Resources
Apr 3, 2025 |
57 min
|
524 views
|

Should you build or buy AI? | Jay Timmerman | Data Science Hangout

To join future data science hangouts, add it to your calendar here: https://pos.it/dsh - All are welcome! We'd love to see you!

We were recently joined by Jay Timmerman, Head of Data Science & AI Platforms at Biogen, to chat about practical applications of generative AI in the pharmaceutical industry, navigating the challenges of AI like hallucinations and data privacy, the evolving role of data science teams in the age of AI, and the strategic considerations for evaluating AI vendors and building in-house solutions using platforms like Posit Connect.

In this Hangout, we explore the strategic considerations for evaluating AI vendors and the concept of "AI grift". Jay emphasized the importance of having in-house expertise to discern genuine value from vendor offerings and suggested that organizations can often build equivalent capabilities themselves by properly framing generative AI problems. This discussion highlighted a pragmatic approach to adopting AI and the need to be discerning when considering vendor solutions versus developing internal tools.

Resources mentioned in the video and zoom chat:
🔗 Abigail Haddad's Blog on Unstructured Text Data with LLMs → https://presentofcoding.substack.com/
🔗 Hard Fork Podcast episode on vibe coding → https://podcasts.apple.com/us/podcast/is-google-search-cooked-were-getting-a-u-s-crypto/id1528594034?i=1000698254448
🔗 Posit Blog on Deploying a Streamlit Application with Posit Connect → https://posit.co/blog/deploying-a-streamlit-application-with-posit-connect/
🔗 Posit Blog on Deploying a Dash Application to Posit Connect → https://posit.co/blog/deploying-a-dash-application-to-posit-connect/
🔗 ragnar package for R users (mentioned in chat for RAG) → https://github.com/t-kalinowski/ragnar

If you didn’t join live, one great discussion you missed from the zoom chat was about "vibe coding", with some attendees sharing humorous takes and a mention of a Hard Fork podcast episode that featured examples of such projects. Let us know below if you’d like to hear more about this topic!

â–º Subscribe to Our Channel Here: https://bit.ly/2TzgcOu
Follow Us Here:
Website: https://www.posit.co
Hangout: https://pos.it/dsh
LinkedIn: https://www.linkedin.com/company/posit-software
Bluesky: https://bsky.app/profile/posit.co
Thanks for hanging out with us! 💛

image: thumbnail.jpg

Transcript#

This transcript was generated automatically and may contain errors.

Welcome back to the Data Science Hangout, everybody. My name is Libby. I'm the Data Community Manager here at Posit, and for those of you who are new around here, Posit is the company behind awesome data science tools for R and Python. You might know us as the folks who used to be called RStudio, and we make things like the RStudio and Positron IDEs, Shiny, and other great tools.

The Hangout is our weekly get-together, every Thursday, same time, same place, where we chat about all things data science across all of our different industries. We talk about how we do things. We get advice. We connect with each other in the chat, which is always a party, and it's a really welcoming space. We love hearing from everybody. It doesn't matter what your background is, your years of experience, your tool preferences, and if you are watching or listening to this as a recording in the future, hello, you're really missing out on the chat. We'd love to have you live, so look for a link in the description to add our call to your calendar.

We definitely want to hear from you, so please jump in the chat, introduce yourself. A lot of you are already doing that. Say hello, mention your role, where you're based, your LinkedIn profile link, or a link to your website, somewhere where people can find you, and something that you do for fun as well. We always want to hear that. You can also share resources, job openings, as well as any roles you happen to be looking for right now, so we also really want to focus on this being a supportive and a safe space. We have built this together over the years, and we're so, so thankful for everyone who has made this space what it is today, so we're dedicated to keeping that way.

Send us some feedback in the survey that's going to pop up in Zoom in your browser when you close this. Let us know how we did, if you had a great time, if you have any additions that you'd like to see in the future to make the Hangout just 10% better for you. This is going to be a conversation with a data science leader that is powered by your questions, and we are using Slido again today. This is going to be another Slido day. It's a great way to see all the questions and upvote the ones you're most interested in, and so what I'm going to do is I'm going to re-share something that I'm testing today because Slido can actually be used inside of Zoom as well as in the browser.

So we're going to share, Isabella just shared the link to the Slido in the browser, and then if you see it pop up with a Zoom invitation inside of the chat, you can click that and it will open in Zoom. I would love to hear in that sort of feedback survey that we send on Zoom how this worked for you today. Like, did you like it? Did it work inside of Zoom? Is the browser better? Let me know.

I just wanted to let everyone know because I just opened it too. If it takes over your chat there in the Zoom bar below, you can toggle between the chat and then where it says apps, and that will bring you back to the Slido too. Nice. You can also pop it out. So if you happen to have other screens, like I have extra screens, you can click the little like pop out button and pop your chat out and also pop out the little window that holds the Slido Q&A. Okay, so I will try to remember to share that. I'll do it again right now, just kind of throughout the call, and I think that Rachel will be sharing the browser-based link as well.

So in Slido, you can put your name on your question. Please, please do that. If you can, add your name to your question in Slido so that I can actually call on you, get you unmuted, and you can ask your question. If you don't want to ask it live, you can put an asterisk in it and I'll ask it for you. No worries. If you really, really, really want to stay anonymous, feel free to leave your question anonymous, but I promise you you will not find a friendlier or more supportive crowd on the internet anywhere. So please, please don't be shy.

