Jeremy Howard Interviews Kaggle Grandmaster Sanyam Bhutani

00:00:00.000 | >> Hi, this is Jeremy Howard,

00:00:01.640 | and you're listening to Coffee Time Data Science,

00:00:04.240 | a podcast for data science enthusiasts where I interview

00:00:07.040 | practitioners, researchers, and Kagglers about

00:00:09.920 | their journey experience and talk all things data science.

00:00:13.440 | Before we begin, I apologize for the change to our schedule.

00:00:17.120 | Of course, usually you would be seeing

00:00:18.840 | Chai Time Data Science on this channel with Sanyam Bhutani.

00:00:22.360 | Unfortunately, he's not available today.

00:00:24.400 | He had a prior appointment on

00:00:25.920 | another podcast and he was not able to join Chai Time Data Science.

00:00:29.560 | So we hope you enjoy this special episode of Coffee Time and Data Science.

00:00:34.240 | Without further ado, I would like to invite our very special VIP guest,

00:00:39.680 | newly anointed Kaggle Grandmaster, Sanyam Bhutani.

00:00:45.080 | Sanyam, welcome to Coffee Time Data Science.

00:00:47.880 | >> Thank you, Jeremy. Usually, I'm very anti-coffee,

00:00:50.840 | but I'll have to allow that.

00:00:52.960 | I still can't believe you weren't kidding.

00:00:55.800 | I mentioned in our message also,

00:00:57.680 | I think I don't deserve this, but thank you.

00:01:00.000 | Thank you for doing this. Of course, I couldn't say no.

00:01:02.400 | >> Oh, it's my great pleasure.

00:01:04.240 | Thank you for agreeing to be an inaugural guest on our show,

00:01:09.360 | and apologies for the inappropriate choice of beverage,

00:01:14.280 | but this chalier coffee is our national drink, so.

00:01:19.080 | >> Actually, at Voke,

00:01:21.240 | we were having this funny thread where people were pulling my leg,

00:01:24.040 | when you shared the expresso to eat,

00:01:27.800 | and everyone was like, "Hey, see Jeremy's drinking coffee.

00:01:30.080 | You should switch over now."

00:01:32.120 | >> I will say though, I just purchased

00:01:35.240 | three different styles of oolong tea from China.

00:01:39.280 | So I also like my coffee.

00:01:41.720 | It's very hard to find good tea here though, honestly.

00:01:44.480 | So I actually had to get it all specially imported.

00:01:48.200 | So I would certainly love to hear about your tips on Indian tea,

00:01:53.600 | because I'm much more familiar with the Chinese variety.

00:01:57.200 | >> My mom makes it for me.

00:01:59.200 | So every time I have a podcast,

00:02:00.840 | she makes it for me even today.

00:02:02.760 | She woke up before me.

00:02:03.920 | She made two chai, one for prep, one for now, so.

00:02:08.040 | >> Great. So that's great for our listeners.

00:02:10.360 | If you want a nice cup of chai,

00:02:11.880 | head over to Sanyam Bhutani's mom's house,

00:02:14.840 | and grab yourself a nice cup of tea there.

00:02:19.080 | Terrific. Well, let's talk about data science.

00:02:24.760 | As we said, I guess the stimulus for doing this interview

00:02:33.240 | is your recent Kaggle Grandmaster anointment.

00:02:38.360 | So why don't we start there?

00:02:39.960 | What are you a grandmaster of,

00:02:42.160 | and how did you achieve that lofty status?

00:02:46.360 | >> So as you can see, I'm talking a lot,

00:02:49.600 | and I've become a grandmaster in the discussions here.

00:02:52.880 | So I'm only right now good at talking.

00:02:55.280 | Kaggle's belief from left to right

00:02:57.600 | is the hardest to easiest category.

00:03:00.640 | Jeremy was the first,

00:03:02.120 | I think Kaggle Grandmaster,

00:03:03.760 | the first ranked in competition.

00:03:05.200 | So I'm trying to move towards the left end slowly and slowly.

00:03:08.520 | >> I would say datasets contributor,

00:03:11.000 | maybe not so hard.

00:03:13.080 | There are probably ways you could game that, I reckon.

00:03:16.360 | For example, I bet if I put all the datasets

00:03:18.520 | from the course on Kaggle,

00:03:19.840 | I would become a datasets grandmaster.

00:03:22.880 | I'm one gold away from being a Kernels Grandmaster,

00:03:27.400 | which I should definitely get back to doing Kernels,

00:03:33.760 | because they're so fun.

00:03:35.520 | So when I ran the Masters Machine Learning course at USF,

00:03:42.880 | I actually, the marking for the course

00:03:46.280 | was based on how many points you got on Kaggle competitions,

00:03:49.800 | Kaggle discussions, and Kaggle Kernels.

00:03:52.680 | I always said to them, if you can write in a way

00:03:56.040 | that people find compelling,

00:03:57.440 | explain data science in a way that people find useful,

00:04:00.520 | or build bottles that are highly predictive,

00:04:02.800 | then you've got a good opportunities ahead of you.

00:04:08.880 | >> I think sometimes kernels are a bit flaky.

00:04:12.240 | I've been spoiled with this box,

00:04:14.320 | and I'm so used to things running instantly.

00:04:16.880 | Sometimes I get a bit annoyed,

00:04:18.360 | but people sharing stuff on there is awesome,

00:04:21.120 | and I always learn so much.

00:04:22.360 | Every time I go on the kernel stuff,

00:04:24.000 | the Kaggle community makes it super awesome.

00:04:26.560 | >> What's the flaky bit?

00:04:28.520 | >> I'm just used to running cells super fast,

00:04:31.720 | like Kaggle kernel takes longer to start.

00:04:34.000 | >> In the actual compute environment, I mean,

00:04:35.920 | I treat it as a platform for

00:04:38.320 | communicating to people rather than a compute environment.

00:04:42.200 | >> For sure. What are you up to when you're

00:04:46.960 | not being useful on Kaggle discussions?

00:04:50.880 | Where are you working now?

00:04:52.640 | >> I work at Meets and Biases,

00:04:54.760 | and we have an awesome fast A culture internally as well.

00:04:57.840 | It's like working with friends,

00:05:00.000 | working on site projects with friends.

00:05:02.080 | I lead our community efforts,

00:05:04.120 | which basically means I just do what I love.

00:05:06.320 | I host live stream for the broader community,

00:05:09.760 | our CEO, Lucas B World,

00:05:12.120 | whom you know and our Head of Growth,

00:05:14.360 | Lavanya, whose team I work on.

00:05:15.800 | Both of them are super supportive of community stuff,

00:05:18.640 | so they just let me do whatever I want.

00:05:21.200 | Every time I go with a stupid idea up to Lavanya,

00:05:24.280 | taking a stupid idea to her,

00:05:25.680 | she's like, "This sounds good to me.

00:05:27.240 | If it sounds good to you, just go ahead with this."

00:05:29.080 | So I get to do what I love,

00:05:31.360 | and for some reason, they pay me for it.

00:05:33.760 | >> Yeah. Well, Lucas is a super awesome human being

00:05:37.280 | and also a very smart guy.

00:05:39.720 | I know a lot of the fast AI community

00:05:41.920 | have made their way over to Weights and Biases

00:05:44.040 | and hopefully are doing positive things in that company,

00:05:48.400 | which is already going very well.

00:05:50.560 | I know it's actually in terms of all of

00:05:54.120 | the logging frameworks and deep learning,

00:05:57.800 | it's the best integrated with the fast AI library

00:06:00.960 | and the most loved by some degree,

00:06:06.640 | which I guess is why it's the best integrated,

00:06:09.040 | so actually genuinely a good product.

00:06:13.240 | Now, I first came across you when,

00:06:18.320 | I think when you posted a message in

00:06:20.760 | the fast AI forums a few years ago,

00:06:23.920 | introducing yourself and saying you're interested

00:06:26.920 | in learning about deep learning.

00:06:29.600 | Now, remind me, when would that have been? What year?

00:06:33.160 | >> I'm so sorry. Just to add

00:06:34.920 | one small point to your previous question.

