Harpreet: [00:00:06] What's up, everybody, welcome, welcome to the comet Kemel Office Hours Harp by the artist Data Science. It is Monday, August 1st. Happy August, everyone. Man, this is is awesome. I can't believe it's going by this quick years gone by super fast, man. Super excited to have all the guys here hopefully got a chance to tune into the episode that was released just a couple of days ago with Lily and Pearson using Ojai in the Data science game. She's been been a bit of a mentor to me over the last few months. So it's really great to to chat with her on the show and get her on to the podcast and all that stuff. Um, yeah, man. A couple of a couple of big announcements. I'm happy to announce that we join the team at Comet Emelle on September 1st, September. They'll be joining the team over at Comet next week, as will be my last week at Rice Industries, and get a little bit of

Harpreet: [00:00:55] Downtime and

Harpreet: [00:00:57] Then ramp up and do some awesome stuff over at that comet. I'm excited about this opportunity and excited to do some awesome work with the crew over there. Other announcement. Let's see. I wanted to pull up something. I did a poll a few days ago and the poll was about, um, you know, I was going to do on this twenty one day learning journey. Right. So let's go ahead and just take a look at the results of this poll. Um, I have to take a look at it. You can look at them. Yeah. But I want to share my screen and put it up here. What we have here and not bad. That's a pretty decent amount of votes.

Harpreet: [00:01:27] Um, looks like

Harpreet: [00:01:29] Nlp and deep learning are tied up, but there's quite a gap in between them, um, even though there's a couple of percentages off. So, uh, pull ads in just a few hours. Three hours. Let's hope you guys haven't voted already. Go ahead and get your vote in. Um, personally, I want to do the twenty one days of papers just because I got like a huge stack of papers that I've been wanting to have an excuse to get through. But my second choice was going to be twenty one days of deep learning. So I'm happy that when one um also NLP as well, I mean honestly [00:02:00] like I want to do all of these, uh, just needed to figure out which way to prioritize them first, but looks like twenty one days of deep learning first unless we get a ton of people to uh to vote otherwise. So I'm excited for that. Over the next few days I'll think about what my learning journey is going to be like for 21 days. I'll try to structure out twenty one days what I'm planning on hitting each one of those days. I'll share that plan with the guys on LinkedIn as well. Have you guys, you know, comment on it or or whatever, but I'm excited for it. I'll primarily be using a couple of resources to drive that.

Harpreet: [00:02:35] I'll be using John Crone's book, People in the Illustrated and then also, uh, grokking deep learning. And then I've got another, uh, PDF that, uh, from Springer. It's a Springer text. Uh, that's things just simply called an introduction to deep learning or something like that. So we bounce around those resources and trying to structure the learning path. I mean, we're limited by, uh, by LinkedIn, you know, character limit. But I'll try to get some creative stuff in there. I'll try to do some interesting, like slides or something to make it exciting and fun. So hopefully, guys, uh, follow along with that. I'll make an announcement probably later today and then probably take another vote on there to kind of get a vote for what you guys think the, uh, hashtag should be. Um, but yeah, I'm super excited to have all the guys here. Hey, if you guys have any questions, whether you are on LinkedIn, whether you're on YouTube or Twitch, if you guys got any questions at all, go ahead. Let me know. I'll go ahead and take those questions. Um, you're also more than welcome to join us right here in the zoom room as well. There's a link to join us right there in the description of the video. But, yeah. Have you guys here yet?

Speaker3: [00:03:50] Can I say one little thing? I just want to like first, I want to say we're super, super excited about the sort of hiring manager for the role we heard for, like, super, super excited. I got, like, so [00:04:00] many ideas that I'm going to be sharing with you and folks in the community and also so super excited for you to join us. And then the other thing was, I wanted to let you know that for folks who have signed

Harpreet: [00:04:09] Up for office

Speaker3: [00:04:10] Hours, we also did a meet up industry Q&A on Monday of last week. And so I'm working on preparing the video and things like that. So I'm going to also send that out to folks on the left. It's sort of like a thank you for

Harpreet: [00:04:22] For for joining

Speaker3: [00:04:22] The office hours here. I'm going to send you that that video so you guys can check that out and we'll be doing more of those in the future. So just in terms of, like community things to be on the lookout for both. Yeah, Harp read all your awesome work that you're going to be doing and we're going to do it together. And then that sort of events that we're kicking off to. So just before we get started on and all that.

Harpreet: [00:04:38] Awesome. Yeah. Thank you so much. I really enjoyed that session. I had an opportunity to to join in, um, and then I really learned a lot from this. Was it radio and deep.

Speaker3: [00:04:51] Deep no. Yeah. So deep down as I could call out like a more collaborative Jupiter type of environment. And the radio is like this graphical user interface for like testing models in real time. Really cool tools.

Harpreet: [00:05:01] Yeah. So I said, do you want to like kind of give it, give the folks a rundown of what they're going to. To see from you over the next that length of my tenure at Comit, I'm excited. Yeah, yeah, I'm excited.

Speaker3: [00:05:11] Yeah, yeah. So many. I think, you know, one of the things I've been really focused on is figuring out how our

Harpreet: [00:05:17] Our products

Speaker3: [00:05:18] Stack and how comments, experiment management tools can be better, can be designed better and more effectively implemented for people who are learning, for folks who want to build portfolios, for either hiring managers or if they're bringing Data science into organizational and building portfolios, using sort of like a system of record and kind of a GitHub system of record for machine learning specifically. So I have some great plans for bringing contributors into the mix. And a couple of those folks actually in the chat right there in the in the Zoome call right now who are going to be working with us. So I'm very excited about that. And I know you'll be very instrumental in helping that happen, but creating a lot of materials for learning and just learning and using these tools in more effective [00:06:00] ways and then building just a community around that. So we're very excited. So many different plans, but some of that's going to be events and it's going to be a contributor program. So it's going to be just us sharing educational content and helping make it easier to use and more accessible to everyone in the Data science machine learning community.

Harpreet: [00:06:16] Yeah, man, I'm really, really excited for that. That's something that's been you know, I've been doing that over the last four years, a year and a half or so. And just it just the opportunity worked out so well. And I'm excited to to help bring all this content to guys and do all this awesome stuff with the guys. We are super excited to be joining the team at Comit in just a little little over a month. But hey, if you guys had questions, go ahead. Let me know. I'm keep an eye out on all

Harpreet: [00:06:41] The all the streaming

Harpreet: [00:06:42] Platforms. If you got questions, you could put them right there into the chat. But we can also start right here. Man what's up with the Cristoff? How's it going to meet the. What's going on? What's going on? Good to see you again on T and Marianne. How are you guys doing. All right, sounds like everybody is doing good. Let's let's kick this thing right off the bat.

Speaker4: [00:07:03] Oh, sorry, boisterous, but yeah.

Harpreet: [00:07:06] Yeah, but you guys got a lot of responses here. So let's kick it off with today with a couple of questions that I want to start off with. Um, how about you are doing this? How about starting off with one of these? I'm a data scientist and I have never done

Harpreet: [00:07:21] A blank type of thing,

Harpreet: [00:07:24] Um, because I see a lot of these pop up on LinkedIn. I think there are a lot of fun. Um, yeah, I see that. So, you know, I'm a data scientist and I've never done Blinkx. Does anybody want to kick us off? We want to start off with this. How about this? I'll do this. I'm a data scientist and I've actually never deployed a deep learning model into production that that's that I'm a data scientist and I've never actually gotten an opportunity to do neural networks in industry. Yet all everything I've done has been classic Emelle. So I'm excited to be that. That's part of the reason why I'm super pumped to be going on this deep, deep learning, learning, training with the guys over [00:08:00] the next 20 ish days. I thought about you guys. Let's the let's start with the the netizens. That's neat. As Mike is on me there. Go for it. Data, OK, about what? About you, man.

Speaker3: [00:08:14] I'm not a data scientist.

Harpreet: [00:08:15] I don't know. You're definitely data scientist. We're doing this nice work. I consider you a data scientist, though.

Speaker3: [00:08:23] I don't know Data science because I'm focused only on machine learning. So I'm a data scientist and I never

Harpreet: [00:08:30] Did things outside of

Speaker3: [00:08:32] Washington.

Harpreet: [00:08:35] How about how about Merrion or ushe happy to hear from any of you guys.

Speaker4: [00:08:40] I'm a data scientist and I have never had the chance to use natural NLP. Yes, that's right. Yes, I can believe it.

Harpreet: [00:08:50] Yeah. That's one thing that I thought I was going to be working on. Cristoff with was doing some projects with the chat logs, not starting at the trialogues, but the transcripts

Harpreet: [00:08:59] From from

Harpreet: [00:09:01] The happy hour episodes. I've got those getting cleaned right now. I sent them off to to a freelancer to help get those transcripts cleaned. And I think it'll be fun to do some interesting NLP with that. But I've never got a chance to do that at work. It's just, um, I mean, the nature of, I guess, the jobs that I've had there, I've always been kind of business

Harpreet: [00:09:20] Focused type of roles where

Harpreet: [00:09:22] It's like, OK, well, I've got to do something the business requires me to do, which is typically, um, activities that will help them reduce costs

Harpreet: [00:09:29] Or no. Generate more revenue. Right.

Harpreet: [00:09:31] And yeah, unfortunately for me, it's not involved any cutting edge stuff. I got some questions coming in right here into the chat. So let's go ahead and take those on.

Harpreet: [00:09:41] And in the meantime, if

Harpreet: [00:09:42] You guys have questions, you could go ahead, write them out into the chat wherever you are, and I'll keep an eye out for them. Questions here coming from I think you I don't know how to pronounce his name, but it looks like ay yi yi yi yi Yi says I'm an Emelle enthusiast and have started learning Python. How best [00:10:00] can one break through into this discipline? I will. If you've started learning Python, that is the right direction. I'd say next thing to do is pick up a good book and work your way through that book. So in particular, yes, doesn't say that one. That is a good one. Hands on machine learning with secular and tensorflow is definitely a good book too. I use to break into Data science, but if it completely brandnew the machine learning in and kind of new ish to Python, he said You just started learning Python. I really enjoyed, uh, there was another O'Reilly book and I think

Harpreet: [00:10:32] It was just, uh, hands

Harpreet: [00:10:33] On machine learning with it, just like it learned or introduction of machine learning was secular. And, uh, that was a really good book. And through that book, I got really good familiarity with just how to use Python and how to work with the second Learn API. So I probably stack both of those books back to back as a way to kind of learn what it is that machine learning is all about. And then from there, I think everybody here already knows what my answer is going to be after you learn that stuff the way project do a project. Right. And that's to make the project really one that's interesting to you. It's Data is actually everywhere, right? Data is everywhere. You generate Data, um, whether it's through the music you listen to on Spotify, Pandora, whatever, you're streaming platform of choices, the movies you watch, a Netflix activity, whatever, you can get your hands on that data and just start doing fun. Interesting stuff with it.

