Transcript#

This transcript was generated automatically and may contain errors.

Hey there, welcome to the Paws at Data Science Hangout. I'm Libby Herron, and this is a recording of our weekly community call that happens every Thursday at 12pm US Eastern Time.

If you are not joining us live, you miss out on the amazing chat that's going on. So find the link in the description where you can add our call to your calendar and come hang out with the most supportive, friendly, and funny data community you'll ever experience. Can't wait to see you there.

Everybody, I'm so excited to finally introduce our featured leader today. His name is Matt Bannert. And Matt works in, wait, let me get this straight, AI & Analytics Solutions at Helsana Insurances. Matt, it's so good to have you here.

I would love to dive in and just ask you, what is Helsana Insurances? What does it do? Is it car insurance? Health insurance? Life insurance?

Well, yeah, thanks, first of all, for having me here. Heart rate is almost close to 100, which is very unusual for myself just sitting around. So I'm really excited to be here.

Helsana Insurances, as the name says, is an insurance company. It's a health insurance. In Switzerland, health insurance is maybe kind of a contrast to what it is like in the US. It's mandatory. No matter how you enter the country, whether it's through birth, fall out of an airplane, no matter what, after three months being in Switzerland, you need a health insurance. But we can't do with the prices what we want. There's rules and regulations. And there is add-on insurances. And that's kind of where there's the room to wiggle.

And yeah, we are doing data science as well, like got about 3,000 employees, which is the largest insurer in the country of maybe, well, our entire country is maybe the size of Queens and Brooklyn. And that's it, citizen-wise. And yeah, so we're the largest health insurer in Switzerland.

Amazing. And you work in AI and analytics solutions. Does the solutions part mean that you are enabling other data scientists to do their job? Can you tell us a little bit more about what you do?

Yeah, exactly. I mean, there's basically two things that we do. There is no centralized data science team for everybody. So there is data scientists with field-specific experience, like actuaries, which is obviously an entire certification. There's marketing people, all sorts of people in different departments are doing data science. And we are trying to enable them different ways.

First of all, we are kind of like the Ghostbusters that you can call when the model or anything gets out of hand. Then we are helping with the modeling, with the classical analytical stuff, with machine learning, also with AI. But then data scientists in their own fields are getting more capable and capable, also enabled by AI, enabled by stuff like Posit does, like making data science easier, better documentation enabled through the community.

So we do DevOps for data science to allow them, like they learn programming, not as software engineers, like in computer science, but to learn it maybe through a math class or stats class. And they're coming in from their fields, they're coming in, and we help them with the software engineering part, how to run their computations on our OpenShift cluster, or more recently on our GPUs and stuff like this. So we are enabling them with the DevOps for data science part, plus community, as you all know, that's why we're here, is a big part of it.

We're trying to connect them to data scientists outside, yeah, making them realize that there's a big community in Switzerland, that there's community all over the world that they can exchange with, and so on, and helping them do their stuff. And really, my motto is kind of to leverage their skill and their field-specific expertise.

And really, my motto is kind of to leverage their skill and their field-specific expertise.

Teaching and community work

I love that you mentioned community, because I feel like that's just like a signal, that signal for me, it gets me really excited. But I also know that you do things outside of work. You teach at a university, and you also organize community stuff for data science outside of work. Do you want to talk a little bit about those two things?

You mean about the fun stuff, of course, I mean, the fun stuff, I mean, fun is the main ingredient like to keep to stay motivated. And so yeah, there's basically three things. I'm a Euro, I don't know anything about baseball, still I wear this ace hat, because it's not to Billy Bean, right? So all the sports stuff, sports analytics, I love it. That's part of the fun stuff. I use it in my teaching. Because I feel like with students can relate to all different kinds of sports, mostly basketball, but every once in a while I do another sport. So this is teaching examples.

So yeah, sports, and then teaching is fun to me, I still teach at university, just because I want to stay in touch. I mean, I'll be turning 45 next year. And I kind of feel old, you know, and I want to stay connected and keep my ear on the track. Is it still relevant? What I do is if I build an R package or Python package, can anybody use it? This is really what how young people think. And yeah, how you reach people. So it's important to me to stay connected. And students, they speak their minds, right? So you want to be in touch with the with the younger generation as well. And that's why I enjoy teaching.