All right, I am excited to be joined by our featured leader today, Jay Timmerman, Head of Data Science and AI Platforms at Biogen. Welcome, welcome Jay. Great to be here. I'm happy to talk to you guys about my experience at Biogen and doing data science. And just so we kind of are clear, I'm representing myself. I'm not representing Biogen here, so I'm happy to provide my personal experience with using Gen AI and how it can be used.

Wonderful, wonderful. Well, we would love to get started by asking you a little bit about Biogen itself. What is Biogen? What does Biogen do? And then a little bit about yourself. So what is your role? What kind of problems does your team solve? Maybe what types of data do you use? And we'd also like to know what you do for fun. So I will try to remind you on some of these questions as we go.

About Biogen and Jay's role

Sure. So Biogen, for folks that aren't familiar with it, is a large pharmaceutical organization headquartered in Cambridge, but I'm actually based in North Carolina in RTP. Biogen's kind of claim to fame is they created some revolutionary products in multiple sclerosis. And so those products came out in the early 2000s and the more recent years been pivoting more into rare orphan diseases.

So one of our major products is called Spinraza. It's for kids that have spinal muscular atrophy. It's a devastating neuromuscular disease. And their product has really just changed the standard of care for folks like that. There's another product we have released recently for Friedreich's ataxia, another sort of devastating disease that is actually changing the standard of care. So it's really cool to work at a company that is actually, you know, making the difference in some of these, you know, obviously more common diseases like MS. And then in these rare diseases, being part of a company that can change that is big. And then more recently, we launched a product for Alzheimer's. So it's cool to work in a company where they work on really hard things and, you know, there's evidence that we're able to bring them to market. So very proud to be part of a team like that.

So what kind of data do we use? So let me explain a little bit about my role. So I sit in one of our, I work in IT, and I sit in one of our enterprise services organizations. And so I get the opportunity to work with everybody, right? I'm not aligned to just research. I'm not just development. It's really kind of interesting to be able to support the entire organization horizontally. It's kind of a job I've always wanted.

So the types of data we get to work with are all over the place. It could be a lot of corporate function data, like, you know, looking at expense data. We can work in our pharmaceutical operations group. We might look at vendor invoice data and customs declarations. We might look at in a commercial organization, some real world evidence. So it really is massive in terms of the scope of data that we have access to. So it's fun to be able to try to do data science and generative AI in this space.

So in terms of fun though, I think I'm getting all your questions here. Yes, absolutely. I was going to say fun's next. Fun. So I have two kids. I have a, my son is five and my daughter's three. So most of my fun activities are spending time with them. You know, my son just started t-ball, so it's kind of exciting to be able to, you know, get him to learn how to hit the ball and play catch in the yard. So I'm really enjoying that.

Awesome. Yeah, the weather is turning amazing here in Texas. I'm not sure what it's like for everybody else. It might still be cold, but it's, it's a sweet spot right now. We get a good March and a good April usually.

Vibe coding and code oversight

Well, I wanted to sort of kick us off with a question from Slido, but Abigail, I don't know what vibe coding is and it looks like Noor asked the same thing. Abigail had asked, vibe coding, yes or no? Also, are you creating any new code oversight processes now that LLMs can code a lot? Abigail has an asterisk, so she might not be able to ask things, but maybe we can get some clarification in the chat. Jay, does vibe coding mean anything to you? It does. Okay, so can you help us understand?

Yeah, so vibe coding is like, so these LLM enabled code generation tools are becoming more powerful. There's things like Cursor, which is like an IDE that is used to help generate code. And there's also other ones, I think Windsurf is another one. And so the idea is these things are becoming increasingly powerful. Why don't I just like give this tool an instruction and just let it rip and see what happens? So I think people are doing this. For us, we don't do that. We're not in that kind of business. I think if you're trying to experiment with yourself, see what you can create, I think that that's kind of a fun experiment.

However, I would caution, if you don't know what you're doing, you have to expect that the quality of what you're creating is probably not great. And also the security components of this. I've seen examples where, you know, embedding keys and code, kind of a big no-no, the vibe coding could do that. So I think for exploration, I think it's a reasonable thing to do if you want to see what these tools are capable of. But putting it in front of external facing people, I would be extremely careful. I wouldn't do it. But, you know, I think people are attempting to do that. So maybe you could be the first. I don't think that I'm brave enough for that.

The other one was code oversight. I'm not really sure what we mean about that. Is it just like, you know, using AI or like, you know, doing code reviews? Or maybe, can you elaborate a little bit about what the question is? Yeah. Oh, sorry. Like, is there, now that it's possible, sorry, you could have people writing a lot of code without really knowing what it does. Are there any kind of steps you've added into your existing code review processes? Or do you feel like you're already kind of managing all the things that needed to be managed?

So this is an interesting question. The way that my team operates is we are very focused on generating repeatable components. So I'm less interested in someone creating some crazy new capability and like it has some crazy functionality. And I'm more interested in people building a component that other people in the team can reuse. So for that to work, everyone needs to be able to understand how the code operates. So I think that in my team, it's a little bit more resistant to that because we're doing kind of very discrete tasks where we're trying to build components like modular components.

But I think in organizations where you're expected to do more, I think that becomes a bigger challenge, right? Like I have a large team. If I was a team of one and I had to do the full stack, I think that the risks of that grow significantly. So I would just kind of work in your domain and I think you'll be all right. Try not to step too far away without having someone that knows what good looks like and you'll probably be fine.