Harpreet: [00:11:24] Um, like there's no

Harpreet: [00:11:25] Barriers to entry, like there are no barriers to entry in this field. I believe that there is no barrier to entry to as a machine learning like HCFA. You got to take a bunch of exams to be an actuary. You got to take a bunch of exams to be an accountant. You gotta take a bunch of exams. You have to do that here in this field

Harpreet: [00:11:45] To definitely get your hands on some

Harpreet: [00:11:46] Data and make it happen. Oh, we have Ben Taylor in the house. Happy to see Ben Taylor here, man. Uh, man is going on. So we got another question coming in from LinkedIn here. Uh, have you ever done energy analysis [00:12:00] related projects? If yes. Can you explain? Um, so I've never done energy analysis related projects, but I know there's a really, really good, um, a blog post. I think the guy's name is like Will

Harpreet: [00:12:13] Caution um

Harpreet: [00:12:15] Or something like that. KUAR as H and I could be pronounced that wrong, but he does this Energy Star, uh, like really well thought out project from start to finish and highlights his entire thought process in a series of like four blog posts. Um, so definitely check that out. I'll see if I could link it in a little bit there and I'll drop that right there into the chat for you, AJ. Um, and now the name of the books that we're talking about was, uh, um, hands on machine learning was like a learning tensorflow an introduction to machine learning with, uh, Ben Taylor. Man, what's going on? Good to see you here, man. Audio might have some issues. It says unmetered, but I can't hear you. Yeah, no problem. Ben, I got to check out his, uh, really interesting project. He sent me. Um, he did this thing.

Speaker3: [00:13:00] Can you hear me now?

Harpreet: [00:13:01] I can hear you now.

Speaker3: [00:13:02] I didn't turn on my fancy Mike.

Harpreet: [00:13:03] Sorry about that. No reason. Yeah. I logged into LinkedIn today. I saw your, uh. Me, that message about that really cool project you did, man, I'm excited to check that out. He did this thing where if he doesn't get up and work out, if this A.I. system will create this weird sound in all of his resume meetings or something like that, what was that about? A project

Speaker3: [00:13:25] About? So it's I built a I've got a camera running and it's looking at a 16 frame buffer. And then it sounds like a temporal representation to a system that it can tell if I'm doing push ups, sit ups, pull ups, taking a break from work, working or if I'm missing from my office.

Harpreet: [00:13:42] And so if it doesn't detect workouts

Speaker3: [00:13:45] On the daily cadence that I'm expected to, then it plays noises in the background or my resume calls, noises that I don't want to be played.

Harpreet: [00:13:52] So, yeah, I need to get some like that from you. I've got this thing where, like, I'll have reminders come up every periodically during the day [00:14:00] and it'll be, you know, get up and do some push ups to get up and do some jumping jacks. Just get up and do something. And it's just so easy, just like.

Speaker3: [00:14:07] Exactly. That's why I like the idea of having a consequence that I can't

Harpreet: [00:14:12] Ignore because

Speaker3: [00:14:13] I have meetings where if something if these farting noises are playing in the background on some of these meetings, that could be really bad for me like that could it's actually not funny. And so for me, it is funny because I'm not missing my routine anymore. But the the next project I'm starting on is about three young kids and they eat in the TV room and it doesn't matter how many times I tell them not to do that.

Harpreet: [00:14:36] And this week

Speaker3: [00:14:37] I was picking up Little Caesar's pepperonis

Harpreet: [00:14:40] Off of my

Speaker3: [00:14:41] Couch and there were grease stains like underneath on my couch.

Harpreet: [00:14:44] And so

Speaker3: [00:14:44] Next week is full on war against the kids with an AIs

Harpreet: [00:14:47] System that turns off the

Speaker3: [00:14:49] Tv if they eat in the TV room. And I'll be like, I'm so excited for this next one. And like,

Harpreet: [00:14:54] You do some really cool, fun, interesting stuff. Like, how did you did you have to teach yourself how to do all this stuff? Like, I mean, because

Harpreet: [00:15:01] Because you're you're

Harpreet: [00:15:03] You're integrating a lot of various systems and a lot of various different things together and doing awesome stuff with them. Like how did you even start when I

Speaker3: [00:15:11] Think it all starts with having a selfish idea. So if you have a selfish, passionate idea, it doesn't matter what steps are needed to get it done, you just get it done. And so for some of my for some of the projects I've done in the past that are the most fun, like this one of the one of the ones I've had the most fun with so far was this Xbox project I did a couple of years ago where I had to learning on

Harpreet: [00:15:31] Call of Duty on a stock

Speaker3: [00:15:32] Xbox. But for these projects, you just have a selfish interest and you just pursue it. And that's the most rewarding thing you get to your side.

Harpreet: [00:15:39] But most of the stuff is pretty. It's not. It's pretty doable. Yeah.

Harpreet: [00:15:43] So when you

Harpreet: [00:15:44] Have some of these

Harpreet: [00:15:45] Interesting challenges that you're working on and you just confronted with like, OK, blank Ed, how do you how do you start? Is there like a like a research phase that happens? Is it looking up what other people have done and then kind of taking a look at this [00:16:00] code? Does this let me try to take some of that adapted to this purpose and

Harpreet: [00:16:03] And work with it like that?

Speaker3: [00:16:05] I think some of it I think it's simpler than that. Most of the I would start

Harpreet: [00:16:10] Selfish because

Speaker3: [00:16:11] Here's the thing is, if it if you are genuinely interested in a project like, I mean, really interested, obsessed about it, there's a really good chance other people will be. And I think the fun thing when it comes to

Harpreet: [00:16:21] Creativity is

Speaker3: [00:16:23] The blank slate approach. So if I have a magic Harry Potter

Harpreet: [00:16:26] Wand and I

Speaker3: [00:16:27] Uncreative all of you guys, everyone has a blank slate. And then I say, who can come up with the first good idea for A.I.? I like this to approach.

Harpreet: [00:16:34] So I just say,

Speaker3: [00:16:35] In your life this week, what do you want less of? What do you want more of? And for some people they'd say, well, I hate mail planning, I want less mail planning. And I'd be like, great, let's figure out how I can help with that. And the funny thing is, if you explore that track, we're going to get to something where people can think that's amazing, like how creative. But really it's

Harpreet: [00:16:51] All starting from what do you want less of?

Speaker3: [00:16:53] What do you want more of?

Harpreet: [00:16:54] Well, I don't want to don't want to read.

Speaker3: [00:16:56] I don't want cats to poop in my kids' sandbox anymore. Like, you can quickly go down this list of projects. And most of those are A.I. projects. They're assistants.

Harpreet: [00:17:03] So I like that. Talking about creativity, I've actually I started reading this book just a couple of days ago. I don't

Harpreet: [00:17:09] Know if you've read or

Harpreet: [00:17:09] Not. I don't know if this is reading, but this article is called The Creativity Code

Speaker3: [00:17:14] And oh, that looks really neat. Yes.

Harpreet: [00:17:15] By Marcus du Toit. He's the guy that took over Richard dockings position at the, uh, Oxford colleges, like the, uh, the public facing scientist type of thing. So he does a lot of cool, cool talks and lectures and he's in a bunch of BBC documentaries. Um, but I like this book so much like I ordered a physical copy of it as well. It's called Art and Innovation in the age of A that's super, super interesting book. I'm really enjoying that so far. Um, I think Christoph has a question. So let's go to Crystal's question and then I'm going to keep an eye out on all the charts here. I see a bunch of stuff coming in on LinkedIn, so I'm going to catch up with that real quick. Uh, shout out to Russell Russell with, uh, I'll give you a link, come into the room and have a Cristoff go for it.

Speaker3: [00:18:00] Ok, [00:18:00] so I've got this kind of question. How do you define hard work and smart work? Because I see such. On LinkedIn and other places and people say you have to work smart and working hard is for, I'd say, losers by different names, and how do you define them and

Harpreet: [00:18:23] How do you tell the difference?

Harpreet: [00:18:25] That's a good question, man. I like that a lot. Ben, do you take a stab at this one?

Speaker3: [00:18:28] Yeah, sure. So hard work and smart work. I haven't heard it framed that way. I like it because it's kind of a different way to set that problem up. It reminds me of this idea of urgency versus strategic thinking and so smart work, strategic thinking. That would be you pondering in your office for two hours, not saying you have to do it this long, thinking really hard about some of the issues you could run into your planning a lot better urgency. You just you're thrown in the fire. You have to make this work and you need the balance of the two. And so if someone said they only did hard work, I might my default reaction might be OK. So you're constantly chasing fires and it's urgency if someone said they only did smart work. I think so. You're a strategic thinker that doesn't actually get stuff done. You're like the the academic on the hilltop. And so you kind of need a balance. And I think the less that that I have before, I'll shut up, as I think with hard work, sometimes with every project you run into unexpected hurdles.

Harpreet: [00:19:22] And that for me, that is the hard work. You're doing a project,

Speaker3: [00:19:24] You want this to happen. And then here comes this brick wall knowing you couldn't plan on it.

Harpreet: [00:19:28] It's now in front of you. And I've

Speaker3: [00:19:30] Spent hours I spent like four hours working

Harpreet: [00:19:33] Through a bug.

Speaker3: [00:19:34] I've taken a break taking a 15 minute nap myself. That bug in two minutes and the same brain, four hour bug versus five minutes. One was hard. The other was just. Yeah.

Harpreet: [00:19:46] So here's what people's reaction is. Yeah, definitely.

Harpreet: [00:19:49] Austin Dannette comments sense that's on this. Working hard versus working smarter.

