Bay Area AI: Lukas Biewald Interview

a hello everybody I'm Alexa crabber of a hello everybody I'm Alexa crabber of the organizer of bi ara I'm it up and the organizer of bi ara I'm it up and here we're on location at crowd flower a here we're on location at crowd flower a very well-known company which does human very well-known company which does human in the loop computing and we have to in the loop computing and we have to look as build founder and chief look as build founder and chief scientist with us here thanks Thank You scientist with us here thanks Thank You Lucas for was not ah so a lot of folks Lucas for was not ah so a lot of folks heard about crop flower and recently heard about crop flower and recently I've seen you guys are the lot of talks I've seen you guys are the lot of talks about the I so can you tell us briefly about the I so can you tell us briefly what is crowd for how to vault into the what is crowd for how to vault into the kind of center of the AI discussions you kind of center of the AI discussions you should have right now sure sure um you should have right now sure sure um you know so I i started crap flower because know so I i started crap flower because i was actually working in AI and it was i was actually working in AI and it was a tool that i really wanted and it a tool that i really wanted and it didn't exist right you know so what didn't exist right you know so what crowd flower does is help people get crowd flower does is help people get training data and then help people do training data and then help people do human in the loop mm-hmm and and you human in the loop mm-hmm and and you know I it was like seven eight years ago know I it was like seven eight years ago I guess now that that I started it and I guess now that that I started it and and basically ever I went I I wanted to and basically ever I went I I wanted to build real machine learning models it build real machine learning models it really worked in the real world and really worked in the real world and every what the bottleneck was just every what the bottleneck was just getting the training data right so getting the training data right so that's just like you know it's like the that's just like you know it's like the most simple thing it's a good thing that most simple thing it's a good thing that you don't think about the most but if you don't think about the most but if you actually want to build machine you actually want to build machine learning it's it's often your biggest learning it's it's often your biggest problem and so you know we started it problem and so you know we started it around getting people training data and around getting people training data and then we realized that you know that the then we realized that you know that the next biggest thing holding people back next biggest thing holding people back is a lot of times they're models only is a lot of times they're models only you know a percent accurate which often you know a percent accurate which often isn't good enough for a lot of isn't good enough for a lot of real-world processes mm-hmm but you know real-world processes mm-hmm but you know if you only use the cases where the if you only use the cases where the model is good if it almost by definition model is good if it almost by definition you have a useful model yeah and so we you have a useful model yeah and so we realize you can help a lot of our realize you can help a lot of our customers actually like deploy things customers actually like deploy things and really do things by adding a human and really do things by adding a human loop component to our software and so loop component to our software and so you know so that's that's how we ended you know so that's that's how we ended up where we are today so I remember up where we are today so I remember using crowthorne levels at cloud you using crowthorne levels at cloud you know 2012 it's also ready can difficult know 2012 it's also ready can difficult to place huh to place huh do you know like this kind of do you know like this kind of specialized you know annotation all specialized you know annotation all right and and then you know obviously right and and then you know obviously Mechanical Turk is kind of the the Mechanical Turk is kind of the the flagship of this kind of services so can flagship of this kind of services so can it just a little bit how I guys evolved it just a little bit how I guys evolved from you know that those years in today from you know that those years in today and how you're different from mechanical and how you're different from mechanical turk and kind of other sharing sources turk and kind of other sharing sources um well I'm surprised you say Mechanical um well I'm surprised you say Mechanical Turk is the flagship I would have to Turk is the flagship I would have to disagree with that ok historical it just disagree with that ok historical it just article historical first maybe like the article historical first maybe like the dinosaur ok you know so you know how do dinosaur ok you know so you know how do we have often from from 2011 you know I we have often from from 2011 you know I actually don't think I think if you you actually don't think I think if you you know if you looked at our product you know if you looked at our product you would you know you would see that a lot would you know you would see that a lot of the kind of like bones of it are very of the kind of like bones of it are very similar to what was happening in 2011 similar to what was happening in 2011 but I think what's what's really changed but I think what's what's really changed is the market has clarified itself right is the market has clarified itself right so you know if you're the machine so you know if you're the machine learning in 2011 you know there are learning in 2011 you know there are probably a lot less people doing it probably a lot less people doing it right and so you know at that time we right and so you know at that time we just didn't have the same customer base just didn't have the same customer base we didn't have you know the company just we didn't have you know the company just wasn't as big because you know the wasn't as big because you know the market hadn't hadn't taken off in the market hadn't hadn't taken off in the way that it has now right so you know I