Beyond FOMO — Keeping Up-to-Date in AI

Speaker 1 [00:07]

Much more people than I expected. Welcome to my little talk, Beyond FOMO, keeping up to date in AI. Let me quickly introduce myself. My name is Carsten, I'm a mathematician, data science consultant and team lead at Sunvert, IT consultancy with special focus on data and analytics. I'm based in Hamburg and I work in the field of data and analytics since 2018, so for seven years now with a focus on forecasting optimization in the last two years generative AI as well I try to connect those two worlds a bit and I have a focus on end-to-end solutions preferably in the cloud and if I look back on the last seven years like at the very beginning some of you might work in this field for even longer what were the topics and tools one worked with classic AI, classic data science so to speak, scikit-learn, deep learning with PyTorch, TensorFlow. There was already the world of natural language processing, hugging phase where one could have a little sneak preview on fancy newer models. Different model architectures like XGBoost, CatBoost, LightGBM were the topic. In recent years, still in the stream of classic AI, Polars, dbt, pandas 2.0, the PyArrow revolution so to speak, so like this data processing topics coming up, XGBoost, there are a couple of auto machine learning frameworks out there nowadays but I would say that especially the year 2020 with the release of of ChatGPT marked a time where things have taken off. ChatGPT based on the GPT 3.5 model. So there were already GPTs out there, but the hype started at that point, I'd say. Followed Followed by Gemini, Google and Claude, Lama opened the development of open source models followed by different architectures, Alpaca, Mistral, then all the topics one needed to own and to work with, prompt engineering, RAC mechanisms, vector databases, how to fine Tune, LLMs, we also had a lot of talks on structured output and function calling today. Then lately reasoning and the mixture of experts, so many different concepts come into play. Agent systems, multi-agent systems and different frameworks, deep seek was a bit hot topic a couple of weeks ago. the moment I would say it's MCP protocol. This is all the block of large language models and there is also like this block of image generation, stable diffusion, deli, mid journey, flux, you name it, Sora. And then there are the capabilities in the cloud. SageMaker, Bedrock, Vertex AI, the Azure ML platform does not really fit to the timeframe here but I think you know where I want to get. Databricks, NVIDIA, and now we're here and have all these nice and interesting PyData talks every 30 or 45 minutes. And I don't know what about you, but I feel sometimes like this Matthew McConaughey meme. How the heck should I follow? And I want to talk about two phenomenons, and the first thing is FOMO, the fear of missing out. I have a definition here, the anxiety that an exciting or interesting event may currently be happening elsewhere. This comes originally from social media, but I'd say that this is a topic in a professional context too. Especially in an AI context, this might be the fear that a competitor is using newer or stronger or better technologies and tools. Someone is more at the forefront or personally concerned about my own career relevance. Is it still relevant, the tools I'm working with or what I do? Both are driven by the anxiety about missing key innovations. And I researched a bit and what research tells us about FOMO, spoiler, just bad things. It is linked to decreased well-being and life satisfaction. It is associated with rumination and a decreased focus. It can create a continuous partial attention state. Think about people having LinkedIn open in the tub all the day and checking it 10 times a day. This is like the same as you would scroll through Instagram, isn't it? It triggers a reward-seeking behavior or it can trigger reward-seeking behavior patterns. Again, LinkedIn maybe. People posting. But yeah, it can lead to reduced productivity up until burnout. And one statistic, no, no, statistics come first. Yeah, it can at the end lead to burnout. Then there is another phenomenon, sorry. What I want to talk about is the so-called impostor syndrome or feeling like a fraud. Why the heck am I standing here in front of 100 participants and giving a talk? Do I have enough experience for that? It is the persistent inability to believe that one's success is deserved or has been legitimately achieved. And people who face this have constant fear of being exposed as a fraud despite the evidence of competence. One attributes personal achievement to luck, to timing or external factors rather than personal skills. and one phenomenon is that one discounts positive feedback while fixing on perceived mistakes and shortcomings. I think this is a typical human behavior in general, but here you have another stronger ratio of it. And it can result in setting excessively high standards and experiencing intense disappointment when I don't meet these standards. And this leads to setting even higher standards, not meeting them, being disappointed, and so on and so forth. And here's the statistics. 