Speaker3: [00:19:54] Yeah, definitely. Lots, lots of thoughts. I mean one example have comes from my time at comment sort [00:20:00] of. I think this is one of the things I've been able to bring to comment is like and I think I've mentioned this before, but I think

Harpreet: [00:20:07] Sometimes the

Speaker3: [00:20:07] Hard work I'll use this example, like when I came to comment on the marketing side, there was not a planned out strategy. There was not really as much of a deep sense of like how this is going to work over the long term. So when I got here, I felt like we were doing this live. Very limited resource

Harpreet: [00:20:24] Team was doing about ten to

Speaker3: [00:20:26] 12 things like 20 percent. They were working very hard to get all of these things aligned and scheduled and done. But then there was this follow through that wasn't happening with any of them. So all this hard work was going sort of almost to waste or it was it was showing it was presenting is like

Harpreet: [00:20:41] Experiments, marketing

Speaker3: [00:20:42] Experiments or content experiments that don't work. And then the reality was because there is all that hard work that wasn't followed up by sort of the smart strategic thinking like you were talking about then. So like what I've been trying to do, it's like we're going to do three, two or three things for things that 80 to 90 percent. So it's that and a lot of times what that is, is the hard work is maybe 80 percent. And you're in the grind, you're figuring out solving tough problems. And the smart work is that extra 20 percent or even flip that around. Right. Depending on what project you're working on, where those proportions represent some smart work and hard work, where if you do much more strategic thinking up front, that last twenty percent of the hard work is much easier if that makes sense. So if you have a good strategic plan in place, you do that sort of smart work up front. Then the hard work is you're executing a plan that you kind of know where it's headed,

Harpreet: [00:21:30] Hard work when

Speaker3: [00:21:32] You when you feel just like lost and you don't know what you're doing. I feel like that can be counterproductive

Harpreet: [00:21:36] Because you're just spinning

Speaker3: [00:21:37] Your wheels on a problem that you haven't even thought deeply enough about or intelligent enough to, like, figure out what the hard work is that you need to do. So it's like one becomes the thing that precedes the other and a lot of cases.

Harpreet: [00:21:48] But yeah,

Speaker3: [00:21:49] I know I have a lot of feelings about that

Harpreet: [00:21:50] Especially. And it's tough in

Speaker3: [00:21:52] In tech, in tech, especially when I think there's this like sort of high growth mentality, you know, like rapid growth mentality. And it's sort of like those wires can get across [00:22:00] very easily around like we just got to do stuff. There's this bias towards action, which I think can be valuable, but I think it also be very, very counterproductive as well.

Harpreet: [00:22:09] So I think it's

Speaker3: [00:22:10] Especially difficult in tech when everything's like, go, go, go, go, go, go, go, go, go, go. All the incentives and point in that direction. It can

Harpreet: [00:22:16] Be hard to convince yourself

Speaker3: [00:22:18] That like that slow strategic thinking is actually pushing you further along than sort of like the I'm just going to keep keep going at this, even though I don't really know what I'm doing.

Harpreet: [00:22:25] Really, really good. Good insights there. Asha, you got some great insights here in the chat.

Harpreet: [00:22:30] Go for it.

Speaker4: [00:22:32] I mean, the same thing is that I think looking smart involves more planning. You realize it, take less time to get to get it done. And working hard is just more like brute force. Try fieldtrip. They'll keep going, keep going. So definitely the difference is in the planning. Sorry.

Harpreet: [00:22:47] So I posted a couple links right here into the chat. I think it's no big secret that I'm a huge fan fanboy. His philosophy is that something that resonates really deeply with me. He's got a couple of posts that are one of the postings from his podcast and it's just talk about hard work. And from there, that's where I got this this quote from, which is hard work is no substitute for who you work with and what. You work on and I

Harpreet: [00:23:11] Think that is incredibly

Harpreet: [00:23:13] Important because he talks about hard work being I mean, he talks about success being a three legged stool, that three legged stool is working hard, who you work with, what you work on. Right. But just working hard by itself is no substitute for the other two. Um, so that's that really just hit me like, holy shit, man. And that's what kind of made me realize, like, you know, the previous job I was working hard at a price,

Harpreet: [00:23:35] But like

Harpreet: [00:23:36] Who I was working with and what I was working on wasn't going to make me successful. It wasn't going to lead me to do anything awesome or interesting. I mean I mean, no, describe to those folks are great, the great people. But we're just working on Data management stuff. It's like, dude, I'm a scientist type of guy. That's not what I wanted to do. But I'll turn over to Ben then. That's more called commentary. Beyond that,

Speaker3: [00:23:59] I [00:24:00] just had a quick thought that came up to this question. So I think in engineering, I'll throw Data science into this as well. There we can all think of examples where we've celebrated heroics, right? Like, oh my gosh, did you hear about that engineer like 48 hours later hackathon or even like your product? Like I've had I've seen this happen to HireVue and some people get some people like this. It's like an adrenaline high and like, yes, I know heroic heroics are awesome.

Harpreet: [00:24:26] Sometimes if there is

Speaker3: [00:24:28] A history of necessary heroics, it's usually it's an indicator of a bigger problem. It's an indicator of

Harpreet: [00:24:34] Technical debt and

Speaker3: [00:24:35] Heroics don't scale and heroics are typically filled with chunk code and shortcuts that don't they they're not self-sustained. They can't be handed off well and integrated. So when I think of hard work, I think of it. But at the same time, we still want heroics, right? Like, I still want

Harpreet: [00:24:49] People on my team that if

Speaker3: [00:24:51] Shit hits the fan, I want you to figure it out. I want to do heroics. But if you have to do heroics every week, that's toxic. So, yeah, super interesting. I had forgotten about that. The heroics are OK, but heroics are also common, like repeated heroics are a sign of a bigger issue. Yeah, it's like it's like figuring out the sort of balance between pros. Actually just talking about this this morning with my girlfriend, I was just thinking about some work stuff and it's figuring out to me it's really figuring out this balance between process versus outcome. Like the outcomes are the thing that give those dopamine rushes, that give that adrenaline, that give you that energy. But then process is what makes those things repeatable and scalable. And so how do you balance that in a high growth environment or in a environment where you feel pressured to, like, acquire new skills or whatever it is you're going through? It's like you really have to like I'm struggling with this or like, you know, this idea of CPI's and like, I got to hit my numbers and my guys. But underneath

Harpreet: [00:25:43] That, it's like if I feel

Speaker3: [00:25:44] Like if I execute the process well, that outcome will come. But I don't get that adrenaline rush of seeing the big spike in growth that day or whatever. It's like the steady long term thing I'm working on. And so balancing that can be really difficult. And I think that plays into it a lot.

Harpreet: [00:25:58] It's like I don't want to get

Speaker3: [00:25:59] Addicted to [00:26:00] that adrenaline rush of

Harpreet: [00:26:01] Like just doing a

Speaker3: [00:26:02] Thing and then having some results and then but not really seeing how that ties into this bigger picture, especially for me, like building community building, a sustainable, positive, reciprocal community is not about one off successes. It's about time those one our successes to a process, to a way of celebrating folks in the community to all of these things that go kind of go unseen underneath that. And I just have to, like, be OK with that. And that's like a very difficult thing for me, because so much of

Harpreet: [00:26:29] So much of our

Speaker3: [00:26:31] Incentives feel very outcome based. And I like my whole thing is to try to figure out how to balance that better in my mind.

Harpreet: [00:26:37] Yeah. I got to give you this recommendation. So it's a book I read a couple of times to some already practicing mind. It's short. It's like maybe one hundred pages if that one hundred, ten pages quick like one hour listen or two hour listen on double speed on audible. But it's all about that entire thing. Worry about the process, execute on the process, let go of the outcome. Very rooted in like a Taoist philosophy and a little bit of philosophy there. But here's the thing for me, man. Like like working smart to me is working on those type of things that are you are uniquely suited to doing that. Your return on investment is going to be high leverage. Right. So, for example, I could spend one hour pulling weeds outside.

Harpreet: [00:27:22] Right.

Harpreet: [00:27:22] But is that a good use of my time? I'm working hard. Yeah, I'll be sweating. I'll be in the final toilet. It'll be hard work. That's not smart work because that one hour of time is better used for me personally. If I spend an hour reading this book right. If I spend an hour writing something, if I spend an hour doing this, I pass that I'm uniquely suited to do where that my return on time investments is just going to have an outsized impact. That's what I think working smart is. Working hard is doing things that you're just, um, probably not as well suited to, are just it's not a good use of your time. That's kind of how I think about [00:28:00] in my mind.

Harpreet: [00:28:00] Awesome.

Harpreet: [00:28:01] Yet let's continue going on. Great question, because we like that there's a bunch of questions coming in from LinkedIn and on. Uh, YouTube, as well as right here in the chat, will take while take some questions right here from the chat first and then we'll go on to a YouTube channel. LinkedIn Parrot has a question here about I could ask on behalf. I think sometimes we have audio issues, but if today is not one of those days, feel free to jump in, but go for it. Hi, everyone. It sounded good today. What's your question? Yeah, so

Harpreet: [00:28:35] This is for me

Harpreet: [00:28:37] And a few friends of mine. We all kind of have this question. How do we decide

Harpreet: [00:28:42] Between

Harpreet: [00:28:44] Which rules to pursue? Like we might piñon lot analysis, paralysis kind of situation between deciding between Data scientist versus machine learning engineer.

Harpreet: [00:28:55] So my expectations are that, uh, if I take a step back, I just

Harpreet: [00:29:00] Want to have more impact and more exploratory role. And I realize both of these could fulfill those boxes, so. Which one to pursue

Harpreet: [00:29:12] And which one is easier

Harpreet: [00:29:14] If we are coming from software engineer backgrounds like that, if we are looking for a transition sort of. I would say like probably I mean, just answer that last question, which is easier transition coming from software engineer background, I would think like machine learning, engineer, Data engineer type of role would probably be easier to transition into just because that is just such a heavy, heavy software development type of role. But then doing take on his first Harp that question or both have.

Speaker3: [00:29:42] Yeah, my short answer to the first half is I think a huge part of that decision is going to play into your employer because I could imagine two different employers for one, the Data science role you're doing. You're not being challenged, you're not innovating, you're not becoming your best self where another company or another startup, you would be really [00:30:00] challenged. And so maybe I'd kind of take a step back rather than deciding on one versus the other. I think for me, I enjoy Data science a lot more because you're coming up with new applications. You're inventing new algorithms and new processes to try out.

Harpreet: [00:30:12] But your employer will

Speaker3: [00:30:13] Be a huge, huge part of that on whether or not that feeds your soul and makes you your best version of yourself or if that stifles your creativity. You want to get a job where the dumbest person in the room, you want to be in a job with the smartest person in the room, because then who's going to teach you? How are you going to learn?