way that it has now right so you know I I don't think like you know there's I don't think like you know there's there's like huge differences between there's like huge differences between you know crowd flower now crowd far in you know crowd flower now crowd far in 2011 but but I think like all the little 2011 but but I think like all the little details are the difference between you details are the difference between you know kind of the the low end PS that we know kind of the the low end PS that we had now the really high NPS that that we had now the really high NPS that that we have do have today mhm because you know have do have today mhm because you know because of you know we now have the because of you know we now have the resources to get all the details right resources to get all the details right and actually make the product really and actually make the product really good for our customers mm-hmm so I look good for our customers mm-hmm so I look at you guys um you basically you become at you guys um you basically you become the platform all right so kind of the platform all right so kind of jealous a little bit you know in what jealous a little bit you know in what sense you are the platform for a hike sense you are the platform for a hike why wouldn't say with a platform for AAA why wouldn't say with a platform for AAA well I guess we're a platform for ya um well I guess we're a platform for ya um you know I think I like to think of you know I think I like to think of ourselves as like the essential missing ourselves as like the essential missing tools for AI I mean what's funny is like tools for AI I mean what's funny is like you know the AI is kind of the least you know the AI is kind of the least important part of our platform earlier important part of our platform earlier right like I think it's the human loop right like I think it's the human loop and and the the training data that that's really the the training data that that's really where you no crowd flower really like where you no crowd flower really like yes but not evil yea yea enable men they yes but not evil yea yea enable men they enable them it was a good way of saying enable them it was a good way of saying it and and you know how we became that it and and you know how we became that platform I guess you know we started off platform I guess you know we started off with like a very specific you know with like a very specific you know solution so it was it was only language solution so it was it was only language and it was is basically collecting and it was is basically collecting training data for language based training data for language based applications mm-hmm and so you know from applications mm-hmm and so you know from that very specific point solution with a that very specific point solution with a small number of customers you know we've small number of customers you know we've expanded its kind of all the different expanded its kind of all the different things that that AI does and I think things that that AI does and I think like really you know people say like like really you know people say like well you know you solved all these well you know you solved all these different industries like you know you different industries like you know you sell to like finance and retail and hand sell to like finance and retail and hand you know consumer internet and and you you know consumer internet and and you know think a lot of people say well you know think a lot of people say well you know how can you do that how can you know how can you do that how can you know a company that's only like 70 80 know a company that's only like 70 80 people like crop flower looking forward people like crop flower looking forward to do that and I think you know the to do that and I think you know the reason that it works is because we reason that it works is because we actually sell it to really one buyer and actually sell it to really one buyer and he's like all over the walls here right he's like all over the walls here right it's we call them dated an all right it's we call them dated an all right basically sell it to data scientists basically sell it to data scientists uh-huh and so you know I think what's uh-huh and so you know I think what's really interesting about data sciences really interesting about data sciences is like you know a data scientist at is like you know a data scientist at Bloomberg has a lot of the same needs as Bloomberg has a lot of the same needs as a data scientist at home depot mm-hmm a data scientist at home depot mm-hmm you know you might not guessed that but you know you might not guessed that but actually really their lives are you know actually really their lives are you know really similar me that both at the mercy really similar me that both at the mercy of data yes you know they want clean of data yes you know they want clean data right that's like you know data right that's like you know something that they both need you know something that they both need you know they're both trying to build accurate they're both trying to build accurate algorithms like when you know for both algorithms like when you know for both of them when they're their arms do of them when they're their arms do different things when the algorithms different things when the algorithms fail it might cause a problem right like fail it might cause a problem right like so this is function breaks okay so you so this is function breaks okay so you know they care a lot about the acuras know they care a lot about the acuras that goes against my disorder the skin that goes against my disorder the skin Linda for basically exactly sever and so Linda for basically exactly sever and so these guys which i think is really cool these guys which i think is really cool about being a data scientist today is is about being a data scientist today is is like you know I think these folks you we like you know I think these folks you we see them bounce around from into see them bounce around from into different industries right because like different industries right because like a lot of times our customer you know be a lot of times our customer you know be surprised like they'll go from you know surprised like they'll go from you know like working at you know working at like like working at you know working at like