we need to do. And it's not just that he does all the data science and I do all the systems thinking we swap and trade skills in that manner.

Harpreet: [00:42:15] So as a team,

Speaker3: [00:42:16] You're kind of building and you know, you've got to build out a

Harpreet: [00:42:18] Team in a way that I don't need to

Speaker3: [00:42:20] Be an expert at the data science and the modeling part of it in order to do it. But I can get the support where I need to, and that's actually important, right? So again, like and maybe I'm lucky and I'm in a company in a team

Harpreet: [00:42:33] Where there's a bit more of

Speaker3: [00:42:34] License for fluidity. But yeah, that's pretty essential, I think.

Speaker6: [00:42:41] Yeah, I guess a similar deal like we have people on our team who are like Docker, Kubernetes, Jenkins experts. And so in some ways, like because they are experts, it allows me to take the extra time to go through the materials and like upskill. Because if we didn't have that diversification of skills, like we have some people who are what you would call sort of technical specialists like they are very deep in building the tools, big brains and years of experience. And like literally, they can look at like an error that spit out of my log and they can go like, Oh yeah, I think that like it was, and it's quickly Google search and go like, Yeah, I think basically their newest release of PIP, like, broke it and I'm like, Oh, you got that out of all of that, like all that red, that's what you got all of it. And then you have some people on the team who are like much more like, I think, strategic. And they like planning. They like figuring out, like, where do the different projects and initiatives fit? And they tend to speak very well with our adjacent teams.

Speaker6: [00:43:42] And then you have people like me where I'm like, Oh, well, OK, like, I got to figure out everything you know, but like, that's what having like a nicely balanced like portfolio of skills like on your team kind of enables. And it does depend on like what is the core function of the team, you know, like what's kind of the North Star that your [00:44:00] team is providing? I will say because if your North Star is like. Um, I meant to be like more internal business facing, for example, you're trying to drive strategic business decisions. Um, having a bunch of engineers that are really deep into. The CS1 Mills, it might it might help, but frankly might be better just to find someone who loves living in the Strategy Analytics space. But if you are, for example, you're building custom tooling for like challenging problems like self-driving cars that you can't just use, like the current abstractions that are available. You really want to invest into those like deep, deep technical skill sets and people with those skill sets.

Speaker3: [00:44:45] So yeah, there's I think there's this. We've kind of over-fished across the last

Harpreet: [00:44:50] Century and this is an ongoing

Speaker3: [00:44:52] Hypothesis of mine personally that I've been trying to test out in the last five years that in the previous century, we kind of over-fished on this idea of the the master

Harpreet: [00:45:02] Of one

Speaker3: [00:45:03] Trade. You know, there was always that saying, you know, Jack of all trades, master of none. You know, you don't want to be the jack of all trades. I think there's room now to expand that thinking a little bit. And if you go back to like, I'm talking like Socrates level errors, right? Some of the greatest thinkers were polymath, and you don't even go

Harpreet: [00:45:21] That far back,

Speaker3: [00:45:22] Right? You go back to Newton. He's a polymath, right? These guys were writing

Harpreet: [00:45:25] Papers in many, many

Speaker3: [00:45:27] Different fields. I'm actually reading a book on called Polymath by Peter Hollen's, which is all about mastering multiple disciplines, right?

Harpreet: [00:45:34] So there's I think

Speaker3: [00:45:35] There's a resurgence of ideas saying that you can have

Harpreet: [00:45:38] People who, you know, wheeler dealers that can deal with

Speaker3: [00:45:42] The business side of things and deal with strategic can deal with a little bit of technical depth. Right. There might be value in having a balance of,

Harpreet: [00:45:51] Like you said, technical specialists

Speaker3: [00:45:52] That can deliver that code value but can think strategically and,

Harpreet: [00:45:56] You know,

Speaker3: [00:45:57] Work outside of that team. And this is why we've got to move beyond having, [00:46:00] Oh, I'm trying to hire this one rock star data scientist is going to solve my whole business. Right? So it's kind of leading me to this thought. That is data science inherently from a business analytics perspective, inherently better as something that you would hire out to a company or to like a consulting service as opposed to necessarily build in-house because it's a difficult team to build in-house, right? So can you leverage it better as a consulting team?

Harpreet: [00:46:27] And go for it.

Harpreet: [00:46:28] And also, you

Harpreet: [00:46:30] Share a link to that book or just type the name of that book out.

Speaker4: [00:46:35] And not to go too far down the rabbit hole of the of the mastery versus a jack of all trades thing, but that's something I think about quite a bit. And I think in twenty twenty one or in today's age like composite specialization is something that rules supreme. So it's not I'm not the best at this individual thing. Like, let's just take like Data science, for example, right? Like, I am probably not even like a seventy fifth percentile Data scientist, right? But if I combine that with my abilities in content creation or video editing, I am probably a top five percent person there. If you combine that with my sports analytics knowledge between those three domains, I'm like top one percent, right? I might be the only one doing those things, but that also suggests that I'm I'm unique and and can create value in that way for someone who has those specific questions. And I think rather than really like narrowing down and specializing in like a specific like technical domain. If we're spreading specialization over a composite group of unique things, we're able to create that differentiation and create the real value in that domain. So it's like we're becoming more specialized, but we're also less specialized because we have these composite skills [00:48:00] and it's like this. I wouldn't say it's a paradox, but it's like a weird,

Harpreet: [00:48:05] A weird refining of that

Speaker4: [00:48:07] Thought process, right? And it doesn't fit neatly

Harpreet: [00:48:10] Into a mastery

Speaker4: [00:48:11] Mindset or a jack of all trades mindset. I think that that the optimal skill set going forward is you're really good at like one or two things and you combine those with other things that you're pretty good at and that makes you valuable to as many people as possible. I still

Harpreet: [00:48:28] Believe that like if if you have

Speaker4: [00:48:29] Some expertize, let's say you're like an 80th percentile and something, and you can combine that with a bunch of other 50th or 60th percentile skills that gives you a lot of scale and a lot of specialization still to maximize on on both of those things. So maybe a little bit different way of thinking about it. Maybe I'm just rambling, but I just love that idea in general. I know Harp and I talked a lot about this too.

Harpreet: [00:48:55] Yeah. So like the rest of that quote, a jack of all trades, master of none, like the second half of it, is oftentimes better than master of one. So that's that talent stacking thing, right, you just, you know, be top twenty five percent in a few different things, combined them in unique ways. All a sudden, you're the best in the world at being yourself. Yes, you know, kind of the direction you want to head towards.