Harpreet: [00:30:27] Yeah.

Harpreet: [00:30:29] Um, anybody else have any insights or comments there. Um, I think Benteke great job answering that question. Uh, I mean it's like how do you decide between machine learning and University of Scientist is like, what do you like doing better. Right. But if you like writing code and uh, I mean, you going to write code in both scenarios, but if you like actually engineering the stuff like writing in stuff to play the production and doing all that, I would say do the engineering type of role. You like the exploration and the research, uh, collecting business value to to the work that you're doing that maybe think about the scientist role. Um, do you have any comments or anything on questions? No. Kristoff, go for it.

Speaker3: [00:31:08] Uh, could I just say I think you should follow

Harpreet: [00:31:12] Your your curiosity. I mean,

Harpreet: [00:31:15] To,

Speaker3: [00:31:17] You know, yourself the best. No one knows you as you do and just look where you are. I'd say lose track of the time. I mean when you do

Harpreet: [00:31:25] Some projects, some

Speaker3: [00:31:27] Exercises, anything, and either Data science or Data engineering, just

Harpreet: [00:31:36] Where you lose your track

Speaker3: [00:31:38] Of time, it means that you really like it and you really enjoy it. And I think you should just follow it

Harpreet: [00:31:46] If you're starting.

Speaker4: [00:31:47] Yeah, you and I have the same question for the state of Michigan because because we know we have learned algorithms [00:32:00] and not going back, not done any algorithm like any. I know we have done a lot of data analysis.

Harpreet: [00:32:08] And so what

Speaker4: [00:32:10] If we are given a task of machine learning algorithm to implement that? Can we be able to do that?

Harpreet: [00:32:16] That is a question

Speaker4: [00:32:17] That I have

Harpreet: [00:32:18] That sure, I fully understand the question. But I would I would say yes, because I think as a data scientist, you still need to be able to code, obviously needs to be able to write good production, really code. Um, but I like the delineation in my mind between data scientist and machine learning engineers that machine learning engineering is going to be more heavily focused on the engineering of it, taking this model and plugging it into somewhere where the data can come in, go through whatever transformations to happen, get past the actual model without the result, and make sure that that model is scalable. It's not going to break. It's engineered properly. Right. That is deployed and, uh, and just do its thing with minimal input. Ben, do really think it.

Speaker3: [00:33:04] I, I like I like what you were saying, uh because it was the question about whether or not you could do something. It really comes back to your passions. Right. Because look at it like I think if you look at myself or other people, I never took computer science in college, but I've been invited to present to like Red Bull, Goldman Sachs and Space X when it comes to like A.I. But I never studied it. And so what that tells you is if your passions align, can you do it?

Harpreet: [00:33:30] Absolutely. Can you invent

Speaker3: [00:33:32] This next algorithm? Can you do it totally? No, no hesitation. You can do it, but you need so instead of asking what's the right

Harpreet: [00:33:38] Decision, you need to

Speaker3: [00:33:39] Kind of look internally and decide which

Harpreet: [00:33:41] Of these excites that passion.

Speaker3: [00:33:43] So if you see yourself being drawn a certain way, then lean that way. And if your passion aligns with the work you're doing, you will become an expert, you will become a leader. And whether or not you can do it is it's no longer an issue, right?

Harpreet: [00:33:54] Yeah, absolutely. Except I didn't I didn't study computer science like at all. But I mean, I was able to [00:34:00] learn it because I found it fun and interesting and I enjoyed it. So what did I do? I picked up grokking algorithms and I picked up some other collection algorithms, books, and I just wrap my head around it. Oh, I get it. I didn't really write code like I like the way I write it now up until like three or four years ago

Harpreet: [00:34:17] And then just started getting good

Harpreet: [00:34:19] At it because I kept doing it, kept actually kept enjoying it. Liking it, it didn't feel like hard work.

Harpreet: [00:34:23] It felt like fun work.

Harpreet: [00:34:25] So, yes, it's always going to come back to what it is that you enjoy doing more, right? That's what I would say. And plus, Data science itself, like

Harpreet: [00:34:33] The data scientist

Harpreet: [00:34:34] Jobholders, we're talking

Harpreet: [00:34:36] About this on Friday.

Harpreet: [00:34:37] We opened up the session on Friday with the question of what type of data scientist are you not. So definitely go back and listen to Friday's happy hour session and you'll have a good hour of just everyone talking about the type of data scientist that they are, not the type of work that they don't like doing as data scientists. So they decided to focus on why they decided to focus on what it is that they focused on. So extremely recommend going back and listening to that one.

Harpreet: [00:35:03] Um, it's you get some

Harpreet: [00:35:04] Good insight from people like, for example, me, like I'm not like a product analytics data

Harpreet: [00:35:08] Scientist. Like I don't enjoy product analytics.

Harpreet: [00:35:10] This is not fun for me. Um, but I love doing research. I love communicating. I love doing this type of stuff.

Harpreet: [00:35:18] And I like I love machine learning.

Harpreet: [00:35:20] So that's the type of data scientist I am, the one that's focused more on research and on discovery and on exploration and then communicating findings and things like that. Um, a couple of great comments here coming in from the, uh, from the chat. I'm just going to read them out real quick to Russell Wilson, the House. And he had a great response here to that hard work question. Working hard usually precedes working smart. It's easier to just start doing stuff, even if it's messy. Then at some point when I started to repeat the same stuff manually, then I switched to smart work to avoid having to work hard and more. If that makes sense. Yeah, makes sense. Because, like, there's this thing that I've read many times and said that people who become [00:36:00] experts in a field start to think less about the work that they're doing. Whereas if you're a beginner, there's a lot of like mental activity happening and it's easy to get wrapped up and tripped up rather than little mistakes,

Harpreet: [00:36:13] Whereas somebody who's done this a lot more

Harpreet: [00:36:15] Has to think about a lot less interesting neuroscience kind of finding. And then,

Speaker3: [00:36:21] Harpreet, I want to comment on something, if that's OK. Now, then, Ben, you just posted this thing in the chat, which I really like, is that your passion can take you in and new made up positions. You can invent a role outside of your typical Data science,

Harpreet: [00:36:32] Like in a couple

Speaker3: [00:36:34] Of things on that is I find myself I'm not a data scientist, but I think that lends itself to be true because I'm sort of married like this strange love of writing and communication and teaching and counseling and all these things into this. Like I you know, I've been a head of community for in the tech industry for like five, four or five years now or whatever it is. And that was something that wasn't it didn't exist years ago. So I'm sort of like at the beginning. And then I

Harpreet: [00:36:55] Think about it. We were talking about you

Speaker3: [00:36:56] Joining Comet Harpreet and not to dwell on that,

Harpreet: [00:36:58] But like sort of that's what we

Speaker3: [00:37:00] Did. We started just like, what are your skill sets? We made up a role. Like, we knew kind of what we wanted and then we just, like, mapped it onto the things that you're very passionate about, because we knew that that would be a thing

Harpreet: [00:37:09] That would connect to

Speaker3: [00:37:10] Our community. And it's like we now we have you coming into this, like, unique role that's

Harpreet: [00:37:14] Going to allow you to

Speaker3: [00:37:16] Theoretically and

Harpreet: [00:37:16] Hopefully express

Speaker3: [00:37:17] All of those things in your work. And I think like the reason why I saw you out for that role specifically was because I saw that passion.

Harpreet: [00:37:25] And I think the good like hopefully,

Speaker3: [00:37:27] You know, good hiring managers and good companies and smart people

Harpreet: [00:37:30] Will want to attract that kind

Speaker3: [00:37:32] Of talent and like create those jobs for you, because this is also nature. It's still also like we act like we know all this stuff about like you think it's like less than a decade, like those key positions, like less than a decade old. There's plenty of room to chart a new path.

Harpreet: [00:37:47] And you just kind of have

Speaker3: [00:37:48] To, like, really follow that curiosity. I going back to what Chris was saying. So I just like I've seen that actually play out in my own

Harpreet: [00:37:54] Life and then in sort of the positions

Speaker3: [00:37:55] I'm just like creating out of thin air. Like I comment in my first few [00:38:00] months here. It's like it's really cool to see. And it gets me excited about sort of being in this role to facilitate that and watch someone carry out their budget, multiple people and carry out their passions and try something new and different. And so I think that's really good advice.

Harpreet: [00:38:15] That's really good advice.

Harpreet: [00:38:16] And when you when you start to do that, when you start to follow your interest and your curiosity, your passions start just doing interesting stuff, you become something really important. And that is the type of person that you can't go to school to become. Right. You become somebody that, you know, I can't get a certificate and become like Ben Taylor. I can't go to school and become like, you know, Harpreet Sahota like that just doesn't happen because I, you know, we each follow our own unique interest, unique curiosity. And then what happens is you're able to find opportunities that are well suited to that. Whereas if you just kind of stuck on that, I want to be a scientist. Great. But what type of data scientists are going to be what are you interested in? Focus on that, then go for it.

Speaker3: [00:38:53] Yeah, I'm a huge fan of like a blank page. Write your dream job like what you want to do. And I think people don't realize how crazy it can become. So like, I feel very blessed to be where I am now. But like I we're going to go on a backpacking fishing trip with like a partners and

Harpreet: [00:39:09] Bring a film crew like, well,

Speaker3: [00:39:10] What the hell does that have to do? They are it's like, well, we'll we're going to build out the system to protect for and catch a fish are like like these types of projects. They don't feel like work. But if you like sail like I really like sails. I like. Meeting with really smart people over dinner, for me, that's been a total joy in my career to meet people all over the

Harpreet: [00:39:26] World and have to talk about whatever we talk about over dinner,

Speaker3: [00:39:28] Whether it's the singularity, religion or whatever the hell we're going to talk about. And so that's become a big part of my job. Like, I just have dinner with fascinating people. And so, yeah,

Harpreet: [00:39:39] There's no doubt knowledge and I would argue

Speaker3: [00:39:41] Like, if anything, the roles that kind of feel more generalized avoid those roles are those companies. I try to find a role where they're trying to match Data science on a marketing or they're trying to match machine learning, engineering or something else where, you know, oh, no, if I go join this company, I'm going to have to be scratching my head and hit my head on two different walls to figure stuff [00:40:00] out. And that's why I'm a big fan of startups, because you have to wear

Harpreet: [00:40:02] Multiple hats, which means you

Speaker3: [00:40:04] Don't know anything. But it means you have to know something quickly, but sometimes these bigger companies, you can just kind of they stifle creativity, not because they try to, but just the reality of red tape. When I was at Micron, I. I wanted a faster laptop for Data Science and they said every engineer gets the same six hundred dollar laptop and it needed like three levels of management to approve me getting a nice free laptop.