70% of people experience it at some point in their careers. Wow. So almost two of three, more than two of three. The imposter syndrome in tech and AI, research shows even a higher prevalence in STEM fields. Why is this the case? I think one reason is that AI in general requires mastery across multiple domains. We speak about math, programming, domain expertise, knowledge of algorithms, systems, etc., etc. And, well, the broader the field, the more knowledge areas there are, right? And what we saw at the very beginning is that there are coming up more and more and more fields and areas and technologies and so on and so forth. also the rapid evolution makes expertise or may make expertise temporary and the public visibility of work and contributions in an open source environment which is a good thing and the community which is a good thing that we share our our work together but it might create a comparison culture by accident and then there's this paradox the more you learn the more you realize you don't know I mean you read a paper 20 references dive into one reference 20 new references and so on and so forth how to handle that okay so far I just described problems but let's step let's do one step back and ask what is AI really for why are we interested in it why do we want to make use of it build something out of it. What actually matters is creating real value using AI to create value and technology in general and my take is that technology creates value if it's embedded in the functioning system that delivers value. if stakeholders and our users trust and accept it a working system then can always act as a benchmark I have just random example here of an architecture some data is loaded by AWS glue to some data warehouse pre-processed whatever and then I have an agent system and two of these agents make use of cloud LLM, just a random example. But in that moment I have this and build it, well, I have a system, a working system, which then can act as a benchmark and I can now react, hey, did you hear about cool, nice unit testing capabilities of dbt? Okay, one can think about using DBT and build this in the system. Or DeepSeq is way more efficient and I tried it out and it does what the third agent does here. Okay, so one can make use of DeepSeq and exchange one part of the system. But in general, problem solving is often more important than the tool mastery. and an incremental improvement should always be prioritized over a revolutionary change because the truth is that most businesses don't require cutting edge solutions so like it's not necessary that something which came out yesterday is to be implemented today or tomorrow maybe it's okay still to make use of a technology in two months, three months and wait a bit until So it's adopted somewhere else. You can get a sense of success there as we get to that later. And I personally had the situation. So I was working in a forecasting project, really large one, for more than two years, when the Gen AI hype started. And I was like, okay, I have this never-ending stream of blogs and posts, and I realized, okay, what to do. and no LLM was necessary in that project so what to do how to keep the ball rolling and I finally managed to bring a football reference to one of my slides which I tried for a couple of years now how to keep the ball rolling and I want to share some practical strategies from my experience and also from from our team how we handle it first thing is that recognizing FOMO and this imposter syndrome is the first step to weaken its impact so if I recognize okay this is a feeling a thought I'm not 100% driven by this thought anymore so to be aware of is the first step. Then when it comes to knowledge sharing I assume that most of you work in teams so my second take is that the team you work in should always be the starting point where one can learn from each other, cultivating a knowledge sharing culture. We are a team of six data scientists and we try to lift this culture. Actually we have two policies for knowledge sharing with dedicated sessions, community of practice, whatever, and the first policy is there are no stupid questions, never ever, because well this should be our safe space this is our team this is not customer facing yeah we work in consultancy but like internally there are no stupid questions every question is welcome and the second policy is we always start on the high level and then dive deeper because well the deeper one gets the more knowledge is required to follow and more or less some at some point you will lose someone who is not that into that topic as you are so this is what we do we always start on a high level and then dive deeper and also feedback is welcome in a sense of hey wait a second let's come back two steps and let's clarify again concept XY