Harpreet: [00:49:22] Greg, great conversation kicked off there. Let's head to Mark's question.

Speaker5: [00:49:28] Actually, I think it ties in really well to the previous conversation. I'm just more of a thing forward, kind of like, you know, next career moves, not necessarily new new company, but more so like where what type of Data role do I want moving forward and specifically moving away from from data science? I've realizing that I love being I see I love being technical, but like analytics and like machine learning.

Speaker3: [00:49:54] Model building just as a career does not, does not provide me joy. I like it for side [00:50:00] projects, but for for the job itself.

Speaker5: [00:50:03] I find it to be way too reactive for my liking. And so trying to think like, what's brainstorming, I'm stuck between Data engineer and Ml engineer as roles, but there are two different paths. Just try and choose one as go go for it. And so essentially, you know, just want to brainstorm. Like what? Where are some career paths beyond Data science that really lean into that technical side?

Harpreet: [00:50:28] That's a great question. It's something I was thinking about earlier today, too is like. I can write code, I get done, but like I don't really enjoy sitting there, writing code all day long or doing like grunt work, type of stuff like I like kind of architecting a solution, right? Like if I could just talk to like some junior level data scientists like I could, we got this Data here, some things that we need to try to do with it. And here's kind of a blueprint and path

Harpreet: [00:50:52] Of things that we should do. Get it right

Harpreet: [00:50:54] Up to the experiment point. You do all the coding up to there, not take it from there. I'll do the I'll do the

Harpreet: [00:50:58] Experiments then and find

Harpreet: [00:51:00] The best fit model and evaluate it and stuff. So like, that's kind of I was having that same thought earlier today. I was like, man, like. I enjoy Data sites, but I enjoy all aspects of it, particularly grunt work, building type of stuff. So I'm very keen on hearing other people's

Harpreet: [00:51:16] Responses here as well.

Harpreet: [00:51:17] But slowly and surely, people are dropping off. I'd love to hear from a listener. Let's go to Ken and then Greg on this than anybody else wants to jump in Mexico. And yeah.

Speaker4: [00:51:35] Yeah, I think, Mark, that's a really good question. I am neither a Data engineer or an

Harpreet: [00:51:39] Ml engineer, so I don't think

Speaker4: [00:51:40] I can provide too much insight on that front. I will say also it can be

Speaker5: [00:51:46] Mellinger Data engineers, the ones I'm aware of. But you can think of, like some wild other ones I'm not aware of, two does have to be those two two buckets. Yeah, I

Speaker4: [00:51:55] Mean. To be perfectly honest, I do like a self [00:52:00] assessment of

Harpreet: [00:52:03] About once a month.

Speaker4: [00:52:04] Sometimes I push it off to a once a quarter where I evaluate where I think my strengths are.

Harpreet: [00:52:10] But I also

Speaker4: [00:52:11] Evaluate my enjoyment of certain activities. I try and like audit essentially everything I do over the course of a week or a month, and I rate how much I enjoy doing those things. I rate how much like effort it takes, how much energy it takes, a lot of different ways that I frame the same thing. And then I try to match that to the roles that I'm currently in. Like what roles would allow me to do

Harpreet: [00:52:38] To make to be

Speaker4: [00:52:39] As efficient and effective as possible? While it not feeling like work for me. I think the more I'm realizing, that's like, let's take the content creation that I do, right?

Harpreet: [00:52:48] I find that nowadays making

Speaker4: [00:52:50] Youtube videos for me is a little bit harder than than filming podcasts. I love talking with people. It's something that feels completely

Harpreet: [00:52:58] Effortless for me.

Speaker4: [00:52:59] When I have a conversation, I just,

Harpreet: [00:53:01] You know, like connecting with another

Speaker4: [00:53:03] Person. That's dynamic. It's it's it's something that flows.

Harpreet: [00:53:06] Rather than making a video, it's pushing

Speaker4: [00:53:08] Like it's it's static. I am doing all of the delivery right and thinking about, OK, like, I still enjoy both. But how do I slant things? So I'm doing more of one of those other things. And that's something I have complete control over, right? But the next logical step, if it

Harpreet: [00:53:24] Was my domain or my

Speaker4: [00:53:26] Work is to say, OK, what types of activities or responsibilities can I pick up that would flex those

Harpreet: [00:53:34] Exact skills?

Speaker4: [00:53:36] And you don't, you know, if you love your company and your management team is receptive, in theory, you should be able to like, just evolve your role to doing more of the things that really feel effortless to you and doing less of the things that you find tedious or struggling through. And rather than looking at it as like, Hey, I'm going to switch, it's like, I'm going to take what I'm currently doing and slowly progress [00:54:00] it towards this other thing. And you can, you know, you can get any title you can do whatever you want. But you know, that's how I would generally approach that situation. Not not like clear cut advice or anything like you probably wanted, but that's sort of my my two cents on that front.

Speaker5: [00:54:20] Greg? Yeah, so I definitely like your your framework, and for me, it's about more where do you see yourself sitting in an organization? I like to think about it in terms of front end of an organization or backing of an organization, and you can have different positions inside of it. Most of the time, the back end, you're building products or you're taking care of infrastructures that serves or enables or enable internal teams. So do you want to be a builder of these infrastructures? You want to build products that serve internal teams versus do you want to build products that serve external customers? So there are multiple places you could go you could become a product and product manager. You can be a systems of architect that builds these systems, that looks at the big overview of these pipelines and kind of like an internal consultant in around these systems. There are definitely different things, but the first thing is, where do you want to be? Because you can be a solutions architect

Harpreet: [00:55:27] For

Speaker5: [00:55:29] External customers, or you could be a solutions architect internally that are looking at ecosystems that are enabling internal teams. So, you know, if you want to go beyond just the data science of things, you'd have to take a look at where you want to sit inside of our organization and see what kind of problems they're solving, what kind of tools they have to solve these problems. And then how can you apply your skills, which is mostly around data science to solve these issues? And then you can figure [00:56:00] it out, right? Most important, you're young. Try as much as you can because time is your best commodity, man. You're better placed than most people right now and the time to try different things until you figure out what you like, what you don't like is now.