And I think you said you put it perfectly, Libby, when you said like, okay, what do you do outside of work? Of course, university, that's also like 20% of my job, but still like, right, fun part and what I do for, for fun. And then yeah, last thing you mentioned, I've been a chair for useR in 21, obviously, during the pandemic, we want to bring it, we wanted to bring it to Zurich, but we couldn't. So that was, I think, the last time I spoke in front of more than 100 people online. Yeah, that was useR. I was useR 2021.

And we were just streaming, like nonstop, like five days of R through all time zones. And we were really like, okay, now the people from Latin America, they are hosting the stream, they are in charge to do everything, like all the euros, they can listen if they want, but the others are in charge because they are awake. Then we shifted it over to Europe, Australia, and so on. And it was like globally all around the entire week of useR. That was, yeah, that was community work, it was the biggest one.

But yeah, now I'm back to smaller Zurich, our user group. I'm one of six hosts. And we are just hosting events, like local local government is super strong. They're super innovative, like we have cantonal governments kind of like states in the United States, though, even way smaller. And they're really doing nice things with stats and also contributing your own packages, pushing to CRAN, all of that stuff. And yeah, I think this this is mostly in person, when when we meet like local Zurich, our user group, or like cantonal meetings somewhere like once a year, but the Zurich our user group is maybe once a quarter.

Yeah. It's great to hear local meetups. I feel like Rachel and I are always interested in hearing about what works for people and in person, sort of making a comeback after COVID put a damper on that for quite a long time. I think we all got really used to being work from home. And for me, things really need to be virtual or else I usually can't attend to them.

But it's so great to hear that local government can be a catalyst for this because local government, municipal government here in the US as well, will often have a lot of data people. And they are often working in centralized locations in person, less so at home, although there is a mix of that here. And so I'm looking at my screen, I can see several people who are working in local government right now and who are really community oriented people. I want to encourage all of you to go forth and and set up some stuff in person and community stuff and invite some other data people.

Slido Q&A

Well, Matt, we have got to get to Slido questions. We have got at least five questions in Slido. As a reminder, everybody ask questions in Slido for Matt now that we've learned a little bit more about him. And let's go ahead and get started with a question from Natasha. There's an asterisk. So I will read it for Natasha. It says, what does a typical day or week look like for you? How do you split your time between programming, meetings, et cetera?

Oh, OK. Let me try and be honest. I mean, there's obviously always too much meetings. So I try. That's like the biggest fight, because one, first, I was working as an external person for Helsana. That's how I got in. And there it was a totally different story because somebody found a budget to hire me. And then I was protected from everything else. The moment I was on Helsana's payroll, all of the meetings, all of the stuff was coming in.

So unfortunately, like a big chunk, maybe one fifth of the day is meeting on average people like acquisition of new projects, like people trying to work with my team. And also what I see that software development, and sorry for extending this a little bit because I feel it's noteworthy through the fact that software development has officially gone mainstream now. Everybody can do this. Everybody does a prototype, something they come, they have an idea, and then they come back to me, to my team, and they realize, OK, this may be a way if we go through data science to speed things up and not go like the usual gyro, take a juggling and stuff. And then we get work that's essentially software engineering and not really analytics and data science, let alone like super AI stuff. It's like calling APIs or something. So this leads to more meetings than necessary.

Then maybe for me personally, a smaller chunk, maybe a half day a week or one day a week at most is analytics, really classical analytics, data visualization, very often putting things together, but more in a didactic way, like, OK, we don't do a lot of evaluation, not as much as we should, like looking at things, intertemporal comparison, stuff like this, like teaching people how to look at data in the big picture. It's kind of similar of what Di Cook does with when she advocates for visualization and looking at data and not just R squared and getting a context, putting context in there. There's a lot of like education, educating managers.