uber to working at you know Thomson uber to working at you know Thomson Reuters and yeah and it's like well Reuters and yeah and it's like well it'll take the tools with them they it'll take the tools with them they always have always have you know they need to clean up some new you know they need to clean up some new data that will come and they will turn data that will come and they will turn back to you an exact way to help on the back to you an exact way to help on the new kind of date exactly exactly yeah new kind of date exactly exactly yeah yeah yeah interesting interesting so yeah yeah interesting interesting so it's actually interesting that you it's actually interesting that you mentioned that you know you start with mentioned that you know you start with language because you know this meetup language because you know this meetup used to be a subtext and what I found used to be a subtext and what I found that essentially people often don't that essentially people often don't realize especially newcomers do machine realize especially newcomers do machine learning data size that you know the learning data size that you know the machinery algorithms are all the same machinery algorithms are all the same right and so basically look at this real right and so basically look at this real wealth of a mile expertise about a lot wealth of a mile expertise about a lot of people did not realize this so once of people did not realize this so once we started calling this data in the eye we started calling this data in the eye we suddenly doubled and tripled nice our we suddenly doubled and tripled nice our audiences am and it's kind of cross audiences am and it's kind of cross pollinating but the core is the same so pollinating but the core is the same so so I'm wondering right like what's your so I'm wondering right like what's your take on this evolution you know it used take on this evolution you know it used to be an LP and then that the science to be an LP and then that the science and later became cool and let's not cool and later became cool and let's not cool anymore and the eyes is cool right now anymore and the eyes is cool right now so kind of you know the design is like so kind of you know the design is like so 2015 so this is the year of AI nvidia so 2015 so this is the year of AI nvidia right and so so what what is your take right and so so what what is your take right like the methods don't change but right like the methods don't change but like the kind of people change the kind like the kind of people change the kind of company support what do you see in of company support what do you see in terms of my customers kind of turning 22 terms of my customers kind of turning 22 you where the the trance kind of what's you where the the trance kind of what's interesting in you new developments well interesting in you new developments well i mean i would say one okay so it one i mean i would say one okay so it one thing like for me you know about the thing like for me you know about the this sort of like hype around AI is like this sort of like hype around AI is like i have to say like i love it I mean i have to say like i love it I mean that's like the first thing that I that's like the first thing that I really loved and I was kind of really loved and I was kind of disappointed when I came out of school disappointed when I came out of school and a lot of stuff that you do like was and a lot of stuff that you do like was an AI so I mean I have to say like for an AI so I mean I have to say like for me like like AI becoming such a thing is me like like AI becoming such a thing is I think it's awesome and I just you know I think it's awesome and I just you know I had a friend who was told me they I had a friend who was told me they asked me their day a grad student she asked me their day a grad student she was saying you know it like what's was saying you know it like what's difference between you know like I difference between you know like I understand the difference in like you understand the difference in like you know you know the regression I do and know you know the regression I do and machine learning and I was kind of like machine learning and I was kind of like well you know there's not much well you know there's not much difference actually I think that's like difference actually I think that's like I kind of astute observation right then I kind of astute observation right then sort of the same thing but that doesn't sort of the same thing but that doesn't mean that it's not like powerful and mean that it's not like powerful and important and interesting so I guess important and interesting so I guess like I'm all for you know people being like I'm all for you know people being excited about AI and doing a and I excited about AI and doing a and I keeping I was a little skeptical about keeping I was a little skeptical about about deep learning and I've totally about deep learning and I've totally changed my mind after like trying some changed my mind after like trying some of it myself I mean I think like you of it myself I mean I think like you know neural networks are doing some know neural networks are doing some amazing stuff in vision mm-hmm at the amazing stuff in vision mm-hmm at the very least and it seems like cool stuff very least and it seems like cool stuff in text as well so you know so I think a in text as well so you know so I think a lot of this stuff that's happening right lot of this stuff that's happening right now is actually like just legitimately now is actually like just legitimately really exciting I mean I think that really exciting I mean I think that comes with a whole bunch of irritating comes with a whole bunch of irritating hype but you know we've seen other hype hype but you know we've seen other hype cycles and you were okay with that yeah cycles and you were okay with that yeah better than a pipe them to you know no better than a pipe them to you know no hype i'm at least it nice Peter takes hype i'm at least it nice Peter takes like yours ya know this is fill the bowl like yours ya know this is fill the bowl since it's good for the context right since it's good for the context right yeah and in the parent everybody is 8 so yeah and in the parent everybody is 8 so Sonya look like yeah my here the vesicle Sonya look like yeah