00:06:37.040 | When I was interviewing for Weights and Biases,

00:06:39.280 | I showed this to Lucas and I was going ahead and explaining

00:06:42.520 | why I spent so much time on our forums and he just,

00:06:44.800 | he stopped me there. He was super appreciative of this,

00:06:47.560 | so he instantly recognized that I'm from fast AI.

00:06:50.920 | Now, I get to work on the best experiment tracking tools team.

00:06:55.520 | But sorry to answer your question,

00:06:57.920 | I got the opportunity to

00:07:00.240 | join the International Fellowship in 2017.

00:07:02.760 | I was really struggling with the university.

00:07:04.600 | I mean, I was doing okay with my grades.

00:07:06.600 | >> So what were you studying at university?

00:07:09.480 | >> This is one of my most controversial messages I put out.

00:07:13.640 | I was studying computer science,

00:07:14.880 | but I wasn't becoming a better programmer,

00:07:16.800 | which is how I envisioned it to be.

00:07:18.600 | >> At what university was that?

00:07:20.320 | >> That was SRM University,

00:07:22.320 | so it's one of the good-known universities in India.

00:07:26.120 | I went there expecting to just become better at programming,

00:07:31.560 | and I just didn't like the syllabus.

00:07:34.120 | I was trying every single thing.

00:07:35.600 | I was signing up for every student club

00:07:37.560 | and I just couldn't enjoy it.

00:07:39.600 | Then I shifted to online courses,

00:07:41.480 | was trying to find something interesting there,

00:07:43.280 | and somehow magically I landed on fast AI,

00:07:45.640 | although initially I was terrified of it.

00:07:47.720 | But for some reason I decided to-

00:07:49.600 | >> So let me just dig in a little bit more,

00:07:51.720 | because I'm very interested, because I think a lot of people

00:07:53.320 | watching will be people who,

00:07:56.600 | like a lot of data scientists or people doing

00:07:58.720 | data science don't have a computer science background.

00:08:01.240 | I think a lot of them,

00:08:03.000 | I certainly used to be like this,

00:08:04.680 | looks over at the computer science world and thinks like,

00:08:07.640 | "Oh, I probably should have done a computer science degree.

00:08:11.800 | Maybe I should go back to university.

00:08:13.760 | I'm never going to understand anything.

00:08:15.120 | I'm not a real computer scientist."

00:08:16.880 | So you're saying that's not necessary,

00:08:20.480 | and in fact you're saying you didn't

00:08:21.800 | learn much useful programming.

00:08:24.000 | So tell us more.

00:08:25.080 | What were you doing in the computer science course?

00:08:28.040 | What were you learning and in what way was it not satisfying?

00:08:32.840 | >> One of the things I learned

00:08:34.840 | afterwards was any computer science degree

00:08:38.400 | is more of double ECS,

00:08:40.080 | so electrical engineering and computer science.

00:08:42.040 | Half of it is just learning about diodes, stuff like that,

00:08:45.120 | and that was absolutely terrifying to me.

00:08:47.040 | I had no interest in it at all.

00:08:49.360 | I just barely passed in that subject.

00:08:52.920 | I remember going to the professor and just asking them,

00:08:55.240 | "Hey, please don't fail me in this.

00:08:56.440 | I don't want to take this class again."

00:08:58.640 | >> Does that work?

00:09:00.360 | Is he like, "Okay, I'll pass you."

00:09:02.440 | >> Worked for a few subjects for me.

00:09:05.440 | But I looked at the programming courses.

00:09:10.360 | First-year syllabus in India is

00:09:13.160 | usually just the basics of everything.

00:09:15.440 | So they teach you chemistry,

00:09:16.520 | they didn't even teach you biology and computer science,

00:09:19.000 | and I just didn't like it.

00:09:20.320 | I decided to go to the library.

00:09:22.120 | I would pick out the senior yearbooks.

00:09:24.360 | I would hang out in the section where

00:09:25.560 | all the sophomores were hanging out,

00:09:27.440 | and they were doing teenage stuff there.

00:09:29.200 | I was the only one with a book in the library in a corner,

00:09:32.840 | and even that was very outdated syllabus.

00:09:36.360 | So it was just very mundane stuff,

00:09:38.680 | and I couldn't see the stuff connecting to real world.

00:09:42.080 | So I would talk to people on the Internet,

00:09:43.840 | they were building amazing stuff,

00:09:45.120 | and I just didn't see the connect happening at all.

00:09:48.080 | >> Yeah. I mean, I remember,

00:09:49.480 | so I ended up majoring in philosophy at university,

00:09:52.680 | but I did try many things along the way to try and

00:09:55.360 | find something that seemed more interesting,

00:09:58.080 | and I did do a computer science subject.

00:10:02.320 | It was like a programming and statistics class,

00:10:09.720 | and that programming was in Pascal,

00:10:12.240 | and I didn't understand any of it,

00:10:14.560 | which I found surprising because I knew how to code.

00:10:19.160 | I remember going, I found that textbook 10 years later,

00:10:24.560 | and I went back and I reread it,

00:10:26.680 | and I was like, even though I know it,

00:10:29.720 | I've written production software in Pascal at that point in Delphi,

00:10:33.360 | I still didn't understand the book,

00:10:34.920 | and I realized like, okay,

00:10:36.160 | the problem is not me.

00:10:37.880 | The problem is the book.

00:10:39.520 | A lot of academics just aren't good teachers.

00:10:44.960 | They make things just so hard to understand.

00:10:49.720 | >> That realization came so late to me,

00:10:52.400 | like two years into the degree,

00:10:55.240 | but still like that was super late for me.

00:10:57.040 | I was just expecting my teachers to be super good at everything,

00:11:00.200 | and I would like go to them with questions.

00:11:01.880 | Sometimes I just wouldn't get answers,

00:11:03.400 | and that would really annoy me.

00:11:04.680 | Like how do they not know this,

00:11:06.120 | like they're supposed to know this stuff.

00:11:08.040 | I remember in like one of my machine learning courses,

00:11:12.720 | there was a question in an exam about what's the latest here,

00:11:17.240 | and I had read a paper.

00:11:19.280 | I drew that architecture.

00:11:20.440 | I was super proud about that,

00:11:21.920 | and I got a zero because they said,

00:11:23.440 | "This is not in the syllabus.

00:11:24.560 | You're not supposed to write this."

00:11:26.080 | >> Oh, no, yeah. Oh my God.

00:11:29.360 | I mean, it's not just university, honestly.

00:11:31.440 | My daughter is six,

00:11:33.320 | and so she's doing primary school,

00:11:36.880 | and it's same issues.

00:11:39.440 | They're like, "You're only allowed to be covering this part."

00:11:42.840 | We're talking about fractions,

00:11:45.160 | but you're allowed to talk about halves,

00:11:46.400 | and if you mention quarters,

00:11:48.160 | you're going to get told to stay in your lane.

00:11:53.800 | It's bonkers.

00:11:56.920 | You were looking for other material,

00:12:03.480 | for other things that might help you learn more pragmatic,

00:12:08.160 | how to code, how to become a good programmer,

00:12:10.640 | and so it was during that search that you came across fast.ai?

00:12:15.560 | >> I first found machine learning,

00:12:18.600 | and then I tried.

00:12:19.600 | I used to proudly say to my peers,

00:12:21.920 | because it was like a status symbol to me,

00:12:24.120 | just a turning group, your effect rate.

00:12:25.800 | I was at the initial bit,

00:12:27.080 | and I would just tell them proudly,

00:12:28.360 | "Hey, I did like 40 online courses,

00:12:30.200 | and then I would go ahead and sign for five more."

00:12:32.400 | >> So it's just like Oceara,

00:12:34.000 | or Udemy, or stuff like that?

00:12:35.640 | >> Every single thing, you name it,

00:12:37.320 | I probably would have at least watched five minutes of it,

00:12:40.240 | and then they did nothing.

00:12:42.520 | I wrote an article how not to do fast.ai,

00:12:46.280 | where I would say, I would discover something,

00:12:48.440 | find a course, study the course,

00:12:50.200 | and not be able to build something,

00:12:52.280 | and during the search somewhere in between,

00:12:54.600 | I found fast.ai, and I signed up for fast.ai, luckily.