Then one day a week, it's like usual, the casual Friday, that's the teaching day where I prepare. I try to get put together an internal curriculum like, OK, this is like I call it the AI and analytics champions that people from outside the IT department where I work in learn all this stuff and get familiar with the stuff. So and the rest is DevOps, really, OK, putting together the platform team that runs our communities OpenShift with the AI layer, writing help charts on my own experiment with Docker locally, container based stuff, figuring architecture out. And once I figured it out, selling it to other people, yeah, prototyping, that kind of stuff. So mostly DevOps followed equally by meetings, teaching and a little bit of analytics.

OK, awesome. I feel like I'm going to have to come back and ask about OpenShift because I am not familiar and I'm wondering if anybody else is. Let me know in the chat if you use OpenShift or if you are familiar with it at all.

But I'd love to ask one more question from the Slido first, because I see that Noor's question has bubbled to the top here. Don't forget, you can read questions and vote for them. Noor, if you're available to ask your question live, I will help you unmute. There you go.

Hello. Hello. Let me dig up which question was it. Oh, yes, I see. So this is a question that's sort of like broadly related to data science slash AI since they kind of are the same slash marketing anyway and insurance. And one of the issues is that with AI data science is that it can often encode biases into models. So how exactly are you avoiding or circumventing that? Is that something you worry about? Is that something more on the data preparation end before you start throwing stuff at models or?

Yeah, well, in general, yes, but mostly we avoid it by not being that far. There are so many like low hanging fruits. We don't have to get to this where our organization like process wide, we are not there. We are not where like this this entire political decision. OK, what what is if machines make decisions? What if there is no one held accountable for it? We are not that far for us. It's OK, picking up the low hanging fruit, like saying we're not making the machine decide, OK, does this guy get an insurance despite having this track record of of health

issues and all of that? It's more OK. People are handing in bills on paper like the non-digital stuff we're using, let's say typical document intelligence, OCR kind of stuff, and then AI making like triaging mails to different teams, like making decisions on that.

So it's not a big risk of failure. So when I can't figure it out, like the wrong team gets the mail and then they do what they do, usually without the like bump stuff around.

And there is already a lot of ground to gain. There's so many things that that people that people do for. Yeah, well, you can still automate a lot of stuff. That's what I said in the beginning. People think we're doing all the fancy stuff and the frontier stuff that basically is calling in an AI, either it be on premise, our own self-hosted models, lots of classification stuff, but not on the hard questions.

But in parallel, we're working with our legal department. Hey, what's possible? How can we go about this? But in production, we are just doing the low hanging, low hanging fruits mostly. And there's a few exceptions to the rule. But but that's basically the honest answer. We're not that far.

Well, Nor had another question that I would love to just sort of like tack on to this to give us a better idea, some more context around the types of data that you work with and the types of problems that you're solving. Because as Nor says, DevOps sounds more on the data engineering side of things. She said, would you say you're more data science these days or data engineering? This is a great question, because I think a lot of us here are a little bit more on the data science side, maybe a little bit more in the analytics side. So give us a little bit of context around what DevOps is and what the sort of problems that your team solves are.

Well, that's that's difficult and it's a really good question. I think the answer is a little bit different for myself and people in the team. So. I'm kind of like if it was sports, I'm Decafon, right? I try to yeah, I do I do everything I do data engineering, I do analytics, I yeah, data engineering as well, like the SQL stuff, like put some technical stuff in here, like much of our own data is on an exosol, which is a SQL database where you put all the stuff together and then you've got views that are prepared for different purposes and then you run analytics on it. Analytics is a lot of our machine learning, a little more Python. But we also used MLR and other machine learning frameworks in R.

So I yeah, I look at it as Decafon, but I try not to be the guy who is a better sprinter than the sprinters and a better high jumper than the high jumpers. And they're just the people on the team. I try to, again, empower the experts that are better analysts than I am or better machine learners to do the stuff by taking away the DevOps stuff from them or enabling to do the DevOps stuff really easy and have proper pipelines that we can share and also taking away often it's fear like I think building R packages, the classic example, like people think it's something for gurus, and then they don't do it and then they work around find their own structure.