my here the vesicle essentially every be company going to essentially every be company going to Tori I i realized the data problems the Tori I i realized the data problems the key problems haha I hear that they all key problems haha I hear that they all have teams working internal and all of have teams working internal and all of them are probably the your customers huh them are probably the your customers huh yeah right so but you know they they yeah right so but you know they they don't talk about kind of this date in don't talk about kind of this date in richmond in richmond is happening I richmond in richmond is happening I don't know it goes chembur so I under don't know it goes chembur so I under right what do you think of this i mean right what do you think of this i mean it's it's possible for business run it's it's possible for business run across basically let's say you know across basically let's say you know Cooper is kind of gonna do it on their Cooper is kind of gonna do it on their own maps who has the maps Yeah right own maps who has the maps Yeah right they called the key to location oh you they called the key to location oh you know Apple basically decoupled from know Apple basically decoupled from google maps and I will bear wants to do google maps and I will bear wants to do its own maps and son they're all its own maps and son they're all probably going to do the same thing with probably going to do the same thing with maps so yeah I'm curious how it reflects maps so yeah I'm curious how it reflects on on your business so do you for on on your business so do you for instance if you're not eight maps for instance if you're not eight maps for over a hot girl yeah welp experience over a hot girl yeah welp experience right so do you reuse this and in kind right so do you reuse this and in kind of and then kind of now you have of and then kind of now you have knowledge let's say how to quickly knowledge let's say how to quickly purify maps huh so what kanokon email purify maps huh so what kanokon email make it like a macular purification a make it like a macular purification a separate business ah how does working separate business ah how does working around dirty well you know so we can't around dirty well you know so we can't share data between our customers for share data between our customers for obvious reasons right um but we do with obvious reasons right um but we do with mood of a program where you can use our mood of a program where you can use our software for free if we can open up your software for free if we can open up your data and you know the reason that I did data and you know the reason that I did that was I was actually really concerned that was I was actually really concerned that you know with with with data not that you know with with with data not being open source and openly available being open source and openly available as data becomes kind of like you know as data becomes kind of like you know oxygen for companies likes just so oxygen for companies likes just so essential for companies it made me essential for companies it made me worried that worried that the best assets were all closed yeah the best assets were all closed yeah actually is a humongous issue yes I mean actually is a humongous issue yes I mean I think like open source has done so I think like open source has done so many good things for software I think many good things for software I think there's a kind of analogous like thing there's a kind of analogous like thing going on and data except that like it's going on and data except that like it's all closed and so here what happens is all closed and so here what happens is like all the academics they want to work like all the academics they want to work on really big really interesting data on really big really interesting data sets they end up going into industry you sets they end up going into industry you know for that reason right and then and know for that reason right and then and then you know just kind of perpetuates then you know just kind of perpetuates the cycle of more and more clothes data the cycle of more and more clothes data so you know I think that's actually a so you know I think that's actually a major issue and and you know we're major issue and and you know we're trying to help that by opening up the trying to help that by opening up the data sets that we can mhm um you know in data sets that we can mhm um you know in a way it's good for business that people a way it's good for business that people want to keep collecting the same data want to keep collecting the same data sets over and over but i think you know sets over and over but i think you know it's one of the things like in a kind of it's one of the things like in a kind of like a bigger longer view I think like a bigger longer view I think there's like plenty of money to be made there's like plenty of money to be made in data and you know the more successful in data and you know the more successful people are the more stuff that they'll people are the more stuff that they'll do yes and Eunice will come with a new do yes and Eunice will come with a new date alright so yeah so do you see any date alright so yeah so do you see any solution to this I mean do you think solution to this I mean do you think people will start sharing my data I will people will start sharing my data I will get it in the initial stage everybody's get it in the initial stage everybody's calling data and then people will calling data and then people will realize like this data can share or do realize like this data can share or do you think that will just continue more you think that will just continue more and more like this and you know big and more like this and you know big companies will amass preparator data companies will amass preparator data small come to be like looked out haha no small come to be like looked out haha no have access to this I don't know I think have access to this I don't know I think it's really tricky issue actually I mean it's really tricky issue actually I mean I think I hope that we find a way to I think I hope that we find a way to convince companies to open up date I convince companies to open up date I mean I think like you know our mean I think like you know our governments doing a fantastic job of governments doing a