Thank you. And I wanted to point out Rebecca Butler had mentioned in the chat that there was a recent Hard Fork episode that had some fun examples of vibe code projects. She says emphasis on the fun. So maybe check out the Hard Fork podcast. Maybe somebody can find a link for that and drop it in the chat so that we can all see it and share it and maybe I can grab it for the resources.

The rise of head of AI roles

All righty. Rachel, you are asking the same question that I was going to ask. You want to hop in and ask that? Sure. Yeah, I was just wondering if head of AI is a role that's popping up in a lot of companies already and just wondering, Jay, if you could tell us a little bit about becoming the head of AI and if that role doesn't exist in our own companies, how might we advocate for that?

So I think the short answer is yes. I think head of AI is increasingly a legitimate role that is not just a manager AD role. I'm seeing it more at the VP, at the executive level. And I think the reason for that is there's a recognition that this capability, if the capability got no better, like if suddenly chat GPT or whatever just stopped, there's still a gigantic opportunity space of how to learn and apply the technology, but it's going to continue getting better. And I think that they want to have, organizations want to have a leader that's sort of plugged in and capable of delivering those solutions.

The other reason to have this, which maybe is a little bit more subtle. One of the things that I see a lot of is software as a service vendors sort of popping up, providing kind of like niche solutions. And I get pulled into unbelievable amounts of vendors meetings to basically kick the tires on these solutions. And what I've seen is basically people want to wrap, they basically build a tool that's a wrapper on open AI and they call it a product and they want to charge you money for it. And so having this sort of head of AI, people that are responsible for being that expert kind of prevents this like, I call it like AI grift, but like you basically prevent these, these grift organizations from getting in and start extracting money. When in reality, what they've built is unsophisticated. They don't have any intellectual property. They don't have some secret data. They basically just glued a bunch of off the shelf stuff together to both, you know, to put it together.

And so having this sort of head of AI, people that are responsible for being that expert kind of prevents this like, I call it like AI grift, but like you basically prevent these, these grift organizations from getting in and start extracting money. When in reality, what they've built is unsophisticated. They don't have any intellectual property. They don't have some secret data. They basically just glued a bunch of off the shelf stuff together to both, you know, to put it together.

The other thing too, is I think that the AI lead part is really, there's really two components to it. One is a platform component. When you, when you build AI stuff, I call them workflows. I think workflows is the right way to think about it. You need a set of technologies to, to support that thing, but then you also need a team that can deliver. So you need this like platform piece, you need this delivery piece. And typically, you know, those organizations can get large and you want to have an executive sort of overseeing that, that stuff.

How would you advocate for it? I think from my time at Biogen, the way that this happened is there was enough groundswell of impact. Like there was enough evidence and, you know, like, so imagine you're an executive, right? You starting to see things

bubble up like a couple of examples. You're like, oh yeah, that is useful. And then you're also reading in the news, CIO magazine or whatever. Hey, you have an AI capability. And I think that these two things together are starting to create pressure to create in these separate functions. But I think every company, the way it will do it, it'll be a little bit different.

Thank you. I love it. These are fantastic questions, y'all. And I've got some coming into Slido that I would love to ask. So I see one from Morgana. Morgana, are you available to unmute and ask?

I think so. Hopefully. I think so. Hopefully. Awesome. So my question was, working with healthcare data, have you noticed any changes in how you think about and manage data privacy as LLMs and AI become more mainstream? Data privacy.

Data privacy and LLMs in healthcare

Yes. But I don't think it's quite as far away from where we are prior to LLMs existing. So identified patient data has existed for a long time in systems. And in many ways, we already had processes in place to keep data private, to protect it. And when you add in the LLM component, it's really more of an extension of that. It's not like something completely new. Sure, the AI has extra stuff. I think the concept of responsible AI, sort of monitoring how it outputs, that is sort of new. But the data privacy stuff, I don't really see that. I think it fits within how data was already being managed today, or at least that's what I've seen in my time here.

And I think that that's the other key thing is, I try to focus on keeping this as kind of an extension of things we already have. I think if you think about it as this brand new capability technology, it makes it less clear how to like act, like how to manage it. And if you think of it more, it's like a new piece of technology. In many cases, you probably have like policies in place that can kind of cover it. You just have to think about it a little bit differently. Like, okay, so instead of it, what if it generates new insights or identified information? How do we make sure that that's managed as the same way as it did before, the data before it went to the LLM? And like I said, we haven't had an issue with this, but truthfully, we don't really work extensively with like PII because we run clinical trials, that's in the whole clinical trial system. We have real world evidence, which is already de-identified. So it's not a huge problem. But if you worked at like a lab testing company, like I used to do or a health insurer, maybe this looks a little different.

That makes a lot of sense. Thank you. Thank you for the wonderful and thoughtful question, Margana. I have some questions along that same line. I think that there are some other people who have really similar ones. Like Sadie asked a really similar question on how to train with secure data. So I think I'm going to count, Sadie, that your question is probably covered by what Jay just said. So I will go ahead and mark that one as answered. Oh, hang on. Let me train with secure data. Yeah, I think that is a little subtly different. Let's go with that then.

What "training" really means

So like, let's be clear about what we mean by training. So I hear the word training a lot and a lot of times it's the person that says it doesn't understand how this technology works. In fact, I had a call earlier today. Someone said, so when you use the model and it learns from your inputs, how do you control that? And I'm like, you don't really understand how this works. Like, so I think when you say training, the thing that comes to my head is you mean either building an LLM from scratch, which nobody does anymore, except the major research labs, where you're talking about fine tuning, which is you take an existing model, you give it a new set of data, and you try to like use gradient descent and deep learning to get it to respond in a way that you like.