So I would say it's a lot of taking fear away, encouraging people to try. And then yeah, trying to be the enabler, trying to find the people I think in the organization is also important. When I was hired at Halsana, I said to my boss, look, what we need to do is we need our team to be the place everybody wants to be part of, like getting others and it really worked. People started, oh, okay, something is moving in here. And people started to do kind of an internship internally. And then they stayed with the team, the team became a little bit larger. This was the strategy.

So we do all sorts of stuff, we do data engineering, data science, and I don't think you can separate those. A senior data scientist should be capable of doing data engineering at least as much so they can talk to a data engineer doesn't mean they can't get help from the experts and so on. But and also, I think there was this book by Alex Gold, which kind of like pushed what I wrote in a book earlier, DevOps for Data Science. And this is the thing that say, hey, data scientists should know that and should be at least understand this, even though they might be experts in like, I don't know, random forest or whatever type of regression modeling, data visualization gurus. Yeah, they should be capable of both this. At least just speaking the language right and having some empathy with what the job entails.

Yeah. Yeah. And you mentioned DevOps for Data Science. That's by Alex Gold, who used to work at Posit. We love Alex. He's worked with us before. So I put the link to the book in the chat. It's fantastic. Everybody should go read it.

And I love my favorite thing about what you said is enabling the people who might be afraid to do it because it's so important for these subject matter experts who are smart, capable people. Maybe they're scientists, maybe they're actuaries when you're in insurance. or some other expert in a knowledge field, for them to be enabled to learn to program, do statistical programming, analytical programming. And you kind of have to feel welcome and feel comfortable and safe in order to push to learn that. So it sounds like you're creating that experience at HealthSANA and that's amazing.

I really enjoy doing this. And this is also I think it was it was funny. I think it was even Tarif, your CEO, I read a piece on LinkedIn where he said about his experience about treating the people not like you want to be treated like that. That's kind of the gold standard. And he raised it like to the platinum and trying to say, OK, I want to treat people how they want to be treated and not how I want to be treated. And I think that that was an interesting read. It was just two days before yesterday. I think I read it and I thought that was great. It's like the same spirit.

Building community through listening

I think the important thing to take away is that in order to know how other people want to be treated, you have to be in community with them. You have to get to know them. You have to listen to them. You have to know things about them, because if you assume that you are the blueprint for humanity and whatever you want is what everybody else would want, you're not going to get very far. So in order to find out what other people need and want and how they feel, you really have to talk to them. So I'm always pushing people, go talk to more people. It's the solution to everything.

So I'm always pushing people, go talk to more people. It's the solution to everything.

Yeah. Yeah. Yeah. I think it's important and I can say that like, oh, it's really personal. I mean, I'm a big time talker, like trash talking everybody all the time. And then not everybody is like that. And when you and I would also say I'm kind of quick to pick up things and then, OK, just because I got a sketched out solution in my mind, I'm not afraid to just drop it and say, look, this is what I think and this is how we can go about it. And then but this is crowding out some people that, oh, OK, he knows the solution. And that that comes over like super convincingly stuff that he says, and then maybe like he knows how to talk and then people are are convinced and you're not realizing that you might like crowd out somebody who could also like out of personal interest could do the work for you to say, oh, fine, this guy knows how to do it. I'm going to take a backseat.

Right. And this this is something we we as like saying it to everyone who also talks a lot, I need to learn. And in Tarif's spirit, say, hey, this is we need to find out about people, not only become good listeners, but but also have like the instinct and feeling people out. And I think this gotten harder because of remote stuff. But it's important and also how to motivate people.

Maybe that's also a thing if we make the connection to teaching again, especially nowadays, I feel like we are totally overtaught by all the information that's coming in and all the stuff that we feel we need to know. But we are undercoached. Like I want my teaching approach is really to to watch people like I was their tennis or basketball coach and then not say the same thing to everybody, but watch them, how they do their data science work. And then just like a tennis trainer talk, talk to me, hey, keep your distance to the ball. You can't swing when you are this close. And just like like this, watch people do their works.