00:12:58.120 | >> What was that experience like?

00:13:03.840 | Was that different to other courses you're taking?

00:13:07.040 | >> I could feel like this was the first time I felt

00:13:10.120 | that I could build stuff that actually works,

00:13:12.840 | so it's not just those three layered neural networks.

00:13:15.640 | I remember in the first two lectures,

00:13:19.080 | you showed us how to get started on a Kaggle competition.

00:13:21.960 | We put out a simple baseline, and we were beating you.

00:13:24.440 | So I remember me and my roommate

00:13:26.600 | were taking the course together,

00:13:27.680 | and we both were jumping for 10 minutes.

00:13:29.840 | "Hey, Jeremy, how is this happening?"

00:13:32.880 | on the leaderboard.

00:13:35.240 | So here is the aforementioned hownottodofast.ai,

00:13:39.800 | which I do remember.

00:13:42.080 | It's very, very helpful,

00:13:43.280 | and I've shared this with other people,

00:13:45.280 | and I think Radek has also shared some of these ideas.

00:13:50.280 | It's interesting to me, yeah,

00:13:56.400 | it's interesting like how few,

00:14:00.600 | even machine learning courses are top-down.

00:14:02.880 | So I always feel like if you're not creating useful models

00:14:07.880 | reasonably quickly, then that's pretty discouraging.

00:14:12.880 | And also, how are you going to know

00:14:14.200 | what's the point of the stuff you're learning about

00:14:16.400 | if you're not actually training useful models?

00:14:18.280 | It seems like it's pretty hard to integrate

00:14:21.480 | the knowledge that you're being taught.

00:14:24.480 | >> All of the courses would--

00:14:27.760 | Sorry. >> Go on.

00:14:29.640 | >> All of the courses would like,

00:14:31.400 | now most of them are better,

00:14:33.160 | but at that time they would start with

00:14:35.080 | showing this super cool style transfer example,

00:14:39.880 | and then they would go back

00:14:41.080 | to just teaching the mundane stuff,

00:14:43.520 | how you do stuff in Nampai.

00:14:45.920 | I was so done with that at that time,

00:14:48.240 | and Fast.ai was super awesome

00:14:50.680 | because there was also this insane community,

00:14:52.760 | and usually when you take a course,

00:14:54.720 | you're in a similar cohort,

00:14:56.480 | which is one thing I didn't like about university.

00:14:58.520 | No one around me was talking about machine learning.

00:15:00.680 | They thought I'm super weird.

00:15:02.600 | I thought they're super weird.

00:15:04.320 | But on Fast.ai, we had--

00:15:05.680 | >> What were they wanting to do?

00:15:08.080 | If they're not interested in ML,

00:15:10.000 | what were their hopes and plans and interests?

00:15:14.120 | >> Just do T-Net stuff, as I mentioned.

00:15:17.120 | >> So it wasn't just ML,

00:15:18.000 | but it's more just actually studying

00:15:21.120 | and learning effectively in general.

00:15:22.920 | >> Yeah, and most of them were just building websites.

00:15:27.400 | I didn't enjoy that bit at all.

00:15:29.040 | I would just download templates

00:15:30.520 | and hack stuff together every time I wanted to do that.

00:15:33.680 | >> Yeah, building websites

00:15:36.400 | is not particularly intellectually interesting

00:15:39.520 | of itself necessarily for sure.

00:15:41.360 | So yeah, tell me about,

00:15:43.520 | okay, so you got started and you said 2017, right?

00:15:47.280 | So how did that go from there?

00:15:49.920 | Was that then all smooth sailing,

00:15:52.000 | or did you hit any obstacles along the way,

00:15:54.800 | like anything that in hindsight,

00:15:56.560 | you wish you had done differently?

00:16:00.720 | >> So you had already pointed to my article,

00:16:03.800 | and I also was asking this question

00:16:06.080 | when I had the opportunity to interview Rachel,

00:16:08.880 | because I'm so new to top-down learning still.

00:16:12.520 | In my entire student life,

00:16:13.840 | which is 15 years I've been studying in the bottom-up way,

00:16:17.000 | and I was so new to top-down,

00:16:18.400 | I would always default to that.

00:16:20.640 | So my issue was not listening to you enough.

00:16:24.960 | The success I have was because I listened to you

00:16:27.080 | 10% of the time, and other time,

00:16:28.920 | I would just like, okay, this is not working,

00:16:31.480 | I need to go and read the basics,

00:16:33.840 | or again, default to bottom-up learning.

00:16:37.240 | >> Exactly like Radek Khosmalski.

00:16:39.880 | He says the same thing, yeah.

00:16:41.520 | >> We both also talk a lot about this.

00:16:45.320 | He even, we spoke about this for half an hour

00:16:47.880 | when I had the chance to interview him.

00:16:50.320 | Even in this book, I think this is really a struggle,

00:16:53.520 | because we always start with aspiring to be someone,

00:16:57.440 | and then we don't want to put in a lot of effort initially,

00:17:01.400 | and you don't see the dot connecting immediately,

00:17:03.480 | although it takes an insanely long amount of time.

00:17:07.040 | >> So can I ask, and feel free not to answer

00:17:09.080 | if this is too private,

00:17:10.320 | but how old were you in 2017

00:17:12.600 | when you joined the Fast AI community?

00:17:14.400 | >> So five years ago, I was 19.

00:17:17.800 | >> 19, okay.

00:17:19.240 | 'Cause I ask because I,

00:17:23.760 | you definitely stood out to me even then, you know?

00:17:27.640 | Like, and I couldn't exactly tell you why or how,

00:17:32.000 | but you certainly came across as,

00:17:34.380 | you know, unusually motivated.

00:17:38.760 | Now I will say,

00:17:39.960 | that's not that unusual.

00:17:42.760 | There's plenty of people who joined the course

00:17:44.840 | sounding pretty motivated,

00:17:46.560 | but then I would say like the big difference with you,

00:17:49.640 | the really big difference with you is your tenacity.

00:17:53.840 | Like you didn't go away.

00:17:57.060 | Like, you know, like if the multiple 100,000 people

00:18:00.960 | who start the course,

00:18:02.260 | you know, seeing it through over a multiple year period

00:18:08.400 | and doing the things,

00:18:12.040 | like I remember you would even post on the forums

00:18:13.920 | and say like, you know,

00:18:16.360 | in order to have something to force myself to live up to,

00:18:20.280 | I'm going to tell you what I'm going to do

00:18:21.460 | over the next few months.

00:18:23.640 | Like, there it is.

00:18:26.060 | What, do you have a sense of like, what's,

00:18:29.880 | how did you get to this point?

00:18:32.560 | Why are you more tenacious than the average bear?

00:18:36.280 | And, you know, what's kind of helped you to stick with it

00:18:41.280 | and to follow through on your goals?

00:18:45.800 | - Yeah, you're being too nice to me.

00:18:47.240 | So like, I would go away from Fast Day

00:18:49.600 | because I was like, maybe only Jeremy can do this stuff.

00:18:52.800 | I'm not like as good as you of a programmer.

00:18:55.760 | I'm not as good as you at reading papers.

00:18:58.080 | So I would like go away to other courses time to time.

00:19:00.960 | And again, like realize, no,

00:19:02.160 | I want to come back to the community

00:19:04.600 | and just the mad passion of really believing

00:19:09.000 | that if I stick to this would be really helpful.

00:19:12.660 | One thing I've learned is like,

00:19:14.960 | we always start aspiring to be some person as I was saying.

00:19:18.360 | And we're like always looking at that end goal,

00:19:21.960 | but instead I've over time learned

00:19:23.840 | to really enjoy the process more.

00:19:25.640 | So like when I was trying to lose weight,

00:19:27.440 | I was like, that was my fifth attempt at losing weight.

00:19:32.160 | This time what made it stick was

00:19:33.680 | I would just show up every day

00:19:35.520 | and walk for like two hours straight.

00:19:37.680 | Like no excuses, just walk for two hours

00:19:40.480 | for at least four months.

00:19:41.480 | And then I started losing weight.

00:19:42.640 | So broadly speaking, I just learned to appreciate

00:19:46.120 | the process of showing up every day and just doing it.