Harpreet: [00:56:16] Let's let's go to Mexico, then coast up and then like, yeah, just that, I just googled Data science solution architect that is an actual job title. There's a bunch of job postings on this really interesting to look into. Let's go to Mexico and then. Anybody else has questions? Let me know. Not a lot of people joining us on LinkedIn. Not a lot of questions coming in here. So this might be the.

Speaker6: [00:56:42] Yeah, so when I had like the I talked to, like my past director, my team, so to complicate the plot, my team Mladic got moved into the Data and Jorg, so complicate the plot or or the Data services, but it includes a bunch of Data engineer. So but when I have the talk, so I had a conversation with two directors, my current director and the past director, who is still in the company, but whose team I or I moved out from under. And one director basically said, Think of it in terms of where you fit along sort of the spectrum from like Data, from tech specialists to tech lead to tech strategists because even within an engineering organization, there is a really wide variety of work that people do. You have some people and this is true even at certain levels, like, for example, you can have a staff or a principal engineer who like is a tech specialist, like they build all the tooling inside. They do the research they, you know, whatever. But you can also have principal engineers who are mostly tech strategists. So they're focusing on like, you know what? What tool stack do we incorporate if, for example, [00:58:00] if

Harpreet: [00:58:00] You're going cloud like

Speaker6: [00:58:01] Gcp and let's say you start out in a monolith, like what? Which of those services do we do we invest in and do we move out of? How do we serve, communicate the company's strategic positioning to like external users or something? So there's even with an engineering, there's a very, very and the technical contributor route. There's a wide variety of work that you could do for me personally, I don't see myself as a tech specialist. Frankly, to some degree, I kind of feel like I started out my engineering career a little bit too late. Um, like really only within the last two years, and I'm like, you know, older than 30 now. So in some ways, I kind of feel like if I had wanted to go like the Jeff Dean route, I probably should have started studying computer science and undergrad. Um, which I'm fine with for me personally, I see myself as like more of a tech lead tech strategist, I'm very, very interested in how a design systems and I'm very interested in like the trade offs between when do we each use some solutions versus others? And I love building products. That's what it comes down to. I really hate getting into the nitty gritty of like, what libraries do we use like debugging stuff. It's it's just a necessary evil to me. I'm not like in love with trying to come up with new languages. Using new languages. Trying out new languages all the time. For me, all my work in all my investment is can I build stuff that people can interact with, whether they're internal or external?

Harpreet: [00:59:38] And so for me that it it makes

Speaker6: [00:59:40] Sense to align with like the tech lead and tech strategist, the very similar. Similarly, like my other director basically was like, you know, said you should be like in people management because they were like, you have an innate ability to lead people and to get them to do things even when you don't want to do it. So. And his advice was very similar. He's like, you're young. Like, he's like, [01:00:00] feel free to bounce in between stuff like try different things because at some some points you're going to want to like, do technical projects, but other points in time, you know, it's like, you're not. He's like, you're not going to deal that bullshit. You're going to want to like, help enable other people and other teams to do really big stuff because at the end of the day, really, really big, like moonshot goals, you have to have teams like maybe you have like three to five 10x engineers, if you can afford it. I mean, but realistically, if you want to get big stuff done, a lot of times you're not building yourself, you're you're going to have like a team. And so for some people, they have more joy. It seems like in rather than saying like, I built this feature myself, they can say I built this platform or solution or I'm not I built. I helped enable the deployment and the creation of this solution by gathering like the best people around me and like making sure that they are developed. So, yeah, so in terms of like in terms of the Data ML and kind of email ops kind of thing, to be honest, it seems like it's the division is a little less. There's not as much of a division as you would think if you develop skills along the and general off site. It probably you can transfer a good amount of it over to the Data and inside and vice versa.

Harpreet: [01:01:28] Those to go for it. And I guess. Sorry, go. I was going to say, and I would love to get Ben's input on this as well so marketers can type

Harpreet: [01:01:38] Out the question he had

Harpreet: [01:01:40] In the chance of Ben is cued up on it. Go for it.

Speaker3: [01:01:44] Yeah, look, I mean, I'm still trying to figure out exactly what my path

Harpreet: [01:01:48] Is as a is an ML

Speaker3: [01:01:50] Engineer, a robotics engineer computer, I

Harpreet: [01:01:53] Don't even know, right? But like so it's

Speaker3: [01:01:55] For me, it's I could go down as a technical specialist, but I know that I also

Harpreet: [01:01:59] Have some [01:02:00] skills and

Speaker3: [01:02:00] Enjoyment in

Harpreet: [01:02:02] A strategic

Speaker3: [01:02:03] Thinking and be thinking about systems at a higher

Harpreet: [01:02:05] Level and as

Speaker3: [01:02:06] Well as, you know, just leading a team like we seem

Harpreet: [01:02:09] To constantly

Speaker3: [01:02:10] Talk about this two kinds of data scientists,

Harpreet: [01:02:12] One which is, you know, your deep

Speaker3: [01:02:14] Technical

Harpreet: [01:02:14] Specialist and your business

Speaker3: [01:02:15] Strategist, right? What about the people leaders? As we start talking more and more about teams, you need people that can lead people but still understand Data science, right?

Harpreet: [01:02:25] Like the best

Speaker3: [01:02:26] Product managers I've ever worked with. And when we were building,

Harpreet: [01:02:29] Like electromechanical devices, were electronics engineers who understand

Speaker3: [01:02:33] The pain

Harpreet: [01:02:34] That the

Speaker3: [01:02:35] Mechanical team of the electrical team are going

Harpreet: [01:02:37] Through.

Speaker3: [01:02:38] So your people, managers and your project managers, there's all of these different roles that are coming into the into the ML, I guess, ecosystem or AI development ecosystem. And really, we tend to overcorrect and

Harpreet: [01:02:52] Kind of say, Hey, I am

Speaker3: [01:02:54] This career because this is my like training. This is my degree in this and that right? But that's not the sum of all parts, right? The sum of all parts is what are your other experiences? What are what are the other things that you've done and your natural skills, your just your personal tendencies. So I I can't remember, I think it was Ken that said he really analyzes his skills as a whole, right? You've got to come up with that Venn diagram and it's something that I'm I revisit that at least once a year, if not once, every six months, and really look at what's changed since the last Venn diagram.

Harpreet: [01:03:24] And that's how I'm hoping

Speaker3: [01:03:26] That that'll inform me going forward in my career and where I can head with that. It's a way I don't know if it's the right way. I'm not at the end of the path yet. Not even close.