because I'm not able to follow it yes side projects always a nice idea to dive into something where one can simply try it out I always would recommend to solve a real world problem whatever out of your private life so like the first rack system I mentioned I was working in this forecasting project was not a customer project but was like just a rack system I built for myself against my internal documents so that I can ask a chatbot, hey, whatever, when was the last time I went to the doctor or whatever. Another cool thing is I built a summary and a wrapping and summary of different football results and it sends me a Slack message every Monday. But I learned a lot on building a system, making use of LLMs and summarization, prompting, stuff like that at that point. Speaking about LLMs, I can highly recommend to make use of what is already out there. Use AI to learn AI. Recently I had a little chat with Claude and I revisited the concept of transformers and multi-head attention. And I tend to forget 50% of the details every time. But yeah, I came back and really dived into that topic simply via a chat. Also, what I can highly recommend is Google Notebook LM. It's quite cool. One can upload different papers or blog posts or some notes and click on create a podcast. And then you get a podcast, an artificial podcast, where two people speak about this topic building an individual input stream also is what I always aim for making use of different podcasts, newsletters blogs whatever still hopefully not too much because then you end up again in this rat race of too much input and like the imposter syndrome and you know, okay, this is out there, this is out there, this is out there. Personally, I can recommend two podcasts, two newsletters. The first one is Handelsblatt KI Briefing, German newsletter. And it's more from a business perspective, like what's going on and different topics and the economy where can AI be adopted and so on and so forth. and the other one is the batch from deep learning AI where one has like a more deeper approach and papers are referenced and repos are referenced and so on and so forth and yeah I think like for me this helps to have these two words and you get an input every Friday. okay and one truth is occasionally I have to block time for deep dives because especially in a working field where one is interrupted by slack messages meetings whatever just in the normal working week I think it's quite hard to take time and really dive into something really deep and this is one of my takes, sometimes that is simply needed to take the time that easy in general focus on application and success stories success stories not involved because success stories also can sometimes trigger the imposter syndrome, I once had a talk where we presented with a client of us, success story two years of work in 40 minutes well but I want to share one cool story from a meetup I visited a couple of weeks ago these guys presented okay this is our business problem it was like from a logistics sector and they made use of an agent LLM system to extract information bring it in the specific structure to semi-automized processes and they showed like okay what we built so far and then they were super honest with okay we are stuck here and opened in discussion can you help us can you share experience and I really like that because it was not super nice and shiny prepared success story hey we are the coolest guys and build this and that but okay here do we stand this is what we achieved and yeah let's let's come into a discussion who has this already had this problem so yeah I really liked like this approach and I already mentioned my chat with a Claude every now and then come back to the basics this always helps to understand okay where technology or where tool XY that is talking on and to get a refreshment of some concepts. Final thoughts before we hopefully have a little discussion. Accept that you can't know everything. It is simply not possible and it's not necessary. Build sustainable learning habits. Accept that growing takes time, one cannot force experience and in general focus on value creation instead of tool mastery when working with AI technologies in particular. Connect with others on the same journey, I think this is what you already do, you come here to the PI Data conference and enjoy being part of this transformative field. Yeah, sometimes it might feel overwhelming, and it is a lot of input. And after a day of, like, whatever, hearing talks six, seven, eight times, I'm exhausted, but it's still a cool thing to be here today. And the cool thing of lifelong learning is that it never ends. So my final take is keep cool, carry on. thanks for listening and hopefully we have some time for a discussion now here in the audience later on with a coffee tomorrow in the evening with a beer reach out at LinkedIn I'm really curious and like to hear your personal journey and your strategies how to handle this never ending stream of AI technology