00:19:49.280 | - Yeah, okay.

00:19:50.120 | That's great advice, is to focus on the process.

00:19:54.760 | I had a similar experience, gosh, how long ago?

00:19:58.840 | Seven or eight years ago,

00:20:01.000 | when I decided to try to improve my fitness

00:20:04.640 | and I started going to CrossFit,

00:20:06.200 | like CrossFit was literally a block away.

00:20:09.420 | And I had really never done any kind of intense training

00:20:13.120 | in my life.

00:20:13.960 | And CrossFit, you dive very quickly into intense training.

00:20:18.620 | And each day I just felt awful.

00:20:24.300 | I can't begin to tell you how much I hated it

00:20:26.920 | and how bad I felt.

00:20:28.560 | And the other thing that makes it difficult

00:20:30.040 | was like everybody else at CrossFit

00:20:31.840 | is just so like gung ho about it.

00:20:34.680 | So I think I must've felt a bit like you was like,

00:20:37.000 | oh, Jeremy's better at coding than me.

00:20:38.760 | I can't do it.

00:20:39.600 | It's like, okay, these people are all CrossFit people.

00:20:43.640 | But yeah, you know, it helped to have Rachel actually

00:20:46.400 | 'cause we both decided to go every day.

00:20:48.680 | So like, okay, we're gonna go every day.

00:20:50.120 | I think it was like 5.30 AM, turn up, hate it, go home,

00:20:54.960 | turn up, hate it, go home.

00:20:56.140 | And after about similar time, I reckon about three months,

00:20:59.480 | I was like, I still hated it,

00:21:01.440 | but I didn't feel horribly sick every time.

00:21:06.440 | You know, I was just like uncomfortable

00:21:09.920 | and a little painful, you know?

00:21:12.060 | Yeah, it's hard, isn't it?

00:21:13.960 | Because like to get stuff out of these,

00:21:16.280 | any of these things, you have to turn up to the process

00:21:19.840 | for months and those months are often not much fun.

00:21:24.840 | - CrossFit is also super hard.

00:21:26.560 | You take, in one day you get super exhausted

00:21:28.920 | and next day you have to show up again,

00:21:30.360 | half exhausted, you take thinking results aren't coming in.

00:21:33.120 | What do I do?

00:21:35.400 | - Yeah, and I mean, so like when we're learning stuff,

00:21:39.140 | I think in university, it often seems to be

00:21:42.340 | the professors are proud of this, you know,

00:21:44.560 | like weed out classes.

00:21:45.840 | They're like, oh, 80% of my class flunked out.

00:21:48.160 | It's like, oh, you're a shit teacher, aren't you?

00:21:51.120 | You know, like you wanna be giving people positive,

00:21:56.120 | real positive feedback all the time of like,

00:21:59.160 | oh, wow, you just trained a model.

00:22:01.360 | Oh, wow, you just fine-tuned a model.

00:22:03.480 | Oh, wow, you just created a web application, right?

00:22:06.480 | That's how computer games work.

00:22:09.880 | You know, with computer games,

00:22:11.160 | you don't like start up Gran Turismo 7

00:22:13.720 | and it's like dumps you straight into an F1 car

00:22:16.480 | on Nürburgring and asks you to, you know,

00:22:20.100 | qualify for, you know, F1.

00:22:24.400 | It's like, okay, we'll start with license test

00:22:26.480 | and I'm not gonna afford focus.

00:22:28.340 | Can you drive in a straight line for 50 yards or something?

00:22:31.280 | And at the end it's like, congratulations,

00:22:33.040 | gold medal, you've done it.

00:22:35.080 | Okay, now we're gonna like show you how to do it

00:22:37.800 | and you have to press the brake pedal yourself.

00:22:39.680 | It's like, congratulations, you did it.

00:22:42.240 | - I think one of the things I learned through gaming,

00:22:47.080 | I used to game a lot.

00:22:47.920 | I would like sit down and game straight for 16 hours.

00:22:50.280 | My parents weren't happy about it,

00:22:51.940 | but like I would just fight and I would fight it,

00:22:55.660 | make that happen.

00:22:56.760 | In games, we call it grinding,

00:22:58.920 | where you like have to do this so stupid stuff

00:23:01.300 | over and over again, that just doesn't make sense.

00:23:03.860 | But you enjoyed so much and like just,

00:23:06.440 | I think many Kagglers see this,

00:23:07.920 | that like bringing that over to Kaggle,

00:23:10.200 | because initially you just suck on the leaderboard.

00:23:12.460 | I still suck on the leaderboard, I'm like bad competitor.

00:23:15.480 | But you enjoy the process so much

00:23:17.120 | because it's the grind that you start enjoying towards the.

00:23:20.520 | - Now you say that, but I don't think that's true

00:23:22.200 | 'cause you've had some very good results

00:23:23.620 | in competitions, haven't you?

00:23:25.840 | - I had the chance to team up with awesome people.

00:23:28.400 | So I was learning from them and contributing ideas.

00:23:31.820 | (laughing)

00:23:34.080 | - Now, the other thing I noticed with you

00:23:39.960 | is in terms of follow through,

00:23:43.440 | is you know, we often encourage people

00:23:45.160 | to kind of put themselves out there,

00:23:46.960 | build their personal portfolio, blog, whatever.

00:23:51.840 | Everybody says they're going to, not many people do.

00:23:57.980 | And it's hard, I know like you've talked about how,

00:24:01.520 | for example, this podcast, when you started,

00:24:05.120 | no one's listening, you know?

00:24:08.240 | It's just you talking out to the empty internet.

00:24:13.240 | Like give me a sense of like,

00:24:15.120 | what did that feel like to like, you know,

00:24:18.880 | write blog posts when you didn't have a readership,

00:24:21.000 | create podcasts when you didn't have listeners?

00:24:23.560 | Like, what did that feel like?

00:24:25.200 | And how did you push through?

00:24:28.420 | And then how did you get people to start noticing you?

00:24:32.060 | - I'll start with the tangent.

00:24:34.560 | I think great people improve and like they reach heights.

00:24:38.680 | Greater people keep, they uplift others

00:24:41.640 | and greatest people keep reaching higher heights in life

00:24:45.320 | and keep uplifting others.

00:24:46.520 | So people like you and the Kaggle community

00:24:48.880 | and Fast hair community are the greatest

00:24:50.440 | because you keep uplifting others also.

00:24:52.680 | My blog got a lot of traction in the podcast

00:24:54.680 | because you and the community were sharing it so much.

00:24:57.920 | But it was like super bad for me.

00:25:01.040 | As you might know, I'm like extremely sincere

00:25:05.660 | towards anything I try and like I give it my best.

00:25:08.220 | Otherwise I'd like just wouldn't sign up for anything.

00:25:10.720 | That's how it is now initially it wasn't the case.

00:25:13.440 | - Let's see, tell us more about that.

00:25:15.800 | Tell us more about that, what do you mean?

00:25:16.960 | So, I mean, that's just such an interesting point.

00:25:19.780 | So you're saying nowadays when you try, you really try,

00:25:23.960 | but that wasn't always the case.

00:25:26.800 | Tell me more.

00:25:28.320 | - I would just like bail halfway through

00:25:29.920 | like three days into it.

00:25:31.880 | One month into anything, I would just bail.

00:25:34.680 | When I was trying to lose weight in university,

00:25:36.940 | I would like work out.

00:25:38.960 | I worked out for one month super intensely.

00:25:41.760 | So I decided I would climb, we had super tall buildings.

00:25:45.040 | So like 15, 16 floors.

00:25:46.600 | And I used to live on the ninth floor of the hostel.

00:25:49.160 | Out of this like nowhere, I decided I'll climb 60 floors

00:25:52.720 | every day, at least 50 to 60 floors.

00:25:55.120 | I did that for one month, damaged my knees a bit.

00:25:58.680 | And then I just like gave up.

00:25:59.960 | I didn't like start slow and build up.

00:26:02.200 | - Right, so what happened, you just became more mature

00:26:06.000 | or was there some kind of like conscious decision or?

00:26:09.600 | - I think maturity, yes.

00:26:10.920 | I'm still not very mature, but now I like start slowly

00:26:14.200 | and like remain at anything I do.