Harpreet: [01:03:36] But yeah, it's really

Speaker3: [01:03:38] About understanding where

Harpreet: [01:03:39] All of that fits, right?

Speaker3: [01:03:40] So there's more than just, Hey, I need to be the technical specialist or going to be this particular thing that comes back down to that pigeonholing thing, right? We need to be a bit more fluid.

Harpreet: [01:03:50] Yeah, I mean, it's interesting, like like this rolling out now at Comet, how it just kind of. Seem to coincide so nicely with kind of my specific knowledge and what it is that I'm [01:04:00] interested in, right? Like like I'm not necessarily like the business data scientist, have a guy like, I don't really care about reporting the stakeholders and doing business stuff and stuff like that. I care more about the creative aspects of Data AIs building models, doing interesting stuff, educating people, communicating data science and and things like that.

Harpreet: [01:04:18] And I seem to have found myself a position

Harpreet: [01:04:20] Where I get to. Old. Data sized bottles write about them. Presentations, communicate, educate and do a lot of creative work while still being. Closely connected to Data scientists and Data datasets working it, you know, I guess, try to find opportunities like that. Roles like this are coming up. If this is kind of like the path you're interested in, like and they tend to be mostly on growth teams or marketing teams like. It's like Advocate is, you know, public evangelist, email advocate or developer advocate, things like this. But I mean, it's all about just finding interesting intersections of unique skill sets. But Ben, go for it. I'd love to hear from you on this

Speaker3: [01:05:05] On the Mark's career path beyond Data science. Yeah. Ok, so I'll I'll start with something negative and end with something positive. So the negative negativity is jobs are not charities. Jobs are not free.

Harpreet: [01:05:19] And having paid payroll,

Speaker3: [01:05:21] You need to have a multiplier and you have a multiplier value. But the positive side is you can actually make up your job like you can make up your job as long as that's true, as long as there's a multiplier, there's that. Attribution is true of every department sales, marketing and data scientists. Really good data science leaders know their attribution, and I see this firsthand. Really good data science leaders will say this is my attribution for my entire org this year, and it's a number with the dollar sign. And so my first year joined Data robot. I don't think I coded like a full year, which is maybe a little weird because I came from very high performance, deep learning, coding all the time, every day. So really, [01:06:00] it's just kind of blending your passions. And I think for some of the people here on the call, they like podcasting. They like speaking.

Harpreet: [01:06:07] They like blogging.

Speaker3: [01:06:09] They like writing. So just I think there's always going to be an intersection of AI with whatever your core passion is and whatever you do, just find something unique and really push into it, and you can make up a role that we've

Harpreet: [01:06:20] Never heard of. But it has some

Speaker3: [01:06:21] A.i. flair to it as long as there's attribution. To that role, which can be there for so many reasons. Yeah. Anyway, I loved Greg's comment on J.R. Brain Show. Our brains been breaking my brain lately anyway.

Speaker6: [01:06:39] Yeah, it's funny. Laurence Maroney over at Google, he's I think he's one of the lead Tensorflow advocates or something. I forgot what exactly how high up he is, but like some people, I've seen it, especially like at Google or Amazon, where they're like developer advocates. They have some of the coolest jobs because like they're like there, it seems like their entire. Public time is they use like either Amazon or Google's tools to build cool shit and go look at all the cool shit you can build using these tools. They do a bunch of projects and tend to talk about how cool the tool and technology is. Publish like content around it, but they they literally get to see projects from like end to end. I feel like in some engineering organizations, you kind of just live in like one of the stages of the pipeline. So if you're shy, you just Q8 everything. If you're ML and you live in some part of your Data and you live in some part, but a lot like the developer advocates or even like the evangelists. It just they can actually take a project to production end to end. They have that creative autonomy to do that because they're their showcase like their companies tools. So to me, I'm like, if I if I get to the point where I'm like, I just don't want to even anymore, I feel like that would be such a cool place to try out. Cool, [01:08:00] such a cool role to try out.

Harpreet: [01:08:03] Yeah. Well, maybe to lean on that real quick. Mexico, I've

Speaker3: [01:08:06] Been I've been teasing other people in the Data science community because I'll send them a text and say, Data science is

Harpreet: [01:08:12] Hard. I'm smoking

Speaker3: [01:08:14] Cigars in

Harpreet: [01:08:14] D.c. or like,

Speaker3: [01:08:16] I'm drinking wine over here. But this is always focused on like partnerships, right? So getting back to attribution. There are so many interesting. So we talk about selling like peer to peer networking. But there's also like business networking, where it's quite complicated. We're trying to prioritize these partnerships. And so for me, that is really, really fun. But it has Data science like data centers all over that. Like, how does a partnership actually work? How does it make sense? What are the integrations? But these all started like human level.

Speaker6: [01:08:46] I will say, like one of my one of the perks I enjoyed working as like a data scientist on like the customer success side was getting all the perks of the customer success team, right? And especially to if you're like a client facing data scientists, right? Like they'll bring you to the meetings, they'll bring you to the kickoff, so bring you to the sales summit. You know, you sometimes that's stressful to like in especially like technical sales, right? Because if you're actually one place I've heard people go to is like sales engineering. That is actually one place I know some data science to go to. They go to sales, engineering or like customer customer success engineering because they like

Harpreet: [01:09:26] The like nine to

Speaker6: [01:09:27] Five aspect of it. And or a lot of times when they do work, it's like around the end of the quarter or like around like the sales peaks, but their entire thing is still once again like building stuff, making it look cool, but working as a day of science, as a as a data scientist on the customer success

Harpreet: [01:09:46] Side was

Speaker6: [01:09:47] Really fun. Yeah, I think I went to I got flown out to like Boston Vegas, Switzerland. Um, could have gone to, I think, England and Italy for Autodesk, [01:10:00] but I just

Harpreet: [01:10:01] Didn't want to, I didn't want to.

Speaker3: [01:10:04] So like, we should push the trend that hashtag Data sites, it's hard any time you're like drinking wine or an African safari, if it's tied to the job, that would be hilarious just whenever the junior Data scientist

Harpreet: [01:10:16] Would think, wait a second

Speaker3: [01:10:18] If that's not data science.

Speaker6: [01:10:20] Oh, yeah, I did that when I went to Tel Aviv for work, that was great. The beach on Tel Aviv, going to the cafes like every on Instagram. Yeah, that was one.