So I think that historically fine tuning, like historically as in like two years ago, one year ago, that was a legitimate technique to try to get the model to fit your output more. I think that I am more skeptical of those techniques now. I think that there are things like retrieval augmented generation, which is, again, I don't know if folks know about what this is, but the cheat, the hack way to get new information available to the model is you basically take your content, the stuff you care about, and you like, you don't try to train it, you don't try to put it into the model. You just sort of like put it over to the side in like this like database thing. And then when you ask a question, you have this mechanism that like looks through that document database and pulls the right piece out. To me, that is the more practical approach to it.

To me, that is the more practical approach to it. And there's a lot of ways you can control that with data privacy, right? If it's a database, you just have the same data privacy safeguards on there. If it's a model, you use the model within your, you know, ring-fenced IT zone. So there's ways in which you can kind of get the data in there and make it available.

And there's a lot of ways you can control that with data privacy, right? If it's a database, you just have the same data privacy safeguards on there. If it's a model, you use the model within your, you know, ring-fenced IT zone. So there's ways in which you can kind of get the data in there and make it available. But really what I'm talking about is more of an engineering solution. It's not like an LLM technology problem. So I think there's a lot of ways you can look at it as like, how do we just make it safe? And there's ways to get there.

Awesome. Thank you for your helpful explanations. I am really appreciating, Jay, the way that you are explaining things in sort of plain language and not using a bunch of huge terms. I appreciate it. Maybe everybody else is okay with the big ones. I have a lot of these types of meetings. Yeah. I really prefer, even if I know the big words, to use the normal words because then nobody misses out on the explanation.

Dealing with AI hallucinations

All right. I have a great question that has come in from Zach. Zach, are you available to ask your question live? There's always the awkward pause as I wait to see if somebody is able to unmute. And if not, Zach, I can totally ask it. Okay. I don't see you unmuting. So I will go ahead and ask, how do you deal with the possibility of AI hallucinations? Is the risk and consequences of hallucinations especially high in the context of healthcare?

So when I embark on a project, so I don't think I've actually explained what I do very well. So let me set the stage of what my team does, and then I'll try to answer the question. So I have responsibilities for two domains. I have AI platforms, which today are not products, our enterprise data lake, and then some other random technology. And then I also have a delivery team, which is focused on generative AI and traditional ML stuff. So I kind of get to do both roles.

So over the past year or so, we've really refined the way in which a project comes into the queue and we execute it. Because what I've discovered is there's unlimited demand, like there's unlimited demand for build a gen AI thing that does something useful. And so when you have that problem, you have to get really efficient on execution. And then you have to get really good at, how do you scope the problem so the people are happy, right? A bad project is one that takes six months and you're chasing hallucinations and you never get to the finish line. A good project is one where there's clear expectations, you can deliver it quickly and stuff like that.

So to this question of hallucinations, how do you deal with it? When I talk with people about building a gen AI workflow or solution or whatever, I say that this is an intrinsic part of the solution. You can't box it out. It's a feature of the LLMs. You will be unable to physically remove hallucinations from it. Now you can minimize them, right? And that's what the goal is, is you built a solution that minimizes that problem, but you need to have a solution that's resilient to the hallucinations.

You can't box it out. It's a feature of the LLMs. You will be unable to physically remove hallucinations from it. Now you can minimize them, right? And that's what the goal is, is you built a solution that minimizes that problem, but you need to have a solution that's resilient to the hallucinations.

And so when you frame it that way, it kind of gets people's minds turning in a different direction. So you're like, okay, we need this secondary verification system. That could be a human in the loop thing, or maybe what it generates, if it's wrong, it's not that big a deal. If you have a question and answer chat bot or something like that, and it's informing you on the HR procedures or what your benefits are or something like that, the risks of the hallucination there are probably lower than if it's like, I just messed up something in the bioreactor, what do I do next, right? So same technology, but the risks of it being wrong are different.

But there's other ways you can do secondary verification. So maybe you have the solution extracts information, and you can verify with independent information. So when you do that, yeah, the model might hallucinate, but it'll at least focus you in on the thing that it hallucinated on. So if the data matched 90% of what was in a source system, and 10% is wrong, well, now you're inspecting 10%, you're not inspecting 90%. So I think framing it that way has been very effective for us. And just like attacking the fallibility head on, just not hiding it and then coming up with really good ways to mitigate it as part of the solution.

Repeatable solutions and the role of Posit Connect

Awesome. Thank you. I have a question here in Slido that's getting lots of updates. That's from David Kegley. David, are you available to unmute and ask your question? If not, I am also happy to ask it for you. Hi. Yeah. So earlier, you mentioned repeatability was really important for your team. And I'm curious what technology choices you've made in support of creating repeatable solutions? And then like sort of a leading part of this question is, do containers play a role in any of that?

So when I think of repeatable solutions, I'm thinking in the context of like a chatbot, right? Like let's say you want to build a chatbot on documents. How do you like, what's the code to do that? So I think of those as, when I say repeatable, I mean like repeatable design patterns. So I think of them in terms of like code. And I think that there's this question of, there's some of these frameworks that exist now to help make this better. So LangChain is one of the most popular ones. There's also something called Llama Index. And then, you know, you're starting to see that the big research labs like OpenAI are starting to embed some of those capabilities within their SDK.