00:26:16.720 | - Okay, so sorry, so go on.

00:26:18.240 | So you're very, so at the point you started the podcast,

00:26:21.080 | you were sincere about putting your effort in and yeah, go on.

00:26:25.120 | - Yeah, so when like I got to interview you

00:26:29.420 | or other agri-grandmasters to me,

00:26:31.560 | it felt like I'm being insulted towards people like you

00:26:35.240 | because it's not getting too large enough audience.

00:26:37.320 | And I felt like that's my job to like,

00:26:40.160 | if I'm asking for someone's time,

00:26:41.640 | I need to like put it out there.

00:26:43.320 | And that part was super annoying to me.

00:26:46.400 | Like how can I, this is something I'm putting out

00:26:49.000 | to the best of my ability.

00:26:50.160 | How can I like not get it to the audience

00:26:52.000 | that I think it deserves?

00:26:53.620 | - Right, so how did you?

00:26:58.200 | And yeah, I mean, like even just to put out

00:27:03.120 | that next podcast or that next blog post,

00:27:05.720 | yeah, how did you convince yourself

00:27:08.920 | to kind of do it anyway and then did that audience

00:27:13.000 | like just gradually trickle in more and more

00:27:16.100 | or was there something that was like,

00:27:18.600 | suddenly you got noticed or yeah, what happened?

00:27:21.080 | - I learned this from Radeko Smulski.

00:27:24.440 | He would post his goals and that's what I started copying.

00:27:27.600 | So from him, I started, as you mentioned,

00:27:30.520 | I now post my goals every single year also.

00:27:33.160 | I just, now I also make a video about it

00:27:35.600 | so that like everyone, I'm terrified then.

00:27:38.560 | Like I think everyone's watched that video.

00:27:40.240 | Now I need to like do that.

00:27:41.680 | And I switched from thinking about what's the outcome

00:27:44.980 | to like, I'm going to, when I was doing the podcast,

00:27:48.000 | I decided I'll put out two episodes every single week,

00:27:51.100 | no matter what for like an entire year.

00:27:53.200 | I just decided to do that.

00:27:55.220 | And I wasn't so much so looking at

00:27:57.620 | how many people are watching,

00:27:58.880 | although I would like pay some attention to it.

00:28:00.960 | And instead I would just like focus on making the best version

00:28:04.060 | of the podcast every single week.

00:28:05.640 | Or before that I was writing blog posts

00:28:08.840 | and I committed to writing one blog post

00:28:12.000 | at least every week.

00:28:13.080 | And I was just making sure I can write anything best

00:28:15.640 | to my ability at the time.

00:28:18.080 | So here is Radex's book where, yeah.

00:28:23.080 | Highly recommended for, he's had a lot of success

00:28:29.400 | and has a lot of great tips.

00:28:30.600 | I thought I'd better just share that

00:28:32.160 | since we've both been talking about Radex a little bit.

00:28:35.000 | Yeah, it's what you described, you know,

00:28:41.040 | which is like just trying to do a better job,

00:28:44.480 | you know, to get there.

00:28:45.780 | It requires a certain amount of kind of like trust

00:28:49.760 | in the system that there is some reasonable relationship

00:28:54.760 | between working hard and getting good results, you know,

00:28:59.760 | which I think is not obvious, you know,

00:29:06.560 | like it's not obvious that that's true.

00:29:09.340 | And so, and if you don't believe that's true,

00:29:14.120 | then it's gonna be very rough, you know,

00:29:18.280 | because it's just gonna feel like, well, why bother?

00:29:25.000 | I guess, you know, actually I wanna share something

00:29:29.680 | which is I, to give a sense that it is true.

00:29:34.160 | And so I mentioned earlier that I'm close to being

00:29:39.800 | a Kaggle Colonel's Grandmaster.

00:29:42.520 | Here is the last few Kaggle Kernels I've uploaded.

00:29:47.520 | And you can see every single one of them

00:29:53.920 | has got a gold medal, which is to say like I put my all

00:29:58.920 | into creating the very best Kaggle kernel I could.

00:30:08.400 | And yeah, it's like, and each time I have done that,

00:30:13.400 | people have reacted to that by upvoting

00:30:21.160 | and liking my Kaggle kernel, you know.

00:30:23.200 | And so like, I think like, if that doesn't happen,

00:30:27.040 | it's important to self-reflect, you know,

00:30:31.760 | same in a competition, a Kaggle competition.

00:30:33.600 | Like I like these transparent metrics

00:30:37.760 | 'cause you can't bullshit yourself about them.

00:30:40.640 | Like, I mean, you can, but you shouldn't.

00:30:42.160 | It's like, okay, if you don't get likes

00:30:43.920 | on your Kaggle kernel, you should assume

00:30:46.960 | 'cause it's 'cause you didn't do a good enough job.

00:30:49.120 | You know, you didn't explain it well enough

00:30:51.000 | or you didn't make it compelling enough

00:30:52.880 | or it wasn't a particularly good approach.

00:30:55.200 | You know, if you sucked in the private leaderboard,

00:30:58.680 | you shouldn't assume you just were unlucky,

00:31:01.080 | but that, you know, this is an opportunity

00:31:04.120 | for you to learn how to do better.

00:31:07.440 | - So sorry to disagree a bit, but initially,

00:31:09.720 | there's like always this period of suck

00:31:12.080 | where things wouldn't just pick up,

00:31:14.040 | at least on Kaggle also,

00:31:15.560 | when you're just putting out stuff or kernels,

00:31:17.720 | I think it takes a while for people to start voting it.

00:31:20.440 | So I would also like get--

00:31:21.760 | - Well, that's true, that you have to notice you.

00:31:23.800 | Not in competitions, you know, not in competitions.

00:31:27.520 | Yeah, I mean, certainly in podcasts and blogs and things,

00:31:31.320 | it's very different.

00:31:33.600 | It's like, trickles up very, very slowly.

00:31:37.600 | Like people often ask me how to get or Twitter followers.

00:31:42.000 | And I'm just like, just tweet.

00:31:44.440 | Like just, you know, and to try to make good ones.

00:31:47.480 | Like I started with zero followers and then one

00:31:50.400 | and then two, you know.

00:31:52.220 | It always goes very, very slowly.

00:31:55.440 | - I'll share this one thing from my,

00:31:59.760 | so this is titled as My First Kaggle Competition Experience.

00:32:04.240 | And on the leaderboard--

00:32:06.160 | - How much is your name?

00:32:07.520 | (laughing)

00:32:09.780 | - There's a story behind it.

00:32:12.360 | We were so naive.

00:32:13.240 | We were trying to rename stuff and my teammate ended up

00:32:16.600 | cleaning up my entire laptop without command.

00:32:18.960 | (laughing)

00:32:22.200 | This is how the first competition felt.

00:32:24.320 | Like I was looking at the Kaggle Grandmasters

00:32:26.520 | and it was like, they're on a supercar,

00:32:28.360 | I'm running barefoot and it's a 100 miles sprint.

00:32:30.720 | I can't keep up with them.

00:32:31.800 | I would submit to the leaderboard, sleep.

00:32:36.280 | And as I wake up, I would have fallen down 50 positions.

00:32:39.200 | I would come back, try it again, keep doing that every day.

00:32:43.040 | And like, I finished in the top 20% I think,

00:32:45.880 | which also was like super awesome to me.

00:32:47.720 | - Fantastic, and again, that's like tenacity, right?

00:32:51.040 | Like most people aren't gonna keep coming back.

00:32:54.560 | And in my experience, the people who do come back,

00:32:58.440 | make it, you know, like it's,

00:33:00.680 | I feel like this is the number one difference

00:33:03.600 | between those who succeed and those who don't.

00:33:06.080 | But yeah, coming back to your point about like for example,

00:33:10.760 | getting people to upvote your kernels

00:33:13.240 | or to watch your podcasts.

00:33:15.600 | I guess what we're saying is having extremely good content

00:33:20.240 | is a necessary but not sufficient condition for success.

00:33:24.920 | So you do need to make sure that you're telling people

00:33:29.920 | about your Kaggle kernels or about your blog posts

00:33:33.880 | or about your podcast episodes.