Speaker5: [01:10:32] This is super helpful advice,

Speaker3: [01:10:33] And also, for context, a lot, a lot of my work and my things I try to go for is like, I view it

Speaker5: [01:10:38] As training for the next time, I try to do my running like a venture backed business and trying to build one. And so a lot of my focus now is like, how can I build really quick and iterate really quick? Because that's like the furthest I've gotten really in, like trying to approach those startups. But the other end of it is like, All right, cool. Once you build it, get funding and you grow will be need, build a team, and that's a huge gap in my learning. But I also hate the idea of management, so still still in the learning phase. But yeah, that's what I'm really focused on, just like building cool shit really fast.

Speaker3: [01:11:11] I'm a big fan of startups early on to only hire principal talent, so you actually don't have to manage. You hire people that know their domain. They are experts at it, and that's awesome. If you have to mentor and manage, then for an early

Harpreet: [01:11:25] Startup, that is awful.

Speaker3: [01:11:28] Yeah, just kind of like just kind of jumping on top of that, I know that I'm I'm kind of I'm not like fresh, but I'm also not nearly at principal talent levels, right? So my focus right now is actually on scaling. I'm trying to work in a team right now that I joined a few months ago where we're working on scaling data science models. So if I can build my niche across the rest of the Data pipeline, there's so many people focusing on how can I develop

Harpreet: [01:11:55] And and do

Speaker3: [01:11:55] These proof of concepts really quick. There is still this [01:12:00] gap in how do we scale that into a large scale product, right? There's not many people who know how to do that. So that's something that I'm trying to learn. It's a different set of skills that's within the Data,

Harpreet: [01:12:09] You know,

Speaker3: [01:12:11] Manifold of jobs and technical skills. But I kind of want to latch on something that Ben mentioned a bit earlier. You said basically, like end of the day, it's about having that attribution, right? That dollar tag figure of what value you're providing like. I don't know if I'm calculating that always the right way. How do you go about processing that and assessing what's the real value you're providing? As some

Harpreet: [01:12:37] Things, it's a hard science and then it

Speaker3: [01:12:39] Blurs into the gray. And so sales are super easy, like they can had attribution marketing as well, but there's things like brand equity or Data science. If you do some innovation, it ends up in the product. What is the value of that? And I think what you see in all these domains, there is a science where people are trying to hammer all the way down to hell on what that what the value is. And so if you go ship a feature, a very mature product would actually know how many customers are using. It wasn't mentioned in Salesforce as a differentiator during the sell. And people attribute it back. I'm not saying that's easy. A lot of companies really struggle with it. So, but it's also not just that I'm the only person creating

Harpreet: [01:13:19] That feature, right? I'm part of a

Speaker3: [01:13:20] Team creating feature, so you know, you've got to assess

Harpreet: [01:13:24] What's your you kind of

Speaker3: [01:13:26] Have to look at

Harpreet: [01:13:26] It at a I mean, I've

Speaker3: [01:13:28] Started to look at it more at

Harpreet: [01:13:29] A level of how are

Speaker3: [01:13:30] The things that I'm doing? Amplifying the speed at which my team can deliver that feature, which delivers this dollar figure value to the company. Right. And conversely, what am I doing outside of this specific role which is scaling this pipeline? What are the other things I'm doing that add value to the company as well? Yeah, that's a bit more of a mixed pot balance. Yeah, I think one of the general themes is how close are you to revenue generation in long term? Like if there's company

Harpreet: [01:13:58] Cuts or pushbacks,

Speaker3: [01:14:00] It [01:14:00] doesn't matter what role you're in. You kind of have a sense of how close you are. So, so if you can't, like, spit out a number immediately, the number should always be greater than one you should always have. You should always feel confident that there is a true multiplier to what you're doing. But some people, they are in roles

Harpreet: [01:14:15] Where they're very

Speaker3: [01:14:17] Far away from revenue, and that just becomes more problematic. I think I'm maybe I'm different. I think it's I like being as close to it as I can be, but also I like blue sky innovation that is maybe falls in the brand equity bucket, which is very messy. It's so hard to assign the value to brand equity. I guess the other side is that maybe this is something that, like younger data scientists like myself, need to be a little bit more aware of is what's the value? Can you tell me the value of the product you're actually building, right? Like, we should know that as a data scientist that, hey, we're building this this piece of software,

Harpreet: [01:14:55] It's going to earn this many million

Speaker3: [01:14:57] Dollars a year,

Harpreet: [01:14:58] Right?

Speaker3: [01:14:59] Or is it $100000 project or $50000 project

Harpreet: [01:15:02] If we don't know that?

Speaker3: [01:15:04] I think that's a critical thing to understand the value you're providing, and that is something that a lot of these engineering projects, they actually do give revenue commits. So for the bigger strategic projects, they don't just say, we're going to do this. It's important they say we're going to do this, here's a revenue commit. And if that revenue commit is not

Harpreet: [01:15:22] Met, then they will.

Speaker3: [01:15:24] They're your names attached to it. So and it's funny those revenue commits when you look at them, they're always nice big round numbers. Was like

Harpreet: [01:15:34] One,

Speaker3: [01:15:35] Two, three.

Harpreet: [01:15:36] It's never like two hundred

Speaker3: [01:15:38] Thousand three hundred and fifty or like two point six million. Like, it's not really a science, right? It's finger in the wind, but the accountability is there.

Speaker6: [01:15:46] It's usually divisible by five. And I think to like so I would so devil's advocate here, I would argue that if you have like leadership that at [01:16:00] all levels is variable aligned on what the North Star is, what the strategy is. As an individual contributor, you don't necessarily need to know what your revenue contribution is in that. If if. So, OK, so what you don't want in some ways is you don't want to be. Working on a sinking ship or like a a feature factory, right? Because with a feature factory, you can have features that do bring in extra revenue, but it's like incremental. It's it's almost like you don't know if it's actually pushing the needle in anything. It could be like an extra lift of. Like so for Amazon and Google level right and actually left of two percent, that's a lot of money. That's a huge amount of money for a company that's doing one million revenue, an incremental left of one to two to five percent. It might not be good enough depending on how expensive your engineering talent is. So I feel like but I think I think that's the hard part, right? Is that like at some point, especially as a company gets bigger, like you'll have teams, which they are just for whatever they're working on this product, which someone knows it's going to get sunset in like the a year or two after it gets launched. The engineers are not necessarily going to know it unless they are willing to kind of go up and go across and talk to the teams like next to them. Or don't talk like understand what the adjacent teams are doing and what the sort of overall roadmap is and where do they kind of fit in that? And ideally, I feel like if you have like a really strong leadership and it's perfect communication all the way down, most people would be able to track that, but haven't got to see that, frankly. Seems like it falls apart at some level.