To me, the jury's still out on a specific framework. We do use LangChain, but specifically, we use their LangChain expression language because it's supposed to be more robust. I think that, I'm not sure yet if there's a mature framework we'll go all in on with that, but I don't really think of it like a container problem. It's really more the amount of code you need to write to build one of these solutions is actually pretty low. And something like Posit Connect product plug works really well for this. So imagine you're building some chatbot that you want other people to use. Sure, you could use Streamlit, but you could also do it in Shiny, like Shiny for R, and you could deploy it to Posit Connect and it could achieve the exact same thing as that. So I try not to get, I operate from a perspective of simplicity. So if I can do it with no packages, that's better than doing it with something else. Like you got to get something for adding complexity. And to me, adding the complexity of container and container management has not yet made sense for what we're doing.

Awesome, thank you. I think that this is a great place to sort of just like ask how you are using Connect. Like how does Connect play into the problems that you solve or that you're able to solve or experiments that you're able to run? So I'm an old R user, I guess. So I graduated from grad school in 2013 with my master's in analytics. And the company I worked at did not have SaaS products. So I learned everything in SaaS, they didn't pay SaaS. So immediately kind of switched over to R. And that's where I really fell in love with the products, saw the tidyverse sort of develop, saw Shiny get created, see the development of Posit Connect. So I think that I am purely from the R ecosystem. I'm an R guy, like through and through. I was in the first R and pharma conference in 2017.

So I am thinking, I think of all these AI, Gen AI workflow things through that lens, right? And when I look at like medium or something, let's say you go say, how do I build a chatbot with like tools? You go to the medium articles, they're all doing it in Streamlit. I'm like, Streamlit sucks. Why don't we just do this in like Shiny? Because that's a much better framework. It's more mature, it's got more reactivity. And so I think that the Posit Connect product specifically is extremely useful because I think someone, the previous guy was talking about containers. One of the things I like about Posit Connect is it automates a lot of the DevOps stuff away from you. You don't have to care about it. You just, you bring your requirements file, you bring your contents file and it just builds it for you and deploys it and you get all the role-based access control. So it really simplifies the deployment process. And when you're trying to like rapidly build something that people can use, that makes a big difference. So people on my team were able to very rapidly understand and use Posit Connect. If we didn't have that product, I would have to go into Azure and figure out how to give them deployment permissions and then talk with the cloud engineering team. Because it's a discrete product, it makes that whole deployment process a lot easier. So that's why I really like it as opposed to a true DevOps pipeline.

Awesome. Thank you. I appreciate the explanations.

I have so many questions in Slido that are so good, I cannot even tell you. So everybody, please hop into Slido. I am going to, if someone will reshare the Slido link for anybody who has popped in partway through. Reminder, put your name on your questions, put an asterisk there if you'd like me to ask them. Otherwise, I will call on you to see if you want to unmute and ask.

Connor, your question has gotten so many upvotes. Would you like to ask it live?

Sure. Yeah, thanks for hopping on here. This is a great talk. So yeah, my question is, how do you talk to stakeholders or management slash VPs if they have some idea or they want to use some new AI thing that they saw on some business project, but the data required for that project is not available slash governed slash cleaned up? I've seen cases where people want to skip straight from the idea on the PowerPoint slide to the finished AI product without understanding the requirements of the data.

So, okay. Number one, you, the person dealing with this, need to have enough knowledge of how these tools work to be able to answer it effectively. I think that there is a prerequisite that you understand what is and is not possible with the technology, and that requires you to have tried and used it in a variety of different ways. So, like, I guess my message to folks is try to use this stuff, get a feel for how it works, what it's good and bad at, and then it'll more prepare you for these types of discussions.

But I think the data, so the way I always frame this is, again, I spend a lot of time trying to explain AI and the way it can be applied, and the way I do it is there's inputs and outputs, and typically I say if you can give the model the adequate context, it can probably answer the question, right? If you can just figure out how to get the information in some engineered solution, generally it works pretty well. The problem is a lot of business process is not written down. There's a lot of context that may not be in the database, right? Like, there's a number, but a lot of times there's the context behind the number, and the model is incapable of knowing that. So, if you kind of get them thinking about, okay, how do we frame this in a way where it can actually understand the bigger picture, and maybe that's possible, maybe it's not. But a lot of times what I do is I don't try to, like, push them off. I'll try to try to deeply understand what they're trying to achieve, and there may be a smaller component you can solve with this. Maybe you don't solve the whole process. Maybe you only solve one piece, and then that actually helps you get, like, build momentum because it'll be faster to build, and then you'll have evidence that it works, right? And so, I think I try to scope them down smaller, and then you get that critical mass that can expand it, and then you attack that problem later. But, I mean, it is a legit problem. It happens a lot, but yeah, that's pretty much what I do.

But a lot of times what I do is I don't try to, like, push them off. I'll try to try to deeply understand what they're trying to achieve, and there may be a smaller component you can solve with this. Maybe you don't solve the whole process. Maybe you only solve one piece, and then that actually helps you get, like, build momentum because it'll be faster to build, and then you'll have evidence that it works, right?

Thanks, Jay. Thank you so much for the fantastic question.

I have a great question from Rohan. Rohan, are you able to ask yours live?

Yes, I am. Can you guys hear me okay?

Yes.