00:33:36.040 | But if you get people to look

00:33:39.000 | and they're not extremely good,

00:33:41.800 | they're not gonna come back and look again.

00:33:45.360 | - One thing, and to credit,

00:33:49.160 | the blog started through Rachel's advice.

00:33:51.360 | I would just read it every week at least once

00:33:54.240 | and I would sincerely follow her advice

00:33:56.200 | on how to get started on blogging.

00:33:58.120 | It's an amazing blog post that she's written.

00:34:00.840 | That's how it got started there.

00:34:02.240 | And I would just transfer my learnings to the podcast also.

00:34:05.560 | But the podcast was like super bad in quality

00:34:08.040 | during the initial weeks,

00:34:09.000 | but at the time that was the best I could do.

00:34:11.680 | And one thing I've learned is I try not to spam too much

00:34:14.720 | with like sharing it over and over again.

00:34:16.840 | So every time I've done a podcast,

00:34:18.360 | I've only shared it once with the world.

00:34:20.800 | And I, if I re-share it,

00:34:22.880 | it's because it's so relevant.

00:34:24.360 | I say, "Hey, please listen to this.

00:34:25.760 | "This is where I discussed this."

00:34:27.000 | But I'm also like extremely sincere

00:34:29.200 | about not spamming the community.

00:34:31.360 | Although like I do spam with my tea puns and tea jokes

00:34:34.560 | just for the sake of branding,

00:34:36.400 | but I try not to spam too much

00:34:38.520 | with like any stuff I'm building.

00:34:40.200 | - So this is Rachel's blog post,

00:34:43.480 | which I've heard what you just said.

00:34:45.600 | I've heard it from so many people

00:34:47.280 | that this is what got them blogging.

00:34:50.320 | And you know what a lot of folks might not realize

00:34:53.840 | is Rachel, you know, she's a math PhD.

00:34:58.840 | She's an academic.

00:35:00.400 | The idea of putting yourself out there like that

00:35:08.520 | is an absolute anathema to her academic training, you know?

00:35:13.320 | And, you know, in the academic community,

00:35:15.200 | which I'm also in nowadays.

00:35:17.480 | Yeah, it's almost look, it is.

00:35:21.360 | It's looked down upon, you know,

00:35:23.200 | of like communicating in a clear way

00:35:26.040 | that the rest of the world can understand,

00:35:27.680 | which is, it's such a shame, you know?

00:35:31.880 | And so it does require a pretty conscious decision,

00:35:34.480 | I think, particularly if you're at university

00:35:36.320 | to be prepared to be different.

00:35:39.280 | Because the people teaching you and supervising you

00:35:45.360 | have probably never written a blog post in their life.

00:35:48.080 | And they probably have no idea what it really is.

00:35:51.560 | And they're trying to get you to write

00:35:53.800 | the most obtuse academic jargony prose

00:35:58.480 | in extremely exclusive PDF-only academic papers.

00:36:03.480 | So tell me more about like

00:36:12.240 | creating educational content in general.

00:36:15.520 | What's the kind of educational content

00:36:18.480 | that you're finding is really gelling with people,

00:36:23.480 | that people are telling you,

00:36:26.280 | thank you so much for creating that,

00:36:27.840 | that made a difference to me.

00:36:30.160 | I'm still surprised when people say that to me.

00:36:33.320 | I'm like, I'm just creating this for my own self.

00:36:35.920 | I just follow Rachel's advice,

00:36:37.400 | which is to create something that I wish was there.

00:36:40.960 | And that's how I started the podcast also.

00:36:42.960 | I was talking to so many incredible people on Fast Day.

00:36:46.840 | To Atini, who used to take the course,

00:36:48.640 | he taught me how to get started in freelancing.

00:36:51.800 | And I would always shamelessly ask question.

00:36:54.600 | I would always start with an apology.

00:36:56.120 | Hey, sorry, this is going to be a super stupid question,

00:36:58.320 | but I want to ask this to you.

00:36:59.880 | And I would like just shamelessly approach people,

00:37:04.120 | ask these questions.

00:37:05.040 | That's how the podcast also got started.

00:37:06.760 | So I'm just trying to fill the gaps that I think

00:37:10.720 | exist.

00:37:11.560 | We have already.

00:37:12.400 | - It's something I learned.

00:37:13.320 | Something I learned in my 20s is, yeah,

00:37:15.480 | that which you've obviously already discovered,

00:37:17.480 | is that those embarrassingly stupid questions

00:37:21.880 | are the questions that lots and lots of people want to ask,

00:37:25.440 | but they assume everybody already knows.

00:37:27.280 | And so they end up never getting asked

00:37:29.360 | and everybody's like, we really don't know.

00:37:31.920 | So, speaking personally,

00:37:34.040 | when somebody asks me those questions,

00:37:36.600 | I'm always very grateful.

00:37:37.880 | Because I'm always like, ah, that's a good point.

00:37:41.160 | I had, you know, like,

00:37:42.160 | particularly if you're an expert on something,

00:37:43.800 | you don't really think to mention it

00:37:46.960 | until somebody asks you the question.

00:37:49.680 | - Yeah, I've been super blessed that people have been

00:37:53.640 | so nice and they always share their knowledge

00:37:55.680 | like they've shared it privately with me.

00:37:57.280 | And also like now on the podcast,

00:37:59.080 | they've been sharing it.

00:37:59.920 | So I'm super lucky in that way.

00:38:02.240 | - Have there been things that have surprised you either

00:38:04.480 | that have been very popular that you didn't expect

00:38:08.000 | or some things that you thought like,

00:38:09.600 | oh, this is gonna be big

00:38:11.120 | and they didn't really go anywhere?

00:38:13.480 | - Honestly, I started with the arrogance of like 20 year old,

00:38:19.560 | hey, people will listen to this.

00:38:21.080 | I know this is awesome.

00:38:22.040 | Everyone should listen to this.

00:38:23.480 | This will be the number one data science podcast

00:38:25.360 | in next year.

00:38:26.480 | No one listened to it.

00:38:27.440 | And then I realized as I got a bit more mature

00:38:30.640 | that it's super hard to get people to click on

00:38:33.160 | and our long video on YouTube alongside with everything else

00:38:36.480 | that exists on there,

00:38:38.120 | especially stuff that's like super technical,

00:38:40.920 | super knowledge heavy.

00:38:42.760 | Now I feel super grateful that people,

00:38:45.160 | thousands of people listen to the podcast.

00:38:47.280 | They click download, they click on YouTube.

00:38:49.320 | So now I feel like.

00:38:51.360 | - Yeah, and I think people are like,

00:38:54.120 | I think it's under appreciated how much people

00:38:57.360 | do want that kind of content.

00:38:59.000 | Like, you know, you look at something like

00:39:00.640 | Lex Friedman's podcast and it's not flashy,

00:39:05.640 | you know, like it's well produced

00:39:08.920 | in that Lex spends time on like,

00:39:11.400 | he has high quality equipment and he sets it up carefully

00:39:14.120 | and he does it in person.

00:39:15.280 | But you know, his actual interviews is just like

00:39:18.240 | him asking questions, generally fairly brief questions

00:39:22.520 | to somebody who's generally highly technical

00:39:25.320 | listening to the whole answer,

00:39:27.640 | rinse and repeat for one and a half hours.

00:39:30.560 | And it's, you know, lots of people do actually end up

00:39:34.120 | tuning in because they're, you know, not everybody,

00:39:36.200 | but there are people out there who want

00:39:38.240 | high quality, not dumbed down content.

00:39:42.960 | - I remember watching your interview on his podcast

00:39:48.160 | and I thought, no, you'll say yes to me.

00:39:51.720 | And when I said I want to fill the missing gaps,

00:39:54.800 | that's what I meant.

00:39:56.760 | Whenever I do a podcast, for your case,

00:39:58.600 | I listen to every single interview,

00:40:00.320 | every single email you've done.

00:40:02.400 | And then I would ask the questions that weren't discussed

00:40:04.680 | there, so just try to bring those things out

00:40:07.360 | and not always start with the repetitive questions

00:40:09.840 | that many people do.

00:40:11.360 | - That's a great point because Lex does that too, you know.