Speaker3: [01:17:52] Kind of a problem of scale, though, right, like in a smaller team, like a mid-sized team, like four to 100 people, maybe that's very easily communicable. Right? [01:18:00] The moment you get a few thousand people in a company, there's competing strategies because you've got multiple business opportunities that you're chasing. You got in different business opportunities that are like ones already. You're already milking a cash cow here, but you're still you've got a rising star that you're developing, that it's just competing strategies, right? So it's hard to communicate that across a few thousand people.

Harpreet: [01:18:22] Article three, that was some good insight for you there. I want to go to to Greg because Greg had a question for Ben. So then we'll go from Greg, then we'll go to Gina's question that we'll call it a day after that. Thank you, everybody on LinkedIn. I'm sorry, I can't get to all of you guys questions, but hopefully you guys are enjoying this.

Speaker5: [01:18:40] Yeah, it's a quick question about your brain, like if you've seen any interesting answers in your post about this one, about that concept, if you want to explain to the audience what what you're working on there. Yeah, I

Speaker3: [01:18:54] Agi, I think is really fascinating for anyone. And I, it's kind of like this this I love how AI has the intersections of psychology, neurology, philosophy like it's kind of nutty that it all or you want to understand your life's purpose, like all of this kind of like blends in this really weird point and people talk about singularity. Agi in the Turing test was the first one, but the thing I really don't like is you have like GPT three, like you have these massive neural networks that will train more data than humans could ever learn. And and I like to focus on randomly initialized a AGI. What could I do? And so the what Greg is mentioning is I posted on LinkedIn

Harpreet: [01:19:33] And I proposed

Speaker3: [01:19:34] This problem. I said called it the AGI game, and I said, Look, if I could rebirth,

Harpreet: [01:19:39] You go back in

Speaker3: [01:19:40] Time, you're not going to learn to speak English and rebirth your brain in a jar and you can consume a video feed. That's it. That's video feed is going to run for twenty five years. I have an AGI. It's literally smarter than you

Harpreet: [01:19:51] Like, that is a fact, it is smarter than you.

Speaker3: [01:19:54] What can it do twenty five years later that you can't? And I think that's an interesting discussion. [01:20:00] Like, obviously things like search would be very trivial. Humans or memories aren't quite good. If I asked you if I showed you a frame or video sequence from three years ago, you would have a hard time knowing exactly in time where that was. Maybe you could place it.

Harpreet: [01:20:13] Maybe you can.

Speaker3: [01:20:14] So I just think it's a really fun discussion. There were some interesting questions. I think Craig kind of put me on the spot a little bit, whereas trying to recall who who had the most interesting answers. I know Greg, you responded. And I liked your some of the insights that you had. I'm trying to recall. Maybe, Greg, since you're on right now, what search would be top of mind predicting would be top of mind? Ais anticipating what's going to do, but just your human brain does it too. Like you're anticipating what I'm about to frog like your brain is

Harpreet: [01:20:44] Anticipating what I'm going to say next,

Speaker3: [01:20:46] And we're pretty good at it, but AGI would be a little bit better. Anticipating what is next, so if I had to plot intelligence? Intelligence is not singular, right? You have different dimensions of intelligence, emotional intelligence, et cetera, et cetera. But if I had to pretend like it was this continuous scale, think of it as being plotted on time and space. Your comprehension of time and space. If an AGI is smarter than you can predict further in the future, what someone might do or what might happen than you could, but it could also comprehend space better things around understanding what might happen. But I like that problem that I brought up because this AGI is not allowed

Harpreet: [01:21:22] To interact with the environment.

Speaker3: [01:21:25] And so it's actually putting AGI in the worst possible position to learn. I don't know, I'm just curious how anyone reacts to that, because people always freak out about AGI. It'll become infinitely smart in a month, and I'm like, Yeah, that's fine if it can do stuff. But what would it do if it literally can't your jar brains sitting next to the AGI jar brain?

Speaker5: [01:21:45] So you it's kind of like you're exposing AGI to to only that environment, right? That even AGI itself is constrained to whatever it sees on that video through the years, and we'll be able to apply brute force to predict [01:22:00] in a longer horizon than than the human brain, right? Our long short term memory is not reliable to perform those tasks. In this case, AGI will be a little bit smarter or much smarter, but there are some other things that humans experience that nature is not aware of.

Harpreet: [01:22:19] Will the AGI have an ego where if somebody in the kitchen drops a glass of water, my human brain might take a person like all this fuckers just fucking with me dropping a glass of water because I can't. Well, the AGI, I think.

Speaker3: [01:22:34] Right? I think with that limited input, it's really hard to have self-awareness like you can't look in the mirror, you can't touch your face. So you even as a human brain in a jar. You're just kind of anticipating what's next, like, oh, here's the humans again, you actually don't even know you're a human. If I could rebirth you in a jar. It's funny because people keep coming up with new Turing tests. I don't know if you guys have seen some of these there like, oh, the Turing test failed. So the new one is this, and I'm like, Oh my gosh, are you kidding me? I think one of the Turing test options out there is the new Turing test is AI will be able to comprehend anything in the sequence of videos. I'm like, That is the dumbest thing ever because I can brute force train. We will brute force train in the future and then everyone will celebrate. We passed the Turing test because you trained

Harpreet: [01:23:21] On like a gazillion

Speaker3: [01:23:23] Videos. So I hate the parlor trick. I don't want a parlor trick. I just want to simplify it to the simplest use case and work backwards. Let's actually talk about what this thing would do if you can talk about what it'll do. You can actually talk about how you build it. So if they would have, yeah,

Speaker5: [01:23:39] I have a good question. So, for example, to a point, Harpreet, I come in the kitchen. I drop something on the floor because I'm mad versus I made a mistake. I think the AGI will see that differently. We'll take that as two events of dropping something on the floor as a Data point, just like it, just as it [01:24:00] is versus a human brain. Would say, OK, there's a why behind it, right? Maybe the human brain will have more intuition into guessing whether the person is in a bad mood versus good mood. But now I just heard you say even the brain doesn't know it's a human brain. Now that brings a little bit more complexity to it now.