Harpreet: [00:40:25] But yeah, I know exactly what you mean. Been there, done that. And I know those struggles and something that Ben was talking about there. It's again, like I said, I'm in the Barbican fanboy. I could never have one officer. I don't reference him, but he talks about specific knowledge. Right. Feels like play to me, but looks like work to others specific knowledge. That's a really interesting concept. Um, yeah. So definitely check out that that quick post here that I that I can listen to it. Click six minutes. All right. Great discussion so far. Who kicked that off, but that's good. Um, let's continue moving on. I got a question coming in from YouTube. Um, Mohammed, he's asking what is a good time to start reading research papers? If I started machine learning a few months ago, will reading research now confuse me during that stage? Uh, probably, yeah. I think that would probably

Harpreet: [00:41:21] Be not a

Harpreet: [00:41:22] Good idea unless you are really interested in it. Ben, what are your thoughts on that?

Speaker3: [00:41:26] I yeah, I discourage that because I think sometimes with these white papers that are not well written, they're very intimidating. Even for people that like math, sometimes they're so heavy on tech that you're cross-eyed looking at this paper. And so some people get discouraged. They'll start reading these white papers and I trying to understand them. And so I'm a much bigger fan and understanding the concept of the high level, what you to watch, understand why you would use a particular algorithm and then study white papers if they get your passion project or if they hit crap. We're having an issue. We're trying to do this first thing in technology. There's some white papers I would never recommend using white papers for a foundation [00:42:00] in a Data science. You'll just feel

Harpreet: [00:42:02] Discouraged.

Harpreet: [00:42:02] Yeah, absolutely. We just spent some time mastering the fundamentals, the basics, and then the research papers. When when you have a problem that you haven't encountered, that you think somebody else might have done something similar to this. Uh, and you find out they have the research paper. That's kind of how I've come come to that. But then, I mean, that being said, like I mean, when it comes to research papers, like I've got a stack right here that I'm really excited to read through. Most of mine are all that I'm reading. So one of the unreasonable effectiveness of deep learning in A.I., uh, there's the tutorial introduction to decision theory.

Harpreet: [00:42:37] That's a good one.

Harpreet: [00:42:38] Uh, the unreasonable effectiveness of Data, the unreasonable effectiveness of mathematics, um, just into unreasonable effectiveness. That's just something that I find fascinating. Uh, but yeah, hopefully that that clarifies that for Muhammad. I 100 percent agree.

Harpreet: [00:42:53] Um, wait it out a little bit

Harpreet: [00:42:56] Better to wrestle with this. Good as the man. Um, there's the question coming in from LinkedIn here from, uh, Christine Seagrave. I think a lot of people forget that Data science is less than a decade old. Do you think that enough time has

Harpreet: [00:43:11] Passed for normal

Harpreet: [00:43:13] Normative procedures to develop in this field? Um, it's enough time passed for normative procedures to develop in this field. I'm not sure what normative procedures mean. Does that mean like a standardized, like workflow or, uh, if anybody has any insight on that? Yeah, I should go for it. Oh, I see you.

Speaker4: [00:43:31] No, I don't have incident that. I just have a different question. Oh, OK.

Harpreet: [00:43:35] Yeah, yeah, yeah. Definitely. Um, wrestle or bend, which is normal procedures. I mean I

Speaker3: [00:43:40] Might have a quick response to this to my twenty second response is if you look at deep learning specifically, it is so sharded out. Right. You have like Tensorflow Petrarch, all these different academic and then you bring up like Kafe like oh yeah. People used to fight Kafe will never go away.

Harpreet: [00:43:54] It's dead now.

Speaker3: [00:43:54] And what that tells you is we're really scattered. There should be one deep wandering library that Invidia [00:44:00] is backing that we've all agreed to, but there's not yet. So I think there's still a lot of scattered efforts, still a lot of technical debt from a humanity perspective. But I feel like that's slowly being cleaned up. Might take another five or ten years before we have the one or two like in. That's just specific deep learning. Right. Then you have the R versus Python debate, but we don't need to bring that up here.

Harpreet: [00:44:20] John Cohn was ignoring the pie torch versus Tensorflow debate, uh, last week. I've been mostly learning high torch, but John's book is all in Kubernetes and Tensorflow. So that's been interesting as well. But I don't know normative procedures, like if somebody can help define that for me, I'd really appreciate that if I have to go.

Speaker3: [00:44:39] Yeah, I have an idea. I think one of the things and this is maybe this is I'm a little biased because I started working a comment. I'm starting to see all these issues arise as one of the things is sort of this mentality switched from treating machine learning and data science teams as traditional engineering teams to more like their own sort of specialized units where they need sort of this large scale like Mellops thing that's developing and right. Like Mistal a lot to learn and find out about it. But I take Comite, for example. What we do is. Sort of the experiment management space is like trying to make my

Harpreet: [00:45:09] Team is able to

Speaker3: [00:45:11] See the work each other's doing, like move away from like Logi putting output logs and spreadsheets and move it to a centralized system in the same way that, like GitHub was trying to do, where you're putting all your software engineering practices and operationalizing it inside of a repository. It's the same ideas that there's just like a whole different set of processes. So there's a space where you have to define what those processes are. Then you have to split out like, OK, how to different roles fit into those processes. How do we build effective teams? How do we build effective systems that do that? So that's still very much in its early days. I mean, we've been talking about this stuff for five years or less, so it's not like I had decades to figure this out. But that's one thing I see as well. In addition to the tooling around the libraries and things like that, I think is actually around how you think about structuring teams within an organization, that it's not just one data

Harpreet: [00:45:57] Scientist just playing around

Speaker3: [00:45:59] In notebooks, [00:46:00] but it's actually full teams that have their own distinct processes and tooling

Harpreet: [00:46:04] And ecosystems

Speaker3: [00:46:05] That they live in sort of existed and how they then they connect those to other parts of the business, because that's that's one of the hardest parts is how do you connect? This is the thing we get a lot about is like how to as a data scientist, how do you communicate the value of what you do and things like that? And I think that's a huge part of the ops space. The operationalizing

Harpreet: [00:46:21] It that to me feels like where this is all

Speaker3: [00:46:24] Sort of head, it is like figuring out what that actually means. It looks like in terms of the life cycle of model development and deployment and retraining and all these kinds of things that we talk about

Harpreet: [00:46:33] Like that a lot that really clarify for me as well. I think that does answer your question there, Christine. That's very good. Comprehensive answer there. Thank you very much, Austin. Yeah, that's, I guess, a number of procedures out of this entire

Harpreet: [00:46:44] Thing of this Mellops.

Harpreet: [00:46:46] And we've control and code and controlling data sets and controlling models and just having that whole lifecycle. Christine, let us know in the comments in LinkedIn if you have any follow up, anything. I'm happy to get to that.

Speaker4: [00:47:02] I should go for it. So my question is the way especially they've been you've said you start a new project every single time. You just take on the challenge. What's the process you go through in setting a new project? Because sometimes you can just copy paste some code and it'll work. You have no idea why it worked. It got the job done. You've got to explain why do you. And also with the research that leads sometimes to a rabbit hole, you read, read, read, try to figure out how to search. But by the time you get to doing the actual project, if you like, tab is nearly up. What's the process for you?

Speaker3: [00:47:34] So that with projects there's normally a lot that are in flight. And the nice thing is they carry on a life of their own. So normally the projects I work on now, they involve partners. So they they actually involve legal agreements and other companies and sometimes they die on the vine and other times they come alive. And I've got a project right now. It's pretty ambitious. It involves a professional athlete like an off road vehicle. And we've gone through legal agreements between three [00:48:00] companies, but we're stuck on the third company right now with their lawyers in that project may never start. And it's nothing. It's actually outside of my control. But it might if it starts, that'll be fun. We'll jump into it. So with projects, I like them, too. Sometimes it's just an idea stuck in my head, sometimes for a couple of years. And that idea will just kind of fester and grow or or things will be bolted out of the idea. It'll evolve and evolve enough that the seed will escape and it'll actually become reality. And then when it comes, becomes reality, there's there can be a lot of chaos with some of these projects

Harpreet: [00:48:31] Where there's no

Speaker3: [00:48:32] Template. How the hell are we going to do this timeline's I'm I'm kind of a fan of procrastination.

Harpreet: [00:48:37] So you you

Speaker3: [00:48:38] Have this you have these two tensions. The one of them is with procrastination. Things magically get done in time. But that's not always a good thing because you can suffer on quality, etc.. The other thing that's top of mind is perfect is the enemy of good. And so for that little project I just shared about like a fitness coach, I figured out out of the gate how to get it to go practice Dream HDR ten, which which is sixteen bit video. And from like a nerd perspective, I'd rather have sixty five thousand unique values per pixel than two hundred fifty five. That's an example of perfect is the enemy of good because I actually invested more time to

Harpreet: [00:49:07] Make that work. But for the project

Speaker3: [00:49:09] It was not necessary. So that's something I'm constantly trying to catch myself with. Am I falling prey to geek fantasies or techno like these technophile efforts that aren't required for Dunn? But that's just kind of the constant dance that will always exist, that I will always get myself in trouble, but I will always try to get myself into bigger projects that intimidate me. I've got a project in

Harpreet: [00:49:30] Q4 that definitely

Speaker3: [00:49:32] Intimidates me, but it's all but that's the thing. You win on smaller projects and you and you hunt for bigger projects. And hopefully you find yourself in a scenario

Harpreet: [00:49:40] Where you're thinking there

Speaker3: [00:49:41] Is no way I can make this project work. And if you can live in that reality, you will have a very exciting reality.

Harpreet: [00:49:46] So that's awesome, man. Yeah, I like that idea of

Harpreet: [00:49:50] That that thing you're talking about, rather.

Harpreet: [00:49:51] Where is the idea? It pops up in your head and begins to fester and take over your entire mind and then all of a sudden it just happens. That's exactly what happened to me with this podcast.

Speaker3: [00:50:00] Let [00:50:00] me not to belabor the point, but it also relates to startups, though. Some people say, I want to be an entrepreneur, I want to go to startup. I like to say, oh, what do you want to do? They'll tell me and they'll say, are you willing to foreclose on your home and burn through all of your 401k savings immediately to achieve this goal? And most people say no. And then I'd say, well, you haven't found the right project.