fantastic job of opening up data I mean I hope that you opening up data I mean I hope that you know people can find incentives you know know people can find incentives you know like like I was pretty excited about you like like I was pretty excited about you know even this is like a while ago but know even this is like a while ago but remember the netflix pride is right with remember the netflix pride is right with you yeah and that actually caused a lot you yeah and that actually caused a lot of data to get open yes but then you of data to get open yes but then you know is right to participate like I know is right to participate like I wrote a fortune for girls openmp nice I wrote a fortune for girls openmp nice I mean so I think it's really sad that mean so I think it's really sad that they didn't do it the next year because they didn't do it the next year because there are privacy issues you know a dude there are privacy issues you know a dude animas ation yeah so you know I kind of animas ation yeah so you know I kind of felt I feel like um I feel like that's a felt I feel like um I feel like that's a major issue that that that people really major issue that that that people really need to figure out because I think like need to figure out because I think like there's a major hidden costs to not there's a major hidden costs to not having open data available like that and having open data available like that and you know like everybody is you know like everybody is flix dataset even now your peer flix dataset even now your peer recommendation engine they always use recommendation engine they always use the netflix data center right and so you the netflix data center right and so you know probably like you know if if know probably like you know if if companies realized that they could get companies realized that they could get so much work done on their data sets you so much work done on their data sets you know maybe they'd open more data set so know maybe they'd open more data set so people would work on them and give them people would work on them and give them free insights kind of like open source free insights kind of like open source all right yeah I'm wondering you know if all right yeah I'm wondering you know if kind of data will follow right open sore kind of data will follow right open sore because open source did not materialise because open source did not materialise immediately right do you realize maybe immediately right do you realize maybe 40 years right after a straight kind of 40 years right after a straight kind of developed so some curious also bought developed so some curious also bought you know you know Madonna so obvious as you know you know Madonna so obvious as an example but you know I think you know an example but you know I think you know from personal experience that you guys from personal experience that you guys are cold when you know that it is are cold when you know that it is proprietary and you have this NDE haha proprietary and you have this NDE haha partners right where you can draw data partners right where you can draw data in companies like Bloomberg right in companies like Bloomberg right ashcraft information yea rather three so ashcraft information yea rather three so I'm curious right how you guys do this I'm curious right how you guys do this I'd like those has never been a leak at I'd like those has never been a leak at least you know I'm not seen a leak least you know I'm not seen a leak rights for a data analysis gold right so rights for a data analysis gold right so how do you kind of convince companies to how do you kind of convince companies to give you all this available data how do give you all this available data how do you sure that this providers will keep you sure that this providers will keep that a secret I how do you manage this that a secret I how do you manage this process and convince people like bloomer process and convince people like bloomer and of course higher standard of Kevin and of course higher standard of Kevin challenge to basically give you all the challenge to basically give you all the data to eternity um well I think there's data to eternity um well I think there's no a monastic like data security is an no a monastic like data security is an issue where there's no silver bullet issue where there's no silver bullet right i mean there's just a lot of lead right i mean there's just a lot of lead bullets as they say all right so i mean bullets as they say all right so i mean you know i think um you know i think you know i think um you know i think that the way you convince a company like that the way you convince a company like Bloomberg to give you data is actually Bloomberg to give you data is actually just like consistently not having leaks just like consistently not having leaks yeah you know why we don't have leaks I yeah you know why we don't have leaks I think it's because we have good think it's because we have good engineers building good software and engineers building good software and then we we work with really good then we we work with really good partners I mean you know I think like a partners I mean you know I think like a data leak would would put any of our NDA data leak would would put any of our NDA channels out of business if it's channels out of business if it's certainly like one data leak we're never certainly like one data leak we're never going to send our customers to to one of going to send our customers to to one of these partners again and so you know these partners again and so you know they're highly incentivized to to make they're highly incentivized to to make sure that stuff isn't leaking and we're sure that stuff isn't leaking and we're feel really proud that we haven't had feel really proud that we haven't had any leaks but I don't want to encourage any leaks but I don't want to encourage any hackers or anything to any hackers or anything to you have to come try that yeah but I'm you have to come try that yeah but I'm hears about the kind of the the culture hears about the kind of the the culture because it reflects on the culture because it reflects on the culture because obviously you're startups are because obviously you're startups are people know who build them that people know who build them that engineering culture is key to successful engineering