00:40:16.360 | And I don't do many interviews at all.

00:40:23.000 | Like I did one with Lucas, you know,

00:40:26.400 | who runs Wets and Biases, I did one with Lex.

00:40:29.000 | I mean, hardly anything else.

00:40:31.000 | Like, and in both cases, Lex and Lucas like,

00:40:35.640 | just had done their homework so well.

00:40:38.960 | And so the questions they asked were genuinely interesting.

00:40:42.360 | And so to the kinds of people who are interested

00:40:45.440 | in my thoughts in Korea, it's gonna be interesting to them

00:40:49.200 | because it's just, it's not just like, oh, who are you?

00:40:51.240 | What do you do?

00:40:52.080 | It's like, oh, do this thing called fast AI.

00:40:54.440 | And, you know, anybody who knows me

00:40:56.520 | doesn't wanna hear more about just like, what's fast AI?

00:40:58.800 | They know what's fast AI.

00:40:59.880 | So, you know, what you're describing,

00:41:03.560 | it's how everybody should do it, but hardly anybody does.

00:41:08.400 | (laughs)

00:41:10.720 | And, you know, probably a lot better for you as well, right?

00:41:12.840 | 'Cause in the process of doing that research,

00:41:14.480 | you're gonna learn about stuff.

00:41:17.240 | - I personally feel maybe it's because of my culture

00:41:21.080 | and upbringing, it's also insultive to the other person

00:41:23.880 | to ask like very basic questions.

00:41:26.040 | Like if I interview you and ask, hey, Jeremy,

00:41:29.320 | can you please introduce yourself?

00:41:31.640 | I won't do that.

00:41:32.680 | That's my job.

00:41:33.720 | I should be asking the interesting questions to you

00:41:35.960 | and to anyone that I--

00:41:37.360 | - I think it's insulting in our culture too,

00:41:39.520 | but people still do it.

00:41:40.960 | (laughs)

00:41:43.040 | Anyway, good on you.

00:41:44.960 | I mean, something I do wanna ask about,

00:41:47.640 | which I definitely remember is the excitement

00:41:53.600 | that you shared about having the opportunity

00:41:57.640 | to interview for a residency at Google.

00:42:02.640 | And I think, you know, the community was excited for you

00:42:07.040 | as well, 'cause it's like, wow, you know,

00:42:08.880 | good stuff, Sanyam, this is gonna be great.

00:42:11.520 | And then it didn't happen.

00:42:14.600 | You know, tell us a bit about that.

00:42:17.080 | Like, what did it feel like?

00:42:18.560 | Like, how did you get that offer, you know,

00:42:20.960 | to interview for Google?

00:42:23.960 | What did you do to prepare, you know,

00:42:28.240 | what did it feel like to fail at that, you know,

00:42:33.240 | at that journey?

00:42:35.480 | And how did you, you know, push on from there?

00:42:39.560 | And what did you do with that experience?

00:42:41.600 | - I was crying at 2 a.m. when I got the email

00:42:45.040 | that I had rejected.

00:42:46.880 | - Yeah.

00:42:48.040 | - And I used the word fail very intentionally, right?

00:42:51.160 | 'Cause that's what it is, and that's what it feels like.

00:42:54.560 | And it's like, okay, I've worked this, I failed, you know?

00:42:58.640 | And that's like, you know, let's not use bullshit words

00:43:03.640 | about whatever it's like when we fail.

00:43:07.800 | How do you, yeah, how does it feel?

00:43:09.600 | How do you get up?

00:43:10.440 | So you, I bet you were crying middle of the night.

00:43:14.000 | Probably felt like that's it, right?

00:43:17.440 | You tried so hard to get here, it didn't work.

00:43:19.240 | You're not good enough.

00:43:20.080 | Is that kind of the feeling you're having in your head?

00:43:22.640 | - Yeah, and the point I was getting to was also

00:43:27.080 | that I was again so distant to my peers.

00:43:29.960 | I remember telling my friends,

00:43:31.160 | hey, I interviewed at Google AI residency

00:43:33.680 | and they feel me.

00:43:35.080 | And they just say, cool, okay, you'll get it next time.

00:43:37.440 | Like they couldn't appreciate how big of a thing it was.

00:43:41.360 | - Yeah, that's hard, yeah.

00:43:42.560 | - And that's why I'm really grateful

00:43:45.080 | to FASTA community when I shared that, hey, I'd failed

00:43:47.640 | and I've done bad for the community

00:43:50.240 | because I couldn't get through.

00:43:51.280 | Everyone was like super supportive.

00:43:53.840 | That instantly got me to getting back.

00:43:56.000 | And in fact, that's how I started the podcast.

00:43:58.040 | I decided, okay, now I'm going to help the community

00:44:01.400 | to at least get to the point of interviewing

00:44:03.440 | and hopefully someone gets through.

00:44:05.440 | - Fantastic, so, you know, that's, I mean,

00:44:08.720 | I'm sure there's a lot of genuinely important outcomes,

00:44:11.520 | but how did you get there in the first place?

00:44:14.080 | You know, like you say, you come from a community

00:44:17.320 | where nobody even knows what you're doing,

00:44:19.320 | let alone gives a shit about it.

00:44:20.680 | How did you find yourself flying to America, to Google?

00:44:24.600 | - I just, I found this courage through applying

00:44:28.400 | through FASTA, I didn't expect to get into the course even.

00:44:31.160 | And since then, I've just found the courage constantly

00:44:34.520 | to seamlessly apply, seamlessly ask for help.

00:44:36.800 | So just stepping out of my comfort zone.

00:44:39.760 | I'm very inherently shy, I've been shy all my life,

00:44:43.080 | but now on the internet, I'm less shy

00:44:45.280 | and I just apply to every position.

00:44:47.160 | Even in my undergrad, I would start applying to positions

00:44:50.200 | that I knew were clearly out of my league.

00:44:52.640 | And my goal was to just get to the interviewer

00:44:55.240 | and ask them, hey, how do I like actually get

00:44:57.280 | through this interview?

00:44:58.120 | And I never got a clear response at that time,

00:45:00.680 | but I just got to the extent where I could comfortably apply

00:45:04.400 | to things I didn't expect to go through.

00:45:06.640 | - And every time you do that, you know,

00:45:08.640 | there's a chance you'll fail, right?

00:45:11.400 | And so, but if you don't apply, then you always fail.

00:45:15.720 | So, yeah, I mean, I'm the same.

00:45:18.360 | I am introverted and shy.

00:45:22.240 | And I remember telling my friend, Chris Latner,

00:45:25.440 | that a few months ago, and he was just like, sorry?

00:45:30.440 | I was like, I am.

00:45:33.000 | And he's like, that can't be true.

00:45:36.680 | He was like, here's all the things you do.

00:45:38.080 | I was like, I know, and they're all terrifying.

00:45:41.080 | And after it, I'm exhausted and I do it anyway.

00:45:45.160 | Yeah, you know, you have to put do it anyway, don't you?

00:45:51.720 | You know, 'cause otherwise you'll always fail

00:45:55.520 | because you never give yourself the chance to succeed.

00:45:58.080 | - Yeah, and just staying at it,

00:46:03.040 | I think I'd agreeing to the fact that initially I,

00:46:06.400 | of course, at that time didn't believe

00:46:07.880 | that I can't get a job just because I was

00:46:09.920 | so arrogant in my head.

00:46:11.640 | Here, I'm ahead of my classmates.

00:46:13.080 | Of course, I should get a job.

00:46:15.360 | India is a big country.

00:46:16.280 | It's very populated.

00:46:17.440 | There's a huge amount of competition here.

00:46:20.080 | So I, of course, wouldn't get the jobs.

00:46:21.920 | And I would continue in my arrogance somewhat

00:46:24.320 | in my tenacity to continue talking to people,

00:46:26.640 | continue applying left and right.

00:46:28.280 | - Yeah, so, okay, so you started the podcast

00:46:36.800 | and, you know, what happened?

00:46:41.800 | What was the next commercial opportunity you got

00:46:45.880 | and how did you make that happen?