But we are undercoached. Like I want my teaching approach is really to to watch people like I was their tennis or basketball coach and then not say the same thing to everybody, but watch them, how they do their data science work.

Also, what I do with my my student at ETH, like, hey, show me your PhD paper. How did you do your data analysis? What's your problem? And then just watch them while they're working or listening to conversations. We have like our own matrix chat at university. And then I just read what they write during the week to each other. So I think they will stop using Slack and WhatsApp so I can read what they saying to each other. They're aware that that I'm reading the message. But based on the conversation, say, hey, look, this is the discussion you most recently had. Wouldn't it be better if you ask these questions? And the point is like watching people work and how they do their thing and not like interrupt them before they step on a court.

I like that. I think you could also be a data community builder if you are a person who talks a lot or like Rachel and I who automatically moderate conversations because we're used to doing this live. You can just ask people to unmute and give you their opinion, too. Or you can say like, oh, you know, I really know that Jared knows a lot about this. He's not speaking up. But what if we called on Jared and we're like, hey, Jared, what do you think? And maybe that's all Jared needs to unmute and share what he was thinking. And then brilliance comes out of Jared. And you're like, oh, my gosh, I'm so glad that we asked his opinion. Right. Whereas he might have just kind of hung out in the background on his own and not send anything.

Yes, the chat is going wild. We are talking about all kinds of fun things. Jared says, don't call on me. I know I just picked your name out of the hat, Jared. Also, I hoot. But by the way, like Libby, can I get like the chat messages? Because I maybe I can just write some people back on LinkedIn. And I think it's about connecting. And maybe I don't know if I know all the stuff that people are asking for. But if I do, I'd love to connect and just answer a couple more questions, because I'm really overwhelmed by all of them.

Matt's journey to data science

We have so many. So Amin had asked on Slido with an asterisk, so I can ask it for Amin about how you ended up in your current position. Do you want to talk just a little bit about your journey to where you got maybe a little bit of education and then how you got to Halsana?

All right. Yeah, where should I start? And then we have lots of questions to get to, too. I didn't do a lot of career planning. I think that's I have to be honest about it. Like when I got interested in what's today data science was back when I was studying economics. So no computer scientists. I had we were talking about this. I think it was off the record before this started when we were doing the mic check. I had a web development company still in high school, which was really basic, like back in the end of the 90s, beginning of the 2000s and kind of did it during my studies. But I was studying economics. And to be flat out honest, I hated it.

I felt like economists were not at least the economists in my bubble. Not no offense to any economists. There's great people out there. But I didn't really connect. I didn't feel like they were the most funny persons and at the same time, not like the most clever, but they were trying to be the new philosophers that explain the world to everybody. That was how I felt about them at the time. Maybe I was a little dumb and naive, but I didn't manage to connect to them.

So I was hanging out with the sports people, but it didn't really help where I wanted to work. So statistics and econometrics was the way out of economics. And I didn't want to start over with my studies. And I started to really invest into statistics and econometrics. And then that was the way out. And then funny enough. that was the way out and then funny enough even though I didn't like economics I technically did a PhD in economics at ETH which is kind of MIT of Europe so at least as I want what they want to be so so and during that I never was thinking about like becoming an economics professor it was more about like enabling economists that are really good at economics to do their job and that's how I did all the stats modeling right and and so I got closer closer to this my PhD was never like the goal was never to become a professor but to get out of economics into what back then was big data first it was called statistics then it was big data then machine learning came now it's AI but yeah same