Fantastic. So, the initial question is, are there any differences in job description between a head of AI and a leader principal data scientist? Some context for this is, I often tell my non-data scientist friends who are talking about AI that, for example, linear regression is a form of AI. When I think of a leader principal data scientist, they are managing multiple data scientists who are doing their own data science projects. And so, I guess an update to my question is, are all data science teams now essentially AI teams that are also adding generative AI projects to their portfolio? Let me know if that makes sense or not.

No, this is a great question. This is actually something I wrote down before the talk, what's happening to data science. And when I look at, so my team, my team are data scientists, historically trained, classically trained, you know, regression, all that stuff. And now they're very effectively running gen AI projects. And the way I look at it is, this is like a tool in the tool belt. Like AI is not, while it's new and shiny and everyone wants to get into it, data scientists are uniquely positioned to be effective at this. The difference between a data scientist that does shiny apps and a gen AI developer that builds AI workflows is an LLM call. The way you frame the problem is the same. You have a business problem. You're trying to figure out how to frame it and execute a series of steps to generate a desired output. Before you had shiny apps, maybe have an algorithm. Now we have an LLM to kind of stick in there, but all of the other things around it are the same.

So, I think that AI teams or excuse me, data science teams should add this as a tool in the toolkit. Will all data science teams turn into AI teams? Maybe, potentially. I think that it's, if you don't and someone does this, that will be bad. I think if you push off AI as a data science team, you're at risk of being marginalized because the capabilities of the models and the tooling is increasing. And so, what do we do when the tool can do many of the things that we can do, right? You don't want to be stuck as the manual verification step. You want to be building the tools because you know what the model should be doing, right? So, I think you bring a lot to the table as a data scientist in that world.

The difference between a head of AI and a principal data scientist, to me, they're fundamentally different. Principal data scientist is a lead. They're a tech lead. They're a person that understands the landscape of technology. They have experience managing teams, but they're not setting the strategic direction. The head of AI, to me, is a person that manages teams. They're understanding how to position AI, the technology within the business, developing relationships with the stakeholders, kind of having higher oversight on the products that the teams are building. So, I think of like a principal to be like a senior manager AD role. Head of AI is like a senior director, you know, VP type of role. Awesome. Thank you. It's a great question, and it looks like other people in the chat said that it's a great question as well.

Q&A: benefits and drawbacks of generative AI

I see Sarah appreciated you asking that one. We have another question that has risen to the top here. Reminder, go in Slido and vote for the questions that you would like to be asked because that feature is really fun, and it's nice seeing things bubble to the top when a really good one gets asked. Arsenis, are you available to ask your question live? Yeah, I am, Libby. Thanks. And thank you, Jay, for joining us today. So, a quick question, and a little bit of context before I ask it. I'm asking about kind of broadly speaking drawbacks and benefits of AI and in two different domains. So, my question is, what are some of the types of work that Gen AI has helped with? And if any, what are some of the drawbacks that you found? We've mentioned hallucinations here, for example, as one of the things. And if you could also talk about it both from a kind of like broad data science perspective and as someone who's worked specifically in the domain of the biological sciences. My experience is primarily social science, so I'm curious about maybe some of the differences.

Okay. Don't know how much I can say about that last one, but I'll definitely try to answer the first two. So, where is the type of work Gen AI has helped in this? So, let me give you a very kind of stupidly simple example. So, in our organization, we use safe agile. So, safe agile is agile, but with more meetings. And so, what you do is you have to write user stories and stuff like that. And it's very formal with how things are written. And you can imagine nobody likes that. No one writes the stories correctly. It's always a pain. It's always a struggle. You always get yelled at because you don't write them properly. Now, one opportunity with Gen AI was what if we built a tool, a Shiny app, a Streamlit app, something that can connect to where the stories are, and it can just rewrite them the proper way. And you literally just have someone that connects to it, rewrites it, you check it, and you click the button, and it's updated. There's nothing there. Like that's not a complicated solution, but it added a lot of value because it kind of automated the thing that people just don't like, they didn't want to do. And I think that there's a lot of opportunities like that in every company. It's not just unique to Pharma.

So, anything where you're reading text and having to extract information or process it, or to me, that is the richest domain of these tools. So, this is actually disproportionately things like corporate functions. You think of people that look at contracts all day. Think of people that have to write a lot, medical writers. So, I think that there's a lot of writing, reading stuff, which is a huge domain here that becomes available with the technology.

I think the drawbacks of it are there's a learning curve to using it. So, when I build a solution, I've said this from the beginning, the distance between failure and success is very, very thin. It actually is pretty hard to get a solution that works well and is robust. You may fail nine times, and then somehow with some little tweak of something, it just magically starts being more effective. So, I think that educating people on that has been a real struggle because people just assume you're reading the news, you got agents or whatever, they can just magically do everyone's job. And the reality is it kind of sets unrealistic expectations that you have to battle with education and experience. And so, I have to spend a lot more energy explaining people how things work versus data science. You didn't really have to do that before, right? So, bring me some data, I'll figure out some information, and we'll be happy. Now, I have to explain how this thing works. Where it's in bioscience, I can't be too specific, but there are, like I said, reading data, doing research, writing, that kind of stuff, stuff that's around text. There's a huge domain of that across the business where you can apply the technology at it, and if you frame it the right way, you can create a lot of value. Awesome. Appreciate it. And fantastic question again.

It actually is pretty hard to get a solution that works well and is robust. You may fail nine times, and then somehow with some little tweak of something, it just magically starts being more effective.