Harpreet: [00:50:18] You haven't found

Speaker3: [00:50:18] The right company. And so to really find the right projects, I would argue your passions are so fired up about this project that you are willing to you don't need to risk the ultimate outcome like a startup does. But you would you're willing to risk something. You're willing to risk your reputation, willing to risk the weekend. You're maybe even willing to risk your job for some projects to get big enough if you're willing to risk your job. And those are the best projects of all.

Harpreet: [00:50:40] It's almost like

Harpreet: [00:50:41] It's the type of thing that you say that you can't you can't not do. Like it would be an injustice to yourself if you did not do this particular thing. I see. Good question. Coming in from Russell. Follow up question, Russell, go for it. I think Russell froze up. He was up there. Yeah. Yeah. So he had a follow up question for Ben. Go for it.

Speaker3: [00:51:04] Yes. I'm only talking about the perfection being the enemy of good enough that. But have you ever tried

Harpreet: [00:51:11] To to

Speaker3: [00:51:12] Have like a temple to developing both at the same

Harpreet: [00:51:15] Time so you can get and you could

Speaker3: [00:51:16] Have gone with your lesson and standard and had that thing working in in in one stream whilst you then try to set up another going for the high quality, the beautiful. Yeah. I think the smart way to do it is to quickly admit a two faced approach like version one and two. So rather than me wasting time on a year ten, I should have quickly said I think that's possible. Maybe like sub hour

Harpreet: [00:51:39] Validation and then just shelf it. Don't worry about it. Go to version

Speaker3: [00:51:43] One with the mindset that there will be a version to version two. I will invest time to improve based on lessons from version one. But honestly, the reality is you never come back to version two.

Harpreet: [00:51:53] But just but just kind of telling

Speaker3: [00:51:54] Your stuff that there is a version too, because you're usually on to the next project or you're on the next customer or something else. But [00:52:00] if it's a if it's valuable enough, then version two is going to come marching

Harpreet: [00:52:03] Off the shelf and become a

Speaker3: [00:52:05] Reality. So that's been helpful for me is to admit that there is a version, too.

Harpreet: [00:52:09] But sometimes

Speaker3: [00:52:10] You. Well, it's tough, I say it's advice to myself, right? Look, I'm not saying like I still screwed that up on this last project. I still wasted too many hours getting the first one to work rather than just quickly admitting I see room for version to pop it over with plans to work on in the future and just avoid it for now. And we're always learning, but

Harpreet: [00:52:29] It's hard to do

Speaker3: [00:52:29] Things in parallel. I like the idea of just like carving it off. The other thing that I'm starting to realize more with some of these creative projects, there's reasons to do them several times. So actually redoing version one, rather than resorting to this really technical, there might be reasons to do version one a few times. And that's because you have maybe you want to shoot the video a couple of times. Like there's there's a whole creative element that's outside of Data science. How do you want tell the story? And sometimes you don't know how to tell the story until you've seen it once and then you want to go retell it or reshoot it a different way. Awesome.

Harpreet: [00:52:57] And that's that. I'd have to come back and listen to that one. Nakash, if you're listening, actually clip out some of that, that one previous one so we can make something out of that.

Harpreet: [00:53:07] Um, thank you very much for that.

Harpreet: [00:53:08] If anybody else has questions, let me know. Um, looking at the charts here, no other

Harpreet: [00:53:13] Questions in any

Harpreet: [00:53:14] Chat.

Speaker3: [00:53:15] I've got one on one of the things I've been some of your comments earlier about like here, sort of selfish projects and sort of the cool different ways that can create these divergent ideas sort of made me think of like and I've been working with this tool. And you were going to talk about Harp about like the transcripts and cleaning things up and creating a Data set there. I've been using this tool called Descript.

Harpreet: [00:53:35] If anyone's familiar

Speaker3: [00:53:36] With it, like an audio transcription tool is pretty good. Like there's a lot of it's a it's a really good

Harpreet: [00:53:42] Product that uses

Speaker3: [00:53:43] Machine learning. So I'd be curious to know like what products or what products that use machine learning that folks have encountered that they really like and sort of why just in the sense of making this like more tangible, because I think a lot of what we talk about is the creating business value free, like fraud detection. I think there's this other space for creating cool [00:54:00] products that detect if your kids are eating pizza on the couch, all the TVs on or are doing these things are the things that you guys have seen or worked with or tools or anything like that that you guys like that use machine learning. Well, so

Harpreet: [00:54:11] Just to follow up on on on that

Harpreet: [00:54:13] Point,

Harpreet: [00:54:14] Since we're talking about transcriptions, I use sonic data. And so when I sonic that and they've been really good and very helpful, I find that fascinating. But yeah, that's a good question.

Harpreet: [00:54:26] Yeah.

Harpreet: [00:54:27] What are some things out there that you guys have enjoyed, some products out there that you've enjoyed, that use machine learning or AI in general. Um, any, anything that's on top of mind.

Speaker3: [00:54:39] Well, I'll, I'll turn this into an Invidia commercial and I'll plug in video. Nano is the cutest little computer. I think it's like sixty dollars and it finally runs Ubuntu and you want to call if you played with like the T one two chips. I hated them with the passion. They were a nightmare waste way too much time setting up. And so find reasons to play with

Harpreet: [00:54:56] That and you will

Speaker3: [00:54:57] Find yourself doing some really, really fun wacky projects and you'll stand out and arrest me because everyone looks the same. And if you can add unique projects to you where you're selfishly interested, I don't like fantasy football, but if you were selfishly interested in that, you did a project that I would celebrate it. I would see that as novelty and passion and so be very self-motivated. I actually had we had Hannah Fry on her podcast and I asked her this question. I said, like, why do you how do you come up with these creative ideas? Like, she does the coolest things, like the mathematics of love. She's written all these books. And I kind of was teasing out like almost like, did she plan these things? And her response essentially was like, no, like they're selfishly interesting to me and kind of that same approach, like find projects selfishly interesting. Check out that little nano in between. It's the coolest computer I've ever seen with the GPU. Sixty dollars. It'll do all sorts of things.

Harpreet: [00:55:45] Going to be picking that up at the Jetson Nano and VHS and Nano.

Speaker3: [00:55:50] Yeah. Yes. So many fun

Harpreet: [00:55:51] Accessories. You can make

Speaker3: [00:55:53] Your house come to life with some

Harpreet: [00:55:54] Of those projects AIs and definitely

Harpreet: [00:55:56] Check that out. Um, yeah. Any other products [00:56:00] that you guys out there man. I'm at a loss um because I just to use that many products do I

Speaker3: [00:56:06] Not not set over not to speak too much on this, on this question, but the one of the things I like is don't predict the future,

Harpreet: [00:56:13] Build it. And so if

Speaker3: [00:56:14] You have things you don't like, like right now we're starting to listen to our podcast. We are definitely going to include A.I. somehow in season two. We do use Descript

Harpreet: [00:56:22] Austin, but there are

Speaker3: [00:56:23] Different things with editing that are a time suck. We don't want

Harpreet: [00:56:26] To do them and so can we pull

Speaker3: [00:56:28] In. Why are we doing this? Because we're building the future that we want so we can produce podcast episodes twice as fast. But that technology doesn't exist the way we need it. And so everyone get even more selfish.

Harpreet: [00:56:39] It's selfish.

Speaker3: [00:56:39] Sun don't anticipate the future that's coming. Just build the future

Harpreet: [00:56:42] You want and think of

Speaker3: [00:56:44] Where you

Harpreet: [00:56:44] Need augmentation. I see some

Harpreet: [00:56:46] Good feedback here from, uh, somebody said Spotify recommendations. Yeah. And he said that Spotify, those recommendation engines have been amazing to Spotify. Always hooked me up with all the gems.

Harpreet: [00:56:58] All right.

Harpreet: [00:56:58] So I guess that somebody else had their hand up for question. Was that Cristoff or was Alysha? I should go for it. Just go for it.

Speaker4: [00:57:05] Just kind of question before we go. Then I'll go

Speaker3: [00:57:11] And. This question and to you mentioned that you had the lock and you said it in your podcast and I think in some office hours also and I've been reading these two books right now, thinking and bed and the psychology of money. And the common thing they both have is, is that

Harpreet: [00:57:36] Luck is there.

Speaker3: [00:57:38] I mean, if there is an outcome, there must be luck. And since you started it

Harpreet: [00:57:44] And my question is,

Speaker3: [00:57:45] How do you apply luck into your decision making?

Harpreet: [00:57:48] So you just factor for things that are not going to be in your control. Right. So any time I make a decision, I just acknowledge that there are things outside of

Harpreet: [00:57:56] My control and there are things that I

Harpreet: [00:57:57] Can't know will happen. And [00:58:00] I optimize on the things that are within my control, the things that I can influence and move the needle on those as as much as possible. So that's kind of how I factored into my decision making. Right. And to the point I was making earlier about not being attached to that outcome is I kind of imagined my head. Okay, if I take these sets of actions, here's what I think the probability distribution is going to look like for potential outcomes. Right. And this is the this is not like the frequenters approach to statistics or probability. This is just probability. And in the subjective definition of it. Right. That it is my personal belief, like, you know, kind of like a Bayesian reasoning type of way, Bayesian epistemology, where say, OK, this is my belief that these are the potential outcomes that could happen from this decision. Right. So if I take this set of actions, then, you know, in a thousand parallel universes, I would imagine that 40 percent of them will have a positive outcome, 10 percent of them will have a neutral outcome and, you know, 50 percent chance of having a unfair outcome or unfavorable outcome. I mean, that's kind of how I use it in my thinking. My decision making is, OK, what can I influence, what is directly in my control? What are the things that I can personally impact, optimize for those things, those actions, those activities. Consider what potential things can happen and then assign a probability weighted distribution to each one of those kind of potential outcomes. That's how I think about things.

Harpreet: [00:59:26] So how do I

Harpreet: [00:59:27] Account for luck in there? Um, I just realized that luck plays a factor in anything, right? There are things just beyond my control that just will happen. And if they do happen, then shit, man, I can't be bad. I did everything that was in my control. So you know that answer your question. Was that in a roundabout? Let me know

Harpreet: [00:59:44] If there is a great answer.

Harpreet: [00:59:48] I mean, when it comes to luck, there's like different different types of luck, right? There's the dumb luck.