shit different name that was how yeah how that happened and then this is my 18th year at ETH now I was helping them I was working at a institute about 90 people Swiss Economic Institute that was doing a lot of data wrangling already at that time like exchanging data with the National Bank with politics with local government federal government but there was no real pipelines and I realized there was there was no economic academic career for me so I helped them automate all the status exchange do the data engineering stuff map out the pipelines kind of do the architecture and then after I had done this and automated this economic data exchange in the country now a small country I was looking for another challenge and then I started with with Syncron which somebody may know it's like a small Swiss company that also resells deposit product really a great place to work you may have met the people at USR or PositConf where they're typically are still headed and still still going some cool R packages like the founder also worked quite intensively on a tidyverse packages so really a cool company really knowledgeable colleagues people you want to work with so that was that was nice and then Halsana was my client and it was really I was hired to bring open source into the company and get rid of some SaaS environment so that was my my job at Halsana first and this is how I got to know the analytics teams and eventually switched over and now I'm co-leading the team and more like maybe as a technical thought leader yeah that's that's making so much sense now how you got into analytics solutions so you were working at a consulting place where you were helping other teams and other companies incorporate Posit products yes and so you were already learning that sort of DevOps and analytical solutions side of it before you decided to go in house at Halsana okay that makes a ton of sense I'm wondering now if we touched on how you all use tools at Halsana I don't think I asked I usually do by the halfway mark I'm falling down on the job can you tell us a little bit about the tools that people use that you are enabling do you have people working in mostly R or Python are they creating and deploying shiny apps you mentioned OpenShift maybe you can talk a little bit about how all those things work together that would be amazing that's an extended question I just needed a sip of water so well okay OpenShift let's let's get that out out of the ways it's it's a platform and so it's it's a Kubernetes cluster it's Red Hat Kubernetes base basically it's just a layer that Red Hat built on top of Kubernetes Kubernetes cluster just putting in a very abstract way a lot of nodes together so you can shift workloads people originally use it for high availability that means if you are running things on multiple parts as they say if one part goes down like there's another one with the same application that backs it up but for data science often being elastic like you have different workloads and two typical sets set up where you like run a website there maybe peak workload is double your workload of like a normal day but in data science peak workload can be like 20 times 25 times what it is on a flat day when people started like high-volume computations doing stuff with GPUs especially in seasonal business like the health insurance there can be increased workloads pretty quickly and then yeah on an office day it's different than the workload at night and so on and workload at night might even be higher when people start to run the intensive jobs overnight when there's capacity and so on and so forth so that's we are using Kubernetes for a little bit of a different thing the elasticity thing as opposed to only stability and high availability because the data science service doesn't need a 99% availability just like a website or something so that's very uncommon for the rest of the insurance with stability is king so this is kind of our thing that we need to get the processes going to the new thing that's new to the company that's also socially not always easy that's that's kind of this so about the products that we built we have like on on our OpenShift we host container based workbench we we got about like 40 people working with workbench really developing their stuff either using Positron VS code some people but also like the traditional RStudio we are in an R shop like in the in the beginning like this is what most people do so sequel is clearly number one a lot of sequel people but not on our team and then certainly more R and then Python but it's it's coming up there there's more Python is gaining ground there's also teams that solely work with Python but then we are using fast API with deploying fast API to connect for example but also to plain OpenShift when people don't need to connect layer so for example we use Posit Connect to enable the data science again it's about in data scientists if there is a field specific data scientists who doesn't know a lot of like the deployment and OpenShift like helm charts everything you don't need to do that with connect so this is where connect enables us to do this stuff but if there is a software engineer they don't even know what R is they don't even know what connect is they can deploy their own web server they don't they don't need that as two types of people but in both cases OpenShift is the platform for some people there's the Posit layer like Posit Workbench Posit Connect and also Posit Package Manager for but we only use that for the internal packages and then people deploy all sorts of things like there's this funny feature I don't know if it's still there where Posit Connect tells you what type of applications you all deployed and we are checking a lot of marks on that report so there is Quarto reports old R markdown reports there's plain R scripts is running just with a scheduled job cron job whatever you call that and shiny apps of course shiny for R shiny for Python we have both yeah and that's that's the apps we we deploy on this service