Classification models and the rise of zero-shot capabilities

I have another question that has risen to the top, and this one says, Libby, please ask it for me. So, I would be happy to. This is from Maxim's, and it is, what kind of models are used in your team for classification problems that are not built from scratch? If you have any classification problems that you're working on. If not, I'll let you take that question in whichever direction you'd like to go.

So, like, is the question, like, what custom classification algorithms have we built? Perhaps. What kind of models are used in your team for classification problems that are not built from scratch? So, maybe pre-built. Pre-trained. Oh, yeah, yeah. Classification models. So, increasingly, these LLM and the multimodal models are eating this.

I'll give you an example. So, a few years ago, I was involved in training a computer vision model. So, the typical recipe for computer vision is you take a pre-trained convolutional neural network model. You get a bunch of images of cats and dogs, and then you have the picture of the cat, and you say, that's a cat. You get a picture of the dog, that's a dog. You give enough examples of this. You put it in this algorithm to train it, and then, voila, on the other side, you have a model that can look at cats and dogs and generate that. What has happened, though, is more and more of these powerful pre-trained models allow you to do something called zero-shot, and I think this is something to pay attention to because as the capabilities of these models increase, the ability to zero-shot things that are useful will increase, and we don't really know what that means, right? So, in my computer vision model, right, it took a person that knew what was going on. You had to have infrastructure. You had to have, like, people to do it. So, there's actually a lot of overhead to training the model, but what happens when I see an LLM, I can give it a picture, and it can identify that at the same level of accuracy? Well, there's no training. There's, like, now a capability is just there. I just need to apply it, and so I think that that's what I see increasingly is the LLMs are going to continue expanding and eating the world of this, and it's really about what I spend a lot of time doing. It's just, I'm not inventing anything. I'm just, like, let me just apply it at a problem and see if it works, right, and you'll know pretty quickly if it works or it doesn't, and if it does, cool. You just solved a huge problem with no effort. Get a lot of credit, and if it's a harder problem, you kind of get a sense of, okay, how close is this technology to solving it, and then when you look at vendors in the universe, are they lying? Are they bringing some intellectual property to the table that actually helps solve this? It kind of helps you frame out how you view other people solving that problem. Yeah, this is a great spot to talk about, like, data quality

and the importance of data quality, especially if you're doing zero shot or one shot or a few shot. For anybody who is Googling what those are, one shot might be your data has, you know, one example of this type of thing that it's trying to predict. A few shot, it has a few examples of this thing that it's trying to predict. Zero shot means it has none, which is a lot harder to do, but it's becoming more and more possible as we are moving ahead in our sort of data science progress journey.

I have a few more questions here that are really great and getting lots of upvotes. I would love to get to them. They're anonymous, so I will be asking them, but please, everybody, if you're willing to put your name on your question, I would love for you to ask it live because these are great. This one is about the hiring process at Biogen. If you are able to answer anything about this, we would love that. What does the hiring process look like for you? How do you weigh data science skill sets versus a background in biology, for example? As a head of AI, do you anticipate skill set shifts as AI becomes more popular? I love this question. Whoever asked this anonymous person, thank you.

I mean, what I can say about the Biogen hiring process is it's probably like most companies, you know, there's a job, you apply, HR is involved. So, I mean, it's pretty typical. For like, for the roles, like, I can answer this from the context of my team. So, my team is a it's a techno-focused organization. We actually don't intentionally resource based on, you know, background and other domains. Because we're this enterprise IT organization, everyone kind of assumes we're clueless. So, they bring the SMEs to the table when we try to solve a problem. So, I think that that background is helpful. But the thing that's much more important in my team, not all teams, but Jay's team, is really the hard technical skills and the ability to kind of manage requirements, you know, talk to stakeholders, kind of like the more traditional data science-y kind of broad-based skill set.

Build versus buy

All right. I had a question here that is anonymous as well, and I will ask it. It's about the difference between grift and value. Software has always had tension between build and buy. What are your thoughts? Okay, build. So, let me answer the build versus buy question, because that one's, that's a really, really important one.

This is an open question, and it actually is related to grift. The way I look at this is, I have a team that's focused on delivery. We have access to all the same tools that every other vendor has in the universe, right? Like, it's not like some SaaS company, S-A-A-S, not S-A-S, some SaaS company has some magical access to models that we don't have. We're all starting from the same level. So, my thing is, okay, if I see a technology, we know the technology is very, very powerful. What is the vendor bringing to the table? Are they bringing intellectual property or expertise or something like a more mature product that is truly differentiated from Jay slapping a wrapper and a streamlined app together to solve a problem? And that's really where, that's my anchoring is, if I had to build this in a streamlined app in three weeks, how much better is what I'm seeing than that?

And as you start to interrogate what the tool does, in many cases, it falls down. Like, a lot of times the solution is a React front end on a RAG chat bot. And you start to ask, well, did you fine tune it? No. Are you using some kind of proprietary technology? No. It's like, you just have a fancy interface on something that's already available. So, I think that when I think of build versus buy, it's like, how long would it take to build that internally? What are the functional requirements of it? Are we supporting 10 people or 10,000, right? Do we have the skills internally to build it? How quickly is the answer needed? So, I think that it's really a balance of power of the product, what's our internal capability, and then it will help you make a more informed decision.

But it's also based on your organization having people to do this, right? If your organization has no people building GNI solutions, then they're extremely susceptible to this because there is no alternative. There is no person to build the Streamlit app. They're just, this is the solution and there's nothing else. So, whatever the price is, is the price of the product. So, I think that's another thing we can do to defend our companies is build this capability so we have a viable alternative to buying something that's not that great.