Harpreet: [00:59:51] Blind luck.

Harpreet: [00:59:52] Things just happen. Right. I can go buy a lottery ticket and just win the lottery, whatever. Like, that's just dumb luck. Like I was born in the [01:00:00] United States of America as opposed to India or Fiji. That's dumb luck for me because now I've got a bunch more opportunities than other people do. Um, then there's the the fortune favors the bold type of luck that hustle luck.

Harpreet: [01:00:11] We just out there doing a bunch of things,

Harpreet: [01:00:14] Kicking up dust. Things are combining, things are happening. And you're able to capitalize on these opportunities that are presented because of the actions that you're taking. Then there's the luck that is, um, like three luck, which is more about becoming really, really good in a field. Right. That chance favors the prepared mind. That's a cliche that goes with that. We just really, really understand a field so well where you could see opportunities happening that other people can't. We see things that are developing that other people won't be able to put together. That's kind of like where we're bend's out with that. He's got that type three type of way on the cutting edge stuff. He could see things before any of us could. And then there's type four left, which is like the most unique, interesting type of like where that you just develop a unique brand, a unique mindset, unique reputation, where that people come to you now when they want to do something because you've been able to create this type of luck for yourself. And that's the type like you want to maximize.

Speaker3: [01:01:11] I think luck is relative to you. So sometimes

Harpreet: [01:01:13] You can sometimes you can

Speaker3: [01:01:14] Regret mistakes made in the past like crap. I didn't do this job and now the iPod and I'm an idiot like. So you don't want to dwell on that. And, you know, when it really comes down to it, we are all lucky to be here today,

Harpreet: [01:01:26] Breathing, able to

Speaker3: [01:01:27] Communicate on this call in. The more you really own that, the reality, how lucky you are to just be here, the more you're willing to take risks. And so they've done studies to show that people that intentionally leave their jobs from there to pursue new opportunities on their own, they will make more money in their career than

Harpreet: [01:01:43] People that wait to be promoted within a company. And so being

Speaker3: [01:01:46] Bold, it's hard to be bold, like why would you ever risk losing your job or switching jobs? What if what if you joined a new company and or purchased on your company? What if you what if turns out you hit the culture like what? It's like six months from now, you're like there are more assholes [01:02:00] in there than you realize and there's always a risk. But the more you do

Harpreet: [01:02:04] This, the more luck will come your way,

Speaker3: [01:02:06] The more people you meet, people in your network are a huge factor. And look, when it comes to startups, jobs. A lot of the a lot of the most exciting positions that are offered are one to one like, hey, new company, come work a new company, come work in your company.

Harpreet: [01:02:18] And I know you can work here.

Harpreet: [01:02:20] That's really been, Russell, that

Harpreet: [01:02:22] I like this thing. You put your

Harpreet: [01:02:24] Time. I look at the fact that luck is a fallacy. Perhaps. Go for it.

Speaker3: [01:02:28] Yes, I'll just step onto it is a fallacy. If we try to look at it in very logical terms and say that is this random chance that happens everywhere in the universe outside of planet Earth. But we experience it directly. And when it does, then there's the ability to observe. And then let's say there is serendipity as well, where things just coalesce for a greater benefit than you would you would expect in a normal distribution. So so I posit that those with a more positive disposition are likely to observe more objectively and more closely to things that are going on around them. And with that random chance, they'll likely be able to find things that can provide a positive outcome to them more than some other things who say someone's really angry or upset about something and just do something that's consuming their minds. They might be taking in everything around themselves. More likely goes beyond something so external. Let's have a look at them and say, well,

Harpreet: [01:03:27] The more

Speaker3: [01:03:28] Positive person was lucky because I found this thing and the negative person was unlucky because I didn't get it may be attributed to the way that they were observing at the time, you know?

Harpreet: [01:03:39] Yeah, I absolutely love that do. That's actually very similar to a Christian Bush talks about in his book, The Serendipity Mindset, which, by the way, if you haven't read the book or read the book, but also listen to a podcast so that we did together, we talk about this as well, but very much so. Just having that open mindset, the open open to new possibilities is very, very helpful. There's [01:04:00] another book I think was called The Luck Factor. I'm Martin something. I can't remember his name, but they had this study where they had two people, one who is self described himself as lucky, one who self described themselves as unlucky. And they had them both go into like a

Harpreet: [01:04:14] Coffee shop, then on the way to the

Harpreet: [01:04:16] Coffee shop, the same path they had dropped money there. So, you know, it was there. And then they they set up the situation at the coffee shop in such a way where, you know, you were forced to sit down with somebody and in conversation, whatever the lucky person found, the money on the ground sat down and started talking to whoever was next to them and realized that this person is like a businessman, can help them launch some product or something like that, whereas the unlucky person just totally stepped right out of the, you know, ten pound note that was on the ground and completely shut themselves off to any new opportunities that were going to be present in the coffee shop.

Harpreet: [01:04:52] Um, they are very, very, very great tachometer.

Harpreet: [01:04:54] Thank you, Russell.

Harpreet: [01:04:55] Um, a couple of a couple of things popping

Harpreet: [01:04:58] Up from LinkedIn chat. Some great comments from Christine, Christine, clarifying this idea of

Harpreet: [01:05:05] The

Harpreet: [01:05:05] Normative procedures that she was talking about. So I guess she says that social scientists adhere to certain normative procedures when conducting research to be able to publish in academic journals. Research on human subjects must meet standards in order to be considered legitimate science and acceptable within community norms and form what behaviors are appropriate and ethical norms. Also help us understand what procedures are useful to answer a problem using a set of data. That's yeah, I wonder if we have that data science. The closest thing I can think of is like Chris. Mm. Uh, methodology, but that's mostly for exploring Data and things like that.

Harpreet: [01:05:39] But I'm going to have

Harpreet: [01:05:40] To digest that and think about that. Thank you, Christine. Christine also says that can be the serendipity factor that happens to assist a person in realizing an outcome can be a belief that a positive outcome is possible, can be used with deductive, inductive and inductive reasoning, I believe infallible. We need to address how the result might [01:06:00] be fallible due to bias and prejudice, that, yes, all beliefs are fallible. Listen to the episode I did with Dave Gray. We talk all about beliefs. Um, and then I just type it out right here, man, if you wouldn't mind verbalizing it for, uh, for the audience listening.

Speaker3: [01:06:16] Yeah, I was leaving Intel and Micron and this is one Data science is really hot, like it was the thing in the market. And I had three job offers and I changed my mind every single day. And and you're trying to make decisions based on it. Be very careful who's giving you advice and not to be too negative, but your parents are coming from a very different mindset. They're going to be more focused on conservative maximum

Harpreet: [01:06:37] Cash, stuff like that. And I had these three job offers.

Speaker3: [01:06:41] I had no idea which one is going to go with. And luckily I talked to someone who'd worked at both these companies to two of the three and he said not to join his ex employers.

Harpreet: [01:06:50] I joined HireVue a year later.

Speaker3: [01:06:51] The first two completely failed in.

Harpreet: [01:06:53] Hirevue ended up

Speaker3: [01:06:54] Being a big

Harpreet: [01:06:54] Success, but

Speaker3: [01:06:56] There was no way to know that. And so that's why the networking comments there that get advice from as many people as you can but be be very aware of their bias. So if you're reaching out to your parents for advice, know that they will have a bias. If you're reaching out to others in the

Harpreet: [01:07:08] Community, they might have a bias.

Speaker3: [01:07:11] Yeah, that reminds me of luck because, like, I got very lucky. It could have gone the other way

Harpreet: [01:07:15] And yeah, I

Speaker3: [01:07:16] Don't know. Is that luck? I think it's like.

Harpreet: [01:07:18] Yeah, I mean, absolutely. I mean, you kind of created your luck in a sense, right? By taking some action and doing things right, you're influencing your luck by your actions. Great conversation over the last the last few minutes there. Yeah, having a good network is important. I think I underestimated the power of any network until very, very recently. You know, over the last year or so, I'm like, damn, like have a network is really important and it's been great. Put me in positions to, for example, be chatting with Ben one on one next week or a couple of weeks from now to

Harpreet: [01:07:50] Get some advice on how I can it in this new

Harpreet: [01:07:52] Role. That comic, you know, one on one call coming up with the with VIN again for more advice on how to just help comment [01:08:00] be as successful as possible in this role while making sure that I enjoy myself. So having these type of connections is it's awesome. And it wouldn't have happened if I didn't build a network and try to build the community.

Speaker3: [01:08:11] Yeah. And one thing about that is it's a lagging indicator. A lot of like business, the outcomes of that

Harpreet: [01:08:16] Are a lagging indicator. Right.

Speaker3: [01:08:17] Like, it's not something where you

Harpreet: [01:08:19] See that it's not a vending

Speaker3: [01:08:20] Machine. I think like this building, this sense of who you are in this community of people and what you want out of it is not a vending machine.

Harpreet: [01:08:26] So that's like some of these questions about like,

Speaker3: [01:08:28] Should I choose this or this? It's like you might not see it is a lagging indicator, like you're saying like, oh, man, like a year on after kind of like taking on this thing, like things start happening,

Harpreet: [01:08:39] Things start coming to fruition.

Speaker3: [01:08:41] And I saw that in my previous role as well, where it's like you have to sort of and to even take the advice from someone else and to know which advice to pay attention to, I think you have to have that sort of built up sense of yourself. Otherwise, it's so easy to oscillate between one person's opinion and another person's opinion and just be ping pong back and forth and never be able to put that advice and those sorts of things in context. So like building that sort of sense of yourself through community, through these different levers is like it ends up being a lagging indicator. But when it comes time to make those decisions and you can weigh in place appropriately based on what you also want, what you're building for yourself, like it's sort of this sort of an abstract idea, but it's super important to like, again, like it's not a vending machine. You can't just put in like X amount of hours into a decision and get out the outcome you want. It's like this thing that's built up over time and that sense of self.

Harpreet: [01:09:30] Yeah, I don't know.

Speaker3: [01:09:31] I feel that very strongly and it's hard to put into words, but it's like a true thing. And like I've seen it with you Harp like just in the we've been working together and myself as well. It's like it's really, really important stuff.