culture is key to successful startup right and apparently guys startup right and apparently guys managed to build the data culture where managed to build the data culture where the humans involute not will do the the humans involute not will do the right thing so yeah the curious right do right thing so yeah the curious right do you have kind of training materials we you have kind of training materials we should give to the partners how do you should give to the partners how do you kind of certify a partner or do you kind kind of certify a partner or do you kind of have long-term relationship is of have long-term relationship is attacking informal kind of attacking informal kind of people-to-people oh how do you have any people-to-people oh how do you have any process do have any manuals that have process do have any manuals that have any kind of training you send you know any kind of training you send you know you're you know managers of this company you're you know managers of this company is that they serve the people how's work is that they serve the people how's work yeah I mean he gets all the above right yeah I mean he gets all the above right so you know we you know it's a pretty so you know we you know it's a pretty long process to get started as one of long process to get started as one of our partners mm-hmm I mean I think the our partners mm-hmm I mean I think the person that we have doing it is really person that we have doing it is really excellent I think you know think one excellent I think you know think one thing that's actually really benefited thing that's actually really benefited us is we tend to work with more partners us is we tend to work with more partners that also have a social mission mhm and that also have a social mission mhm and and I think that tends to get higher and I think that tends to get higher quality companies okay so you know for quality companies okay so you know for example one of our partners you know example one of our partners you know gives work to Muslim moment that in gives work to Muslim moment that in India have trouble like leaving they India have trouble like leaving they can't go very far from home right I can't go very far from home right I remember she spoke at the regime yeah remember she spoke at the regime yeah trata yeah yeah so yeah she's a longtime trata yeah yeah so yeah she's a longtime partner and now when you talk to these partner and now when you talk to these women that do the work they're there women that do the work they're there they love crowd flower there they feel they love crowd flower there they feel really connected to crowd fire they even really connected to crowd fire they even feel connected to our customers it's feel connected to our customers it's funny they know so much about American funny they know so much about American culture because a lot of move works as culture because a lot of move works as crowd flower tasks you know night and crowd flower tasks you know night and day for the last four years mhm right so day for the last four years mhm right so um near they're telling me like men um near they're telling me like men we've learned a lot about American we've learned a lot about American fashion you know we've learned about you fashion you know we've learned about you know I mean just like all kinds of like know I mean just like all kinds of like interesting things that they that they interesting things that they that they know and and and you know because you know and and and you know because you know they're the highest earners in know they're the highest earners in their family you know they're there I their family you know they're there I mean I guarantee it'll never leak mm-hmm mean I guarantee it'll never leak mm-hmm anything on purpose because you know anything on purpose because you know it's it's such a meaningful thing for it's it's such a meaningful thing for their lives and their families mhm no their lives and their families mhm no this is great yeah it's I think it's this is great yeah it's I think it's really awesome and I remember this from really awesome and I remember this from the summit right i think it's the summit right i think it's great context you guys out there so i great context you guys out there so i think i'll probably you know wrap up think i'll probably you know wrap up with this question this meetup is kind with this question this meetup is kind of the first in this kind of reeva of the first in this kind of reeva grated you know I I cycle we're doing grated you know I I cycle we're doing right we're going to actually wrap up right we're going to actually wrap up the the temple gonna do you know to the the temple gonna do you know to adopts a month and we also want to do adopts a month and we also want to do kind of systematic exploration of why I kind of systematic exploration of why I so on the one hand we want to have kind so on the one hand we want to have kind of you know horizontal exploration so of you know horizontal exploration so different verticals different industries different verticals different industries and you guys probably see a lot of them and you guys probably see a lot of them but on the other kind of want to have but on the other kind of want to have this kind of build up as they do in this kind of build up as they do in science but in the startup context right science but in the startup context right so we teach people i think is very so we teach people i think is very appropriate that we start with lowest appropriate that we start with lowest level service data is data right here in level service data is data right here in the bottom or right of the foundation so the bottom or right of the foundation so but the christian you know is but the christian you know is experienced practitioner and what would experienced practitioner and what would be your advice to a lot of people get be your advice to a lot of people get into the area i feel right so we want to into the area i feel right so we want to kind of you know down with hands-on kind of you know down with hands-on talks haha right from different talks haha right from different partitioner so kind of build kind of a partitioner so kind of build kind of a mini course right huh so if you were mini course right huh so if you were kind of learning a I you know kind of if kind