amazing all right I feel like I just learned a whole lot about OpenShift and Red Hat and all the things that Mariana is very thankful very thankfully sending us links to in the chat thank you Mariana for your expert help in link getting today this is so great we have many many questions left in the Slido I would love to see if we can do a rapid-fire round of getting some of all right okay I'll answer more quickly and don't know 16 all right let's do it Tim you had a question I'm gonna ask you to me could you help asking that one live sure yeah I'm just looking at insurance what what types of use cases that were traditionally you would use a deterministic model to address have you found that the probabilistic models and the use of LLMs has really given you better performance hmm actually I'm only at Halsana for two years so I don't know that much about the past you you would have to ask like longer tenured people I feel it's a like this is a good question because people tend to throw an LLM and everything this is certainly something that happens especially now that they're even local models they don't even look at the other model they just bring the LLM in because it's the easier interface especially for the younger generation classifiers is a good example I mean I often say hey it could be it could be better like to use a traditional classifier whatever supervised supervised machine learning and they still come up with an with an LLM and also this LLM takes longer eats more resources but makes people wait longer so what we try to do is like this email classification was about like there's a general contact form hundreds of mails coming in a day and they're a text classifier that works on subject or on a text with traditional NLP would work kind of well it's super fast you can pre-train it like last year's data like you got you got all the forwarding like from a year full of data like 150,000 mails perfect learning ground you'd be faster on that but people oh okay wait this is integrated Microsoft integration oh there's copilot there's some sort of way throwing it to when there's power automate okay we got Microsoft in the house like then okay it's easier to direct it to an LLM and then people that are not even into statistical modeling make the decision to just throw it at copilot and I think I promise to answer quicker and system integration is like the one-word answer system integration is more key to the problem whatever integrates better is making this choice not like a technical or even statistical mathematical expertise it's most of the time system integration that drives it thanks awesome thank you Tim for your question all right let's see if we can lightning round another one we had an anonymous question that says everyone says artificial intelligence is a fast-moving area from your point of view how has AI changed over the last 12 months AI itself mmm for use cases not that much like token prices have increased that's one one change as a debate like people are trying to I'm trying to save on that that is one thing then on the business side I feel like this is the biggest change because these AI companies they raise a ton of money they really are running around with a lot of money for example like hugging face got acquired I thought it was 13.1 billion then I was talking about this in our team meeting and one of my teammates just said hey when you know duck TV like duck lips got acquired to by Amazon we don't even know about the price yeah yeah AWS right so they are quite so this is I think that the big business thing is it's the bigger change to me personally then like this compression of everything like they are picking the big companies are picking it up they're compressing it so much that's the biggest change for me like during the last year that a lot of like my favorite companies like that nobody knew got acquired I think I put a piece on on LinkedIn where I said hey look it's just like rappers we used to be like ghetto secrets but now they are fortune 500 listed or forearm tattoos like ultimate distinction for everybody now mainstream everybody does that and that's that that's that's the thing like the duck laps used to be a small company nobody knew and now they get acquired and they get really big time money that's that's the bigger change to me than the more capable models all right awesome well there is a follow-on to that I think because I mean had also asked in Slido with an asterisk given those current advances in AI how did your teaching change at all well very cool question I encourage people like the motivation fat factor has the coaching factor has become so much more important like I don't have to teach like details I don't talk a lot about like object orientation I tell people two things basically what I learned in eight years you can learn in one and a half or two years but only if you use AI I tell people don't use AI like somebody who forgot their homework like and realizes 10 minutes before the course and then okay I need to put it together something you certainly will be able like as good of a talker AI is to convince me oh this is good work because I can't I can't tell right and then people put stuff together but this makes him an expensive copy paste machine when you are staying out of the loop but when you use AI hey look this is what what the professor gave us is the code I you have an incredibly patient co-teacher now that you can ask hey what did you do by this line how could we challenge this why does he do that explainer that explains every line and you can choose on your own how deep do you want to dive into this okay this is front and maybe I'm not too much into front end but some other person says hey I want to understand every detail of the CSS and framework and how that happens like you choose where you want to dive in like super specifically follow your interests follow you but stay interested and stay motivated I kind of trying to put the energy that I have like even on a screen like tell people this is it so