Harpreet: [01:09:40] Oh dude, I was thinking about this just last night. I thought we had to work so damn hard of, like, I'm doing all this stuff and now like, will the seeds ever, ever blossom? And then I was like, you know what? Typically when you plant a seed, you don't just get a tree that you could sit under in two weeks. Like it takes time, it takes energy, takes effort, and [01:10:00] you just keep working at it. And then opportunities do happen. Right. And and I was thinking I was like, you know, like, yeah, actually all the hard work and effort did pay off because I put myself in these positions to take advantage of these type opportunities that that pop up. Right.

Harpreet: [01:10:14] Um, yeah.

Harpreet: [01:10:16] Yeah. The point to that is, yeah, you can't plant a seed and then sit under shade in two weeks. It takes time. Been like this little anecdote you have here is anecdotal story and it would

Speaker3: [01:10:29] Be, but I was reacting to the lagging indicator. So I've had I've had things happen that were seven years later, two years later, you never know. Like I random talk. I gave a talk to university, to the math department. It's like the graduate math department. And I remember leaving that talk. I was angry. I was angry, like, why what

Harpreet: [01:10:46] Am I doing here? Took time away

Speaker3: [01:10:48] From the family and my wife's not like she's frustrated with me. And then two years later, someone reaches out. So say, I saw you in my math class. I'm a VP at Goldman Sachs, become present

Harpreet: [01:10:56] To 70 business analysts. We're like two years later.

Speaker3: [01:10:59] And then seven years later, that story of chat about Jeremy tried to hire this random person. And then seven years later, I sold my company to his bigger company.

Harpreet: [01:11:06] Like you

Speaker3: [01:11:08] Never know. You never know when it's going to come back around.

Harpreet: [01:11:11] Yeah. You like that. That we want it now type of mentality. You got to you got to put in effort and wait. Just let them have it happen. Right. What's the what's this thing I got to let go in the gut. I like that thing like I mean I thought that was interesting. Just do the work, handle business like go to the outcomes and like things that happened, like, you know, Ben's talking about eight years later, they just pays off massively.

Speaker3: [01:11:38] So there's a lot to be said to for giving one way. And maybe that's not a good way to say it. But like, if you are helping someone or doing something because like pretend if I come to you and if I say, hey, I'd like to present to your college and then you bring me

Harpreet: [01:11:49] In there and I

Speaker3: [01:11:50] Present, but I very much turn it to like a bi Data robot and like, come here and then you feel it. You kind of feel like you've been like a shady salesman, like I've kind of swapped [01:12:00] the value prop. But the more you can kind of give one

Harpreet: [01:12:01] Hundred percent with nothing in

Speaker3: [01:12:03] Return, that's when I've been so surprised when it comes back around.

Harpreet: [01:12:06] So see how

Speaker3: [01:12:07] You can help people in community be one hundred. Authentic and never have like this tit for tat

Harpreet: [01:12:12] Approach, like they are free

Speaker3: [01:12:14] To come visit you when I'm in town, but only if you sit through Data about demo like that would be like, no,

Harpreet: [01:12:20] Go away.

Speaker3: [01:12:20] Like your girls don't want to talk to you like it's all about these authentic human connections first.

Harpreet: [01:12:25] Yeah, that's how you

Harpreet: [01:12:26] Start putting positive sum game tonight. That's how everyone starts winning. I know you had a question that said close it off with that and then we'll wrap up the session. I don't see any other questions coming in from YouTube, Twitch or LinkedIn. So we'll wrap it up

Harpreet: [01:12:40] With Osseous question. And my question

Speaker4: [01:12:42] Was a model validation when you built it. So my process this far, I'm pretty sure it's I mean, I've Googled it, but I'm pretty sure mine is very effective.

Harpreet: [01:12:51] My first step is to build a

Speaker4: [01:12:53] Model, the easiest icon. Then the next step is to build something more complicated that needs to be done. Then after that, I try and see if there's a third option. Are there any tips and tricks you've picked up along the way in terms of model validation?

Harpreet: [01:13:05] Yeah, I think that's kind of like the the artistic side of it. You can be a little bit creative the way you do that. So definitely I like the first step is creating a baseline, right. A naive baseline. So what I will do is I'll create like the let's just assume we're working on regression problem and I'll create the most, uh, the absolute most naive baseline, which is I'm going to predict the mean value of my train set for the test that. Right. And then get the error term for that very, very naive. Then from that, I'll probably go into another just super simple method. And maybe that's super simple method is just linear regression with maybe L1 regularization. Right. Shrinking unimportant terms to zero and see how that improves over the baseline. So now I have two baselines. Then from there I'll start with maybe three or four different candidate models. Like I'll have an intuition like, OK, well engineer the features this way. And I've, I've done this. I kind of feel like that the way I've engineered the features lends itself [01:14:00] to pre based models.

Harpreet: [01:14:01] But maybe I'll try like four or five different tree based models from there, whichever ones end up being the top performer. If there's one top performer, then I'll just take that top performer and fine tune that. And so this stage I just talked about, you have spot checking candidate models like these are like out of the box default type of parameters that I'm filling them with. I haven't optimized anything. So if there's a tie, then I'll take both of them. There's two models that TYL take both of them and then I'll tune both of them. And then just sort of the average prediction and then maybe later in production, I will test both models. So I'll serve the prediction on the end. But then on the back end, I'm collecting data on how each individual model had performed and then I'll do some statistical test to see which one actually does a better job. That's the fitting. And then if if if the results warranted, then just swap out the average model for the one that is better.

Speaker3: [01:14:59] Um, then what do you think? I love this question. I was going to say real quick, I've mistakes. I've made mistakes in the past where I try to validate the model for me and that's very short sighted. How do I validate the model for who matters and who matters is not the data scientist who matters is going to be the domain expert, because you and I could be having you and I could be celebrating and saying, oh my gosh, can you believe that our eighty six point nine three who knew this is going to change the business like we're going out for drinks. We just can't believe how awesome this is. If we go show the domain expert that model, they might look at the feature importance impact, like what is driving your model. They don't know what that statistical metric means, but they might look at it and say, I absolutely hate this because you have customer color as a leading indicator to churn. And that's bullshit in the data science team. Don't even like what are you talking about? And so I'm a huge fan and get or sometimes you're in a same scenario where you're talking to and executives like, hey, the models, that's accurate. Do you think we should turn it on? Like you're actually asking the executives? And that is totally wrong mindset. Like you need to take ownership and so get the subject matter expert in the room, have them validate what's going into the model and [01:16:00] then see if they can translate the output of the model to not a statistical metric. But how would you measure this KPI? How are you measuring this KPI today without the model and how would you defend value for this? Because you win if you have a speech defending your model with a dollar sign, and that's really hard to get there, but you can get there with assumptions. But most people I've made the mistake that I don't even think about them. I'm just like a marketing sales easy. I'll just like build this model in sales force. Like, I don't need people to help me. And that's where you get in trouble as

Harpreet: [01:16:25] An excellent point. And like real world example, something I do not price. Like, I built a model and it was getting decent results, but I was talking with the people who are actually using the model and they noticed that I was using a particular feature in there. And that's a particular feature was the basis for other features that I had engineered. And they're like, oh, we don't like that. Yes, it is ranking as the most important features. Yes. But we don't want to include that at all. Like like we would rather have a less accurate model than include those particular features. And, um, so we had to rebuild everything. We wasn't terrible. But still, just to Ben's point, that's. Yeah. Get the subject matter experts involved often. This is a, uh, it's a great point I'm making here. We talked about this last week as well. Go for.

Speaker3: [01:17:10] Yeah, so it's just that this was very much referenced in the industry and we talk a lot about domain expertize because the female vote was collaboration and Data science machine learning and tooling around that, it processes around that. And the CEO of radio, which I was talking about, is this guy that helps you build models and sort of test versions of them very quickly in the sort of like interface webapp type of environment. And he was talking about how there's like an echocardiogram type model that mimic this process. And the domain experts knew how to test how to test ideas like pacemakers, basically, and how to when they were testing the model, they knew the edge kind of edge conditions or education they wanted to test against, like, you know, like, OK, my expertize. Is the model able to perform something I think is difficult as a domain expert. And if it was, it's like, holy shit that actually works [01:18:00] versus like you're just testing the model's predictions and accuracy in these metrics that are sort of divorced from that domain expertize. And it's really tough to know, like, is the end user actually going to be able to use this in any intelligible way? So I think like collaboration in terms of machine learning and model development has to involve that end user to some extent or another, whether it's in the medical field or elsewhere. It's like, you know, if you like, just kind of I was interested in this idea of products that have a name because like how are you testing? How are you actually bringing in your end user to to validate and that you can put a first version of a model in front of someone and get so much more feedback about what it actually needs to be doing as opposed to just running it against the validation test.

Speaker3: [01:18:42] So it's like optimizing the model for real world conditions and not just a validation set, because the validation set does not include that range of domain expertize that's actually needed to see. Is this going to work in real world settings? Is this going to work as a product that people can use that automate those things that are like what do I want to do less of and what do I want to do better? That sort of thing? You're talking about Ben, earlier, too. It's like that's those are the conditions under which we can iterate and develop more quickly. That struck me as a really great point that I hadn't thought of. And so it's just something I wanted to throw in there on that. And that point is, well, I love all that. I to stress MSG's sprinkles on top of that, I'd say when you meet with the SME, you should say, how is this model going to get you promoted? But like, how am I going to get you promoted? Like, let me understand your OK as your KPIs. Like you'll learn more in an hour meeting with this man than you will in six hours with an army of PhDs trying to bash their brains together about that. Yeah, the dumb things we do in the past that we'll never do again. Right.

Harpreet: [01:19:36] And just for clarity, anybody wondering Smy that's s m e subject matter expert. Right. Any follow up questions or follow up comments or anything on that note?

Speaker4: [01:19:47] Thank you. That has helped they say so in effective ways and the right. But apart from the asking questions that have been skipping that beat a lot,

Harpreet: [01:19:55] Don't see it that bit. That's very important. I guys. So let's go ahead and wrap it up. Thanks, [01:20:00] everybody, for joining in. Definitely be sure to check that episode that are released earlier this week on Friday with Lillian Pearson. Got one next week with Jonathan Tesser. I think everybody here might be familiar with Jonathan. He's quite big on LinkedIn. So that was great chatting with him then. More and more interesting episodes coming out then. Thank you so much for for joining in with us here today. Shout out to everybody else joining Austin, Chris Russell and one of those here in the chat and on LinkedIn. Appreciate having you guys here. I will go ahead and wrap it up, my friends. Remember, you've got one life on this planet. Why not try to do something? Here's one.