Speaker 2 [21:48]

Awesome. Thank you very much. Thank you for this great talk. But I'm afraid you have to play a psychologist a bit now because you have some questions.

Speaker 1 [22:00]

Okay.

Speaker 2 [22:01]

As a team lead, how do you encourage your team members to cultivate a knowledge of sharing culture?

Speaker 1 [22:08]

Well, as a team lead, I set the rules, and I could then give a hint, okay, well, can I give you some feedback? I don't think that everybody was able to listen to what you just explained in the last five minutes, 10 minutes, or in the last sharing session, or whatever. like this is this is and I always try to encourage that people give feedback each other but well I'm always the person who can do this as well so making use of feedback short story

Speaker 2 [22:49]

then how do you tackle the struggle of taking time for deep dives in the corporate setting with tight schedules and a busy personal life with Family and kids and the dog I will add as well

Speaker 1 [23:03]

Yeah, very, very good question. Yeah, sometimes one has to prioritize, right? And to maybe, whatever, skip something or to ask the question, okay, what is now worth it to take the time for? And I still need some months to finally answer this question because I will have my first child being born in August, so I don't know that yet.

Speaker 2 [23:37]

Congratulations! Then I really like this one. I could start a whole rant about it. Do you think we should allocate some time doing working hours just for learning? Well, I think yes, but again, coming back to the previous question regarding tight schedules and deadlines and actually, you know, getting work done. What do you think?

Speaker 1 [24:01]

Yeah, so we do it, yeah. We have like one day a month, which is occasionally blocked and like not being in the project, but coming together also in person and to try something out, read something, code together, whatever, yeah. So I would say dedicated time of your working time should be allocated for learning, of course. This is how you move forward.

Speaker 2 [24:31]

Then I also like this one one problem nowadays is that businesses are searching for people who are Experts at cutting-edge technologies how to deal with FOMO when searching for a job I'm telling you here psychology session now, whatever

Speaker 1 [24:48]

Whatever, making use of humor, maybe. I wrote a blog post.

Speaker 2 [24:52]

wrote a blog post that's difficult in Germany

Speaker 1 [24:56]

So like I wrote a blog post on this topic last year and I started with a quote and like it was four weeks or so after the release of GPT-4.0 and on LinkedIn I saw like this quote of searching for someone for a GPT-4.0 developer Requirements five years of experience with it It's always unrealistic what is expected

Speaker 2 [25:24]

How to explain to stakeholders who want AI that value is not in the technology only.

Speaker 1 [25:32]

Again, yeah, I mean, in that case, I always would start with, okay, searching for where I can make use of AI to create value and to step, not one step back, but two steps back, where is really the problem? what do you want to tackle what what is it about is it being more efficient in the in the process x y z do you want to uplift your revenue your sales your whatever to understand that first and then come from there from the problem and then ask the question okay how can i use uh ai and does this does not necessarily always have has to be ai sometimes a simple heuristic or a process which is digitalized also helps.

Speaker 2 [26:23]

Thank you. Then this is also a question I really like because I work in education myself. How do you implement there are no stupid questions? I hear this all the time, but rarely being positively practiced. And yeah, I completely agree with it, because especially if it's in a group, some people are still ashamed to ask it out loud. So I get that in a classroom setting a lot. That they might come to you and approach you personally, but then like in front of others, they don't want to, you know, humiliate themselves.

Speaker 1 [26:49]

Well, even as a senior and team lead, I can sometimes show that I have no idea of whatever latest update or whatever. So to signalize, okay, it's not just you, it's also me, it's all of us as one tool maybe.

Speaker 2 [27:15]

Then somebody's asking, do you use techniques like building a second brain for the Zettelkasten-Methode to manage your new knowledge?

Speaker 1 [27:28]

Oh, good. Very good call. Work in progress how to collect our knowledge. At the moment we are experimenting internally also with the chatbot and like a SharePoint solution where we can put blocks, nodes, etc. etc and then like we have a rack system which can interact with that but yeah still someone has to start so like um what i what i see a lot is people have their knowledge and their own nodes and obsidian and logsec whatever but are not really willing to share and um yeah that would be a first first step simply to to share even even if it's not super super nice and shiny so also to share something which is a work in progress

Speaker 2 [28:24]

Okay, one more I think we have time for one more question I will change the question a bit because FOMO is fear of missing out, but I know what the person means So I will change it a bit. So just in general use anxiety or fear Isn't there also anxiety or fear just coming from competitors or also because AI has potential to automate our work possibly fully in the future?

Speaker 1 [28:48]

Yes, of course.

Speaker 2 [28:50]

I was just about to say, I think the easy answer is just yes. It's not just FOMO, it's many things.

Speaker 1 [28:54]

things yeah that's that's true and I mean like this is now a super super big topic but the best thing I heard so far is a quote and he or she I don't know said that it's not AI who will replace your job but it's someone who uses AI

Speaker 2 [29:14]

I think that's a wonderful closing quote. Thank you very much, Karsten. I think that deserves a round of applause.