Speaker3: [01:24:21] So yeah, well, it is. You can give examples where you could. The example I throw out that the the general community maybe doesn't talk about that much is for that for those two systems. If I come down there at night because you can't see in, if I put a picture upside down or if I shrink the fridge or if I do something, if I change a number on the microwave now, the nines there, it's not there. Aggies can notice that immediately.

Harpreet: [01:24:49] As soon as it comes on BAM, the novelties,

Speaker3: [01:24:52] Their eggs are something different. The human would, too, but the human is more likely to miss subtle pixel level changes where AGI would immediately get it. Something's different,

Harpreet: [01:25:03] Like the human consciousness is a filtration mechanism, right? Like, we filter out more than we take in. That's otherwise we just. Wouldn't be able to handle it, all right, so. The machine wouldn't

Harpreet: [01:25:19] Let the machine handle everything,

Harpreet: [01:25:21] Processed everything, right, so yeah, I don't know where I was going with that, but. We feel I mean, we have to.

Speaker3: [01:25:26] But doesn't that kind of filter into the main issues of AGI, right?

Harpreet: [01:25:32] You're talking ability to represent

Speaker3: [01:25:35] The world at the same kind of fidelity that that we do in our brains, like we've got the kind of sensing, like the sensing resolution effectively for lack of a better word. Right? To be able to construct these high fidelity maps of the world around us in our brain, right? Like in this field of robotics

Harpreet: [01:25:54] That like the sensor

Speaker3: [01:25:55] Challenges around that alone are starting to add up, right? Then you've got the power consumption [01:26:00]

Harpreet: [01:26:00] Issue, right is sure they

Speaker3: [01:26:02] Might might be able to do

Harpreet: [01:26:03] Things to greater levels of

Speaker3: [01:26:05] Detail and

Harpreet: [01:26:05] Attention than a human

Speaker3: [01:26:07] Brain would over a long term, short term kind of inefficiency that we have built into us. But then where do we where do we come

Harpreet: [01:26:13] Across this point, where this is

Speaker3: [01:26:14] A case where a guy would really work because it's more like power efficient than also it's all about intelligent, efficient intelligence as well, right? And to your point, then

Harpreet: [01:26:26] You're right, it isn't like the latest

Speaker3: [01:26:28] Turing test that just, hey, can you do this task that's seemingly difficult for current day technology to do,

Harpreet: [01:26:34] But if you

Speaker3: [01:26:35] Really want to know about a Turing test, you

Harpreet: [01:26:37] Should be asking, Can

Speaker3: [01:26:38] It like us? Can it develop a sense of ego in the philosophical sense of ego where you're saying, I identify myself, you know, a sense of identification, right? That's when you start getting these higher levels of thinking. And that's the difference that you see between like

Harpreet: [01:26:52] Between

Speaker3: [01:26:53] Lower level ego

Harpreet: [01:26:55] Creatures

Speaker3: [01:26:55] In the world. Like, you see that difference between dolphins and monkeys and, you know, right down to insects, right? There's different levels of ego representation and understanding of who I am as a as a person, as a placebo, as as a human, we're able to see that we are this physical person.

Harpreet: [01:27:11] But then we're also this

Speaker3: [01:27:12] Observer behind that watching that physical person. And then we're an observer behind that, watching the observer right. We can see these layers of abstraction and symbolic learning. So and all of this, we do this taking what, eighteen hundred calories in a day? Right. So the significant challenge for something like AGI and my gut sense is how do we develop? How do we find those challenges that AGI would be apt for within the power constraints?

Harpreet: [01:27:40] That it would

Speaker3: [01:27:41] Just be unreasonable for us to just get humans to do it right? And do we then boil it down so far that it's just turned into a more specific

Harpreet: [01:27:49] Artificial intelligence and no longer

Speaker3: [01:27:51] Considered really AGI? It for the power

Harpreet: [01:27:56] Limits I I do like to go

Speaker3: [01:27:57] The opposite way and just gift it infinite [01:28:00] memory and infinite compute just for just for the thought experiment experiment, because I think as you get into it, like here's a great example. One milestone that would really impress me is to have an AI system that learns to speak English through experience like literally in your home and you have a baby. You have this AGI system, you talk to it, you interact with it. It's going to be focused, based learning. You'll smile, you'll you'll give the same cues

Harpreet: [01:28:25] That it's

Speaker3: [01:28:26] You're excited that it repeated more if it can begin to learn the English and we actually know those different milestones with humans. That is very impressive. But the compute getting back to your compute efficiency, I would argue to do that today, the compute would be disgusting, like it would actually kind of make you cry a little bit and there would still be a big question mark of whether or not there would even be possible. Because the big thing that's missing is search, search, recall. Because you would actually have to do search recall on petabytes of data in 100 milliseconds, because if I say hot dogs, you immediately bam, your AGI exploded that term out the first time it heard it. One shot learning you can talk about zero shot learning like. Yeah, so compute and memory already kind of hurt when you start talking about that example. But look, your brain. Our brains did it. You know, the brain's miracle. The question in my mind is that like, you're obviously right, like I look at my niece, look at any toddler, right?

Harpreet: [01:29:27] You show them

Speaker3: [01:29:28] A really vague, conceptual, cartoonish drawing of a tiger. But then they understand every representation of a tiger, whether it's hand-drawn, a realistic sketch, a photo, a 3D model. They understand that it's a tiger. They re as a human brain, we do the symbolic learning, this representative learning. That's. Inherently different to the experiential learning that we're

Harpreet: [01:29:48] Seeing with

Speaker3: [01:29:49] Most eye models these days, right? Like I think I don't know if that's a

Harpreet: [01:29:55] A

Speaker3: [01:29:55] Construct that we're missing, like a software layer that we're missing or we're [01:30:00] missing some piece of this puzzle in terms of reaching that symbolic level abstraction. And I think that's when we're going to start seeing a significant step up in the efficiency of learning. Right. And that's going to stop making these more general artificial intelligence systems more and more viable. When you bring those power costs down, you bring the ability to abstract out like, like

Harpreet: [01:30:19] You said,

Speaker3: [01:30:20] We're not going to hone in on that detail of a number nine. How do we get a machine to do that? And then you get a philosophical question of should we be building that at all anyway, right? Should we not just let humans do what we're good at, which is those abstract learnings

Harpreet: [01:30:33] And and

Speaker3: [01:30:34] Augment us with these with these specific A.I. that can look at the detail that we miss every time. Well, I think, yeah, I think the transition you're going to see is we head there is think of it as an incredibly useful AIs, so A.I. systems that actually exist to make your life easier. And so they're constantly working to like, imagine an air system that's cleaning the house and it finds a problem.