00:46:50.480 | - I started it inspired by you

00:46:53.600 | where I decided not to monetize it,

00:46:55.760 | which wasn't the best decision

00:46:57.200 | because I was investing my money

00:46:58.760 | and not getting any returns.

00:46:59.840 | My parents were definitely upset.

00:47:01.560 | - Sorry.

00:47:03.680 | (laughing)

00:47:05.520 | - No, but I didn't mean it in that way.

00:47:08.040 | So initially, I started the podcast

00:47:11.720 | and then someone at H2O, H2O AI,

00:47:15.640 | it's an awesome company, shared my podcast internally.

00:47:18.320 | They were like, this guy's interviewing everyone.

00:47:20.960 | I want to interview people at our company,

00:47:22.760 | but he's interviewing them before I do.

00:47:25.040 | That made the CEO reach out to me.

00:47:28.560 | The CEO of H2O decided I want to hire this guy.

00:47:31.600 | And then I got it to do as part of my day job.

00:47:34.080 | - And he's another Indian guy, right?

00:47:35.560 | If I remember correctly.

00:47:37.160 | - Yes.

00:47:38.000 | - Yeah, cool.

00:47:40.440 | - Instantly, that's how the dots connected.

00:47:42.480 | And through any job I have had,

00:47:44.720 | this is I think the fourth stint in my career.

00:47:47.200 | It's always been through the FASTA community

00:47:49.400 | directly or indirectly, I feel.

00:47:51.160 | - Yeah, no, I mean, that's actually amazingly kind of direct

00:47:58.360 | is like your personal branding.

00:48:01.480 | I mean, not that unusual though, you know,

00:48:02.920 | your personal branding efforts was the thing

00:48:05.120 | that made people not only know who you are,

00:48:07.520 | but you know, that you're doing something that's so good,

00:48:11.120 | like that's literally what they were wanting

00:48:12.400 | to do themselves.

00:48:13.440 | That's amazing.

00:48:17.840 | Okay, I have reached the end of my questions.

00:48:22.240 | Did you have anything either that we haven't covered

00:48:25.280 | you'd like to cover, or is there anything

00:48:26.880 | that you would like to ask me before we wrap up

00:48:30.280 | the inaugural coffee time data science episode?

00:48:33.440 | - I would ask you for advice.

00:48:36.240 | What do you, as my guru and my teacher,

00:48:38.640 | what do you expect me to do next?

00:48:40.840 | What do you want me to do next

00:48:42.120 | that'll help make you happy?

00:48:44.560 | - Oh no, I never have opinions, you know?

00:48:47.920 | I try to like, I try to kind of say like,

00:48:52.360 | here's how you can move in a direction

00:48:55.280 | that you want to head, you know?

00:48:56.840 | Like, I don't have any sense of like what direction

00:48:59.280 | should anybody head.

00:49:01.080 | I mean, hopefully, in a direction

00:49:03.640 | that's at least not destructive to society,

00:49:06.000 | and that you've got the tools and thought processes

00:49:09.040 | in place to think about how to help society.

00:49:11.440 | But like, you're, I mean, you're doing exactly

00:49:14.840 | what I would have thought makes perfect sense,

00:49:16.760 | which is to surround yourself with not just good people,

00:49:20.120 | but people who appreciate you and value you

00:49:24.200 | for who you are and what you do.

00:49:27.880 | You know, and particularly,

00:49:29.320 | you know, in a geography where that isn't gonna happen

00:49:35.400 | just by virtue of the people who happen

00:49:36.960 | to live down the street, you know?

00:49:38.880 | You know, the only thing I kind of tell people

00:49:43.960 | of around your age is you could consider

00:49:48.480 | changing your geography at least for a year or two.

00:49:51.280 | Like, for me personally,

00:49:55.840 | one of the few regrets I have in my life

00:49:58.000 | is that I didn't do that.

00:49:59.760 | You know, growing up in Melbourne,

00:50:02.400 | very different to India, but in some ways,

00:50:06.840 | you know, it's got some similarities

00:50:09.280 | in that it's a long way away from,

00:50:11.720 | you know, at least kind of culturally

00:50:14.840 | and in Australia's case geographically

00:50:17.200 | from anything like the kind of stuff

00:50:19.880 | that I was interested in.

00:50:21.280 | And nobody in my life cared about anything

00:50:23.780 | that I was interested in.

00:50:24.880 | And that made me feel

00:50:28.760 | like there was something wrong with me.

00:50:32.440 | You know, like I was weird and that was a problem.

00:50:36.400 | And, you know, people would sometimes fairly directly

00:50:41.400 | give me that feedback, you know?

00:50:43.440 | - That sounds very similar to my experience also.

00:50:46.640 | - Yeah, so going to San Francisco to me,

00:50:49.420 | like there's a lot of things I don't love

00:50:52.560 | about the US in general in San Francisco

00:50:54.740 | in particular, but it was such a confidence building exercise

00:50:58.000 | to be surrounded by people who were interested

00:51:03.000 | in what I did and did try to do similar things themselves.

00:51:08.080 | And to also realize that the people that I admired

00:51:13.320 | and looked up to were not on some other level of existence.

00:51:18.320 | They went on some other plane, but they would,

00:51:20.960 | you know, they made mistakes and had setbacks

00:51:24.560 | and redoubled their efforts and just normal people, you know?

00:51:29.560 | And so I kind of thought like, okay, I'm, yeah,

00:51:34.880 | they're just other people like me

00:51:36.200 | and I can do interesting things like they're doing.

00:51:40.920 | And like I'd always wanted to do interesting things

00:51:42.920 | like they were doing, I guess.

00:51:45.360 | So that'd be the one thing like to consider.

00:51:49.040 | And I'd say like a couple of years

00:51:52.800 | in a invigorating culture full of people

00:51:57.800 | that respect the kind of work you do can be of value.

00:52:03.480 | It's not possible for everybody, you know?

00:52:07.600 | And honestly, it wasn't really possible for me

00:52:10.500 | for most of my life.

00:52:11.360 | So it's not really a regret in that it wasn't something

00:52:14.920 | that I had access to or something I wish I had access to.

00:52:18.500 | And it's definitely not needed.

00:52:21.600 | Like you can totally do it all, as you've shown online.

00:52:25.700 | But yeah, it's something to consider if it's an option.

00:52:30.460 | - Thank you.

00:52:32.460 | Now I've gotten to the stage where it's somewhat possible.

00:52:35.920 | I'll try to explore that option.

00:52:38.120 | But before we wrap up,

00:52:40.800 | I also want to really thank you for all the knowledge

00:52:44.800 | you've been sharing with the community.

00:52:46.120 | Your interview will always be pinned on my channel.

00:52:48.960 | This is the 150th episode on the podcast,

00:52:51.800 | but I'll always be grateful to you and the community.

00:52:55.720 | All of my small achievements are through Fast Day,

00:52:58.120 | so thank you so much for-

00:53:00.480 | - It's actually the first episode of Coffee Time Data Science,

00:53:03.260 | but I know what you mean.

00:53:05.260 | Well, we'll let you keep it as the 150th.

00:53:10.260 | You're more than welcome.

00:53:12.360 | And thank you, because I mean,

00:53:13.920 | you've given back so much to the community.

00:53:16.560 | So I know so many people are grateful to you.

00:53:21.480 | So thank you.

00:53:23.040 | - They've been too kind, just to add one more thing.

00:53:25.680 | I know I have a lot to learn still.

00:53:28.120 | I've gone off on this tangent of creating content,

00:53:31.120 | and I feel like an imposter

00:53:32.460 | that is only a discussion's grandmaster,

00:53:35.320 | is only a competition expert,

00:53:37.520 | and that knowledge needs to transfer somehow.

00:53:39.720 | So I know I have a lot to learn still,

00:53:42.280 | and the imposter syndrome is taking over, but I'll-

00:53:45.640 | - Wow, you're still young.

00:53:47.040 | You'll get there.

00:53:47.880 | Thank you, mate.

00:53:50.880 | Have a great day.

00:53:52.160 | - Thank you so much, David.

00:53:53.000 | - Bye.

00:53:53.840 | [BLANK_AUDIO]

Jeremy Howard Interviews Kaggle Grandmaster Sanyam Bhutani

Chapters