of learning a I you know kind of if you tell your previous self lonely I you tell your previous self lonely I know in what you know right now and to know in what you know right now and to be the most efficient person in this be the most efficient person in this like least amount of time yeah right like least amount of time yeah right like how should you go about it like like how should you go about it like should you learn you know linear algebra should you learn you know linear algebra and calculus and all should you just and calculus and all should you just play with our aggression like what is play with our aggression like what is kind of the most efficient way to become kind of the most efficient way to become a data scientist and stuff playing with a data scientist and stuff playing with all the schedules mmm do a question um all the schedules mmm do a question um I'm actually teaching a lot of data I'm actually teaching a lot of data science classes these days it is really science classes these days it is really fun and I would say you know you don't fun and I would say you know you don't necessarily see the most efficient thing necessarily see the most efficient thing like I i mean i'd say like I I love math like I i mean i'd say like I I love math you know I'm uh you know I'm math guy I you know I'm uh you know I'm math guy I thought was a mathematician and and so thought was a mathematician and and so you know for me like i love all the math you know for me like i love all the math but i think if if you want to be fishing but i think if if you want to be fishing don't want any math you know any math don't want any math you know any math right i mean actually like I think um right i mean actually like I think um you know I think like you know you learn you know I think like you know you learn some Python and you learn I mean some Python and you learn I mean scikit-learn is such a fantastic piece scikit-learn is such a fantastic piece of software what I mean I just like very of software what I mean I just like very practically i would say learn Python hey practically i would say learn Python hey learn scikit-learn start building models learn scikit-learn start building models and then focus on the applications and then focus on the applications because like I think there's so many because like I think there's so many machine learning people that machine learning people that get so kind of like narrow mindedly get so kind of like narrow mindedly maniacally focused on the the algorithms maniacally focused on the the algorithms and making it like a percent or two and making it like a percent or two better mhm and I think like the real key better mhm and I think like the real key to making machine learning work is not to making machine learning work is not like you know can you get from eighty like you know can you get from eighty percent to eighty-one percent but it's percent to eighty-one percent but it's like you know how do you make eighty like you know how do you make eighty percent okay yes you know like those percent okay yes you know like those giggle data get yeah you get more data giggle data get yeah you get more data that's one way or more like or maybe that's one way or more like or maybe like make it in some way that like the like make it in some way that like the mistakes are like okay or like fun you mistakes are like okay or like fun you know like I mean you know I think know like I mean you know I think there's just like so much creativity there's just like so much creativity like I actually think like a lot of like I actually think like a lot of stuff is like you know like actually stuff is like you know like actually it's more human-computer action what you it's more human-computer action what you think about like you know I could make think about like you know I could make like a ten percent accurate self-driving like a ten percent accurate self-driving car they could be really useful if that car they could be really useful if that ten percent of the time it's accurate is ten percent of the time it's accurate is its parallel parking right right a car its parallel parking right right a car that pillow pressed for you that's great that pillow pressed for you that's great i love it right yeah if I make a 99 i love it right yeah if I make a 99 percent accurate self-driving car that percent accurate self-driving car that doesn't know when it's gonna crash yeah doesn't know when it's gonna crash yeah it's gonna kill me immediately like I it's gonna kill me immediately like I said it's like incredibly dangerous oh said it's like incredibly dangerous oh that's right you know I think like that's right you know I think like focusing on the accuracy I mean let the focusing on the accuracy I mean let the like what the nerds like you know nerd like what the nerds like you know nerd out on that and I think if you're if out on that and I think if you're if you're kind of just coming into the you're kind of just coming into the field I think you can bring your like field I think you can bring your like you know whatever experience like you you know whatever experience like you have to bear by building interesting have to bear by building interesting applications in a field that you know applications in a field that you know particularly well mm-hmm it's actual particularly well mm-hmm it's actual device right I mean when you mentioned device right I mean when you mentioned this i think you know stuff that people this i think you know stuff that people know that you know much much better and know that you know much much better and easier thing to do is just ask the user easier thing to do is just ask the user right right totally the loop right yeah right right totally the loop right yeah i said okay i'll try to squeeze a i said okay i'll try to squeeze a percentage from a matrix just get percentage from a matrix just get another oh there yeah just ask them another oh there yeah just ask them right yeah I'm some some some kind of right yeah I'm some some some kind of active learning and see what happens active learning and see what happens totally yeah cool thank you very much totally yeah cool thank you very much we're looking forward to talking on we're looking forward to talking on great great two story here awesome great great two story here awesome thanks