Harpreet: [01:30:57] Hey, this power outlets broken.

Speaker3: [01:30:59] It's bringing it to your attention. And if you don't like it, think of like the ring camera false alarm. That is something I will quickly react to. It doesn't. It doesn't want to interrupt you unless it's actionable insight. And so or the Mars rover actually being useful for 30

Harpreet: [01:31:14] Months without or like, you know,

Speaker3: [01:31:16] For a year without human interactions. It's kind of these autonomous systems. They're not conscious, they're not the singularity. But you're going to see this natural trend in the next 10 years where A.I. is desperate to to please you with actionable insights. And if it's overwhelming you, you will be upset and it will have some reaction to that. It doesn't mean it can't actually be sad, but I think that's a funny. That's a funny dynamic that might exist any smart homes where you begin interacting with them. And I think we've maybe we've talked about how that could be ethically bad to personify I. And think of like the therapist, so I come home, I'm talking to the home, my wife's mad at me, why I'm mad at you, I'm telling the home it's not alive. But if it has this personification [01:32:00] that could actually mess with me, it would definitely mess with kids. But it could mess with me where I'm laughing. Greg thinking about this because think of this system actually kind of became tuned to my experience. And then there's a firmware update and it

Harpreet: [01:32:14] Screws up my system.

Speaker3: [01:32:17] We would all laugh and say, Oh, it's OK, don't worry, you'll get the new initialize personal assistant. I would actually experience emotional loss because I have had a I have developed. Does that make sense like I am developing an interactive relationship with my personal assistant who's adapting to me, and it's been personified a year later, something screws up and it's not backed up. I will actually experience emotional loss as a human. And now what does that mean? Is, yeah, so personification of AIs super problematic. Yeah, her it's got to be mindful of. Like, like for me, one of my favorite quotes from the year twenty thirty five is I told you so by Isaac Asimov. So you've got to be a little bit mindful of that.

Harpreet: [01:33:02] The interesting discussion, for sure, I sent a link right here in the chat

Harpreet: [01:33:06] Room, a video

Harpreet: [01:33:08] By the YouTube channel called Pursuit of Wonder. Name of the video is called digital psychosis. It's a it's a very Kafka esque 12 to 15 minute video which plays on the

Harpreet: [01:33:21] Metamorphosis plotline where

Harpreet: [01:33:23] You know how the guy woke up and was a cockroach. This guy woke up and was a computer, and it's kind of exactly what Ben is talking about. All he could do is sit in his room and has a camera and just observes things happening as they are as a computer. And they're talking about how in the background he has this program running, that's actually his base level consciousness is pretty interesting. Good 15 minute watch. Definitely give it a watch. It's also here on LinkedIn PDF to check out as well. Ok, guys, well, thank you very much for that.

Speaker3: [01:33:56] Really great. I feel like I always take things [01:34:00]

Speaker5: [01:34:00] Down the rabbit hole of useless. I wanted you to go there, man. Don't be sorry about that.

Harpreet: [01:34:06] Hey, Ben, this is this type of stuff I like. I mean,

Harpreet: [01:34:09] I just I enjoy hearing it and hearing other people talk about it.

Harpreet: [01:34:12] So.

Speaker3: [01:34:13] Fun fact. If you put a night shade over the Ajai system in your kitchen, it actually can't learn to speak English, but neither can you. Because humans need focus based learning, you actually need so like that quote, you can't learn English listening to the radio. It's actually problematic for your brain. You can learn patterns, but you you need the focus based learning, which is fascinating because you look at how kids learn a language. Everything's focused based. Well, babies tracking AIs. Of course it is because that brain is ready to. It's in overdrive to learn everything that you do, which is so amazing because we even see this in the animal kingdom. Empathy, emotional intelligence like a baby hippo is very sensitive to behavior of adults and other hippos right out of the gate. And so and even your baby is sensitive to your novelty reactions. Oh wow. Like you as a parent, you have a novelty reaction. Your baby is

Harpreet: [01:35:07] Very like keen

Speaker3: [01:35:09] On that, which I think is just so fascinating.

Harpreet: [01:35:12] Oh, it's true. Like, I remember my kid said something yesterday and just made me chuckle, and he just kept saying it, and they kept saying it and saying it makes me laugh. And he was laughing. All you saying it to is cute. My babies are crazy, like like. Like, simultaneously, like learning, deep learning and then watching my kid grow up and doing the research and drawing the parallels between human intelligence and deep learning, it's just so fascinating to see in action what has been excellent discussion. Hopefully, you guys

Harpreet: [01:35:40] Tune in to the podcast or

Harpreet: [01:35:41] At least this week with Eric, and I got a few awesome stuff coming up in the next couple of weeks. Next week, I am interviewing live streaming with Natalie Nixon. She wrote a book called The Creativity Leap. It's here somewhere. I'll be giving away a free copy of that book as well. She's kind enough to give me two copies of that book. Also speaking to at the Data [01:36:00] professor himself, Mr. Chanin or Dr. Chanin. Rather, on the 23rd, I'm speaking to Marcus du Sakwa. He just sent me a new

Harpreet: [01:36:08] Copy, a copy of his

Harpreet: [01:36:09] New book. Really excited about this. He also wrote the book The Creativity Code, which I've talked about multiple times on this podcast. Excellent book, so super excited to talk to him. He's a professor of mathematics at Oxford, the Sami

Harpreet: [01:36:24] Simoni Professor

Harpreet: [01:36:25] Of Public Education of Science, or something like that. So that is going to be exciting as well. Also got a live stream happening with Danny Marr on the 28th and got a livestream happening with.

Harpreet: [01:36:38] They've linger, and eventually

Harpreet: [01:36:40] I'll get to all of my friends and get to all of my friends and will be on the show at some point. So will Tom, who will all the friends, so I'm excited about having all them on like you've already been on the show. I got Greg on the show before he was where he was famous. Guys, thank you so much for taking time out your schedule to be here today. I appreciate you being here. Remember my friends, you've got one life on this planet. Why not try to do some big cheers, everyone?