But it's also based on your organization having people to do this, right? If your organization has no people building GNI solutions, then they're extremely susceptible to this because there is no alternative. So, I think that's another thing we can do to defend our companies is build this capability so we have a viable alternative to buying something that's not that great.

Career advice

Thank you. I have a question that has popped up from Rachel and has shot straight to the top. Rachel, would you like to ask? Sure, or you could ask it. So, we like to always ask people at the Hangout about career advice. And I think we usually frame it like, what is a memorable piece of career advice that you either received from somebody else or that you've given to someone on your team? Because we would love to learn that from you.

I give a lot of unsolicited advice. That's unfortunately a flaw in my personality. So, I would say, I think when I try to solve a problem, I try to come at it with coherence. And I think in the AI space, this is very important. Because there's so many pieces moving and there's so much technology, a lot of times if you just anchor on, does this make sense? Can I explain this to another person and it makes sense? And if you can do that effectively, that is a very useful skill because it's actually a tool you can apply to any situation, right? Like you're dealing with a consultant or you're dealing with a domain you don't understand. And it's like, if you can try to get the coherence in your understanding and explanation, that's a very, very useful skill. And so, that's what I try to do with everything I put together.

We love the career advice questions. They are my favorite. I put together a blog post of some of these bits of career advice and I smell another one coming because we get such great ones.

Stakeholder engagement in Gen AI

Alrighty, let's see with our very few minutes left, we have five minutes left and we should probably wrap up in a couple minutes. So, let's see what else we can ask here. I do want to make one more point before I do that. There's an important thing that is different or that is uniquely different about Gen AI versus traditional data science. Stakeholder engagement is the difference between success and failure. It actually doesn't matter how tangible the problem is. If you don't have that stakeholder engagement to sort of direct you on what looks right or wrong, that is where it's failed for me. It's not even like, can it solve the problem? But if they won't show up to meetings or they don't give you the good guidance, the product that you produce will not be very good. So, I emphasize that when I start to engage with people about building solutions, they need to commit to building out the product. And when you get that commitment, it's significantly better.

Stakeholder engagement is the difference between success and failure. It actually doesn't matter how tangible the problem is. If you don't have that stakeholder engagement to sort of direct you on what looks right or wrong, that is where it's failed for me.

Human-centered language for AI

Awesome. Thank you. All right. I am considering this next question, dealer's choice. I'm going to get to pick one. So, I went through and I found an anonymous one that I like, which is, what is your opinion on using human-centered terms for AI? So, learning, thinking, reasoning, when you're describing the operations of large language models. Do you believe that using these terms might lead to misunderstandings about their actual capabilities or do they serve as useful analogies?

I avoid them. What I do is I try to frame this again. I try to use simple language and frame it in the context of things people already understand. So, what I tell people is, an LLM is a text computer. You have a CPU in your computer, an LLM is a text computer. And if you can provide information to the LLM, it will be able to compute and output that way. That is a more like intuitive way I try to explain it versus, yeah, I can learn and reason. And then you're like, well, is it really doing that? Well, what is it actually doing? And then that becomes this very long tailed, you know, rabbit hole discussion. So, I try to avoid that. I try to treat it like it's just another technology. And I feel like that gives you credibility because you're not using jargon. Like, I really avoid jargon. I try as much as possible to avoid jargon.

And that's what, you know, for the RAG example, right? Like, could I have said embeddings, vector databases? You know, like, I just said, no, it's like this like thing next to the model where the documents go. And then, you know, it just goes over there and gets it. That's about the level of understanding people need to have to use the products. It doesn't really matter how much depth you have there because the application of the technology is where the value is. It's not in them understanding how it works.

Awesome. Well, I think that we should probably wrap up and give everybody a chance to hop to their next meeting. I would love to let everybody know that we have Bill Wilkins next week, SVP of advanced analytics with practical applications at Safety National Casualty Corporation. That one's a mouthful. So, please come and join us next week. We love having you here. Jay, this has been a fantastic discussion. If you would like to go find Jay, he's Jay Timmerman on LinkedIn. I think you're the only Jay Timmerman that I found on LinkedIn. But we will make sure to gather some of these resources and put them up with the YouTube video. Jay, thanks so much for joining us. Absolutely. It was a pleasure. Thanks, everyone. Appreciate the questions.

Thank you. Thank you, everybody. Fantastic questions today. Have a wonderful Thursday. And Libby, somebody gave us feedback to just leave the Zoom running for about 15 more seconds so they can grab any links from the chat too. Yes. And you can also click the three dots and save the chat if you would like it. If there were a bunch of resources and you're like, oh, no, they're going to go away. There's a little three dots. I think Zoom moved it. So, if you have an older version of Zoom, it might be in a different spot, but there should be three dots somewhere that you can click and then click Save Chat. For me, it's up at the top right-hand corner. Yeah. Mine's top right too. Older versions, it'll be bottom left. I see Uday asking Jay's full name. It's Jay Timmerman. And you can find him and his picture and everything and a link to his LinkedIn on our website. If you just Google Posit Data Science Hangout, the first hit is usually our website and you can see Jay's picture there.

Thanks, Sundae. Thanks, Sheldon. It's so nice seeing everybody. Thank you all so much. Have a great rest of the day. All right. We'll close it out now. Thanks, Uday. Bye, everybody. Bye.