but this makes him an expensive copy paste machine when you are staying out of the loop but when you use AI hey look this is what what the professor gave us is the code I you have an incredibly patient co-teacher now that you can ask hey what did you do by this line how could we challenge this why does he do that explainer that explains every line

my teaching more motivation then like really okay just syntax is the way you do it this is clean code whatever yeah all right well I saw that nor had asked a question that I think is pretty career-oriented so why don't we shift into asking you a career advice question Matt do you have a piece of career advice that you like to share with people or something you wish you had known before you got into a data career

yeah definitely so I have this pretty old and somewhat of a dated book I think I published this in 2024 and it was exactly about this it one it collects all the things that I wish I'd known when I started my PhD and this and I'm really proud of the fact that this book is still being bought it's still being read like two years later because it's so hard to to have something that that stays up to date and of course there's few things we not my need to syntax that that I wrote there a posted there because Claude can do it much better but the the advice is okay try to understand the big picture follow your your instinct what you are interested what you are interested in and because you need you need a driving force try to find out what what drives you and follow what you what generates an energy more energy than it drains and then and then very important get get rest I mean just because Claude has unlimited condition doesn't mean you have that type of endurance figure out what you want to focus connect with people don't be alone

this is the same answer I have a in this book I have a question sorry I have a chapter on the question R or Python and I said it's the choice that doesn't matter that's the name of the section I wrote and I said hey use whatever your peers are using eventually you're gonna learn both if you stick to data science and know that was a couple years ago already but the point is connect stay connected because it's always faster alone further together like build a group build your community build your network but also people that give you energy surround yourself with good people and yeah make these make choices yeah and driving state stay driven some somehow not mad but but driven and try to find the energy generating processes and then maybe not to make it only a blah blah answer not only a like a pep talk type of answer content wise I would say big picture stuff is important so when people people say to me I can leverage AI better than they do or I can use the fact that Claude can program stuff quickly for me better because I know okay this is OpenShift what's the role of OpenShift and I think I'm passing on advice that Emily Riederer who's also a keynote speaker at PositConf I learned I really recommend to listen to her talk and she was like helping me finding a red line for my book and she had this great idea we said okay put it around your toolbox around the stack stack is the developers toolkit so you want to have a programming language you want to have like an IDE you want to have version control you want to have data management and all of that you might not need all these bins for every box but you need to know the category of what the tool yet the category of the tool that's what you need to need to know and that's important like stay ahead of the big picture stay at home of the big picture and try to find that out for your field for your yeah expertise and that's I think the way how you can stay in the mix with all the AI stuff everything that's my career advice thank you Matt that was beautiful

energy surround yourself with good people and yeah make these make choices yeah and driving state stay driven some somehow not mad but but driven and try to find the energy generating processes

oh we had people asking you in the chat what is the book and I found research software engineering yes yes that's it this is miss with CRC like there is a print and but there's also free version like CRC was enabling me to like do a creative common license so you can take a look don't have to buy my books read what you can use and especially like the motivating chapters I think this what we discussed here great yeah wonderful I put that in the chat for everybody and I will save the chat for you Matt if anybody else would like to save the chat there are three dots at the top of the chat and that will let you save things and have it for later and I will also try to put a chat file up in the discord server so please please hop on the discord hang out with us there you can make your way to the data science hangout channel if you join with the link that is in the chat that will also give you a data science hangout and data science lab crew member role so your name will be yellow so definitely use that link and if you are a data science hangout person and you don't have a yellow name or role ping me in the day science hangout channel and I will give it to you or Rachel will if somebody will Matt this was so much fun today thank you so much for hanging out with us giving us all of your wisdom hyping us up about posit comp also because I'm really excited to hear Emily's talk if anybody knows Emily reader she has joined us it was a great experience thanks so much for engaging questions was really cool I'm gonna put your LinkedIn in the chat if that's okay of course of course happy to connect with anyone and every once in a while I'm even in the US happy to so probably not this year anymore but happy to meet for every single one of you at some conference or whenever and say hi when you in Zurich right yeah all right Matt thank you everyone for hanging out reminder again we will not see you next week we will be at posit comp Rachel will be there in person Isabella and I will be there online and we hope you have an absolutely fun and wonderful week I will see you on the discord very shortly and for all of posit comp bye everybody