LLM Updates and Sports Illustrated's Authorial Escapades

Speaker 1: 0:00

OK, let's go, let's do it Hit it music playing].

Speaker 2: 0:16

Hello, hello, welcome to Data Topics. Unplugged. A casual, light-hearted weekly question mark? We'll see about that. Something like that Short discussion on what's new in data, from Ella Lims again to Sports Illustrated, everything goes. Today is the. What's the date? The 8th of December. Today's date, 8th of December of 2023. And my name's Morello and I'm joined by the one and only Bart, bart. Well, however, how are you doing last week? Have we missed you? Well, last week.

Speaker 1: 0:52

Last week I was a bit sick.

Speaker 2: 0:54

Yeah.

Speaker 1: 0:55

Just two weeks ago, I think.

Speaker 2: 0:56

Yeah, two weeks ago Under the weather. I know You're dearly missed for sure. I heard stories that you did not look good, so it was good First for the best that you had.

Speaker 1: 1:06

Just for the best, I wasn't there, it wasn't, it wasn't pretty.

Speaker 2: 1:10

That's the word on the street, so we're good to have you back.

Speaker 1: 1:13

Thank you, happy to be here.

Speaker 2: 1:16

All right For people listening. I think we're doing a little experiment, right.

Speaker 1: 1:22

Is it the first time that we're recording? Well, video recording, exactly.

Speaker 2: 1:26

So in the future, I think we hope to have these things livestream on YouTube. But all the lies or truths that I've set up on my height, I think now they will be put to test. I think they will be visible. You'll put a face to the name, so let's see how we keep that up. Now I have to start properly showering every day, cutting my hair, all the works.

Speaker 1: 1:46

Yeah, the good thing is we already have a Blanda, so there's that.

Speaker 2: 1:49

That is true, and we have a very, very high end setup as well, with the camera and whatnot. We have the couch. We have the couch. Yes, we do. We do All right, and maybe anything else that we plan to change in the future.

Speaker 1: 2:06

Well, we've been discussing a bit, so it's been a bit somewhat weekly, sometimes monthly, sometimes bi-weekly, exactly, and we're thinking a bit about going weekly as of more or less 2024.

Speaker 2: 2:22

Yeah, I think that's true. I think that's true. I think we're getting our vibe now. I think we can kind of do this weekly Like new year, new me kind of thing, right Like 2024.

Speaker 1: 2:34

Nice, that's it we have a what's the resolutions?

Speaker 2: 2:37

the new year resolutions, you know.

Speaker 1: 2:39

Never been too good with resolutions, though.

Speaker 2: 2:41

Me neither, but you know.

Speaker 1: 2:44

Not sure if we should link this to that.

Speaker 2: 2:47

In any case. In any case, we have a lot of topics today. Yes, A lot of data topics.

Speaker 1: 2:53

A lot of data topics. Ok, this is too late, oh, ok.

Speaker 2: 3:01

I really need to practice this. You've been out of the game.

Speaker 1: 3:03

The difficult thing is like there's nothing on the buttons.

Speaker 2: 3:06

So I need to have like an icon on the button. An icon, I mean what you prefer an icon than a text.

Speaker 1: 3:13

I'm a visual guy.

Speaker 2: 3:14

OK, ok, thought you'd been to the typography part of it. Ok, we'll get to it. There's more typography coming your way. So tech corner.

Speaker 2: 3:22

First thing I actually wanted to bring up this week Google introduced Gemini, our largest and most capable AI model. Did you hear about that part? I heard about it, yeah. So basically, according to Google, it beats chat GPT-4 in a lot of benchmarks. I think we already discussed previously that how good is a GPT model? It's a bit subjective, right? Like how do you measure this objectively? But according to Google, they beat.

Speaker 2: 3:53

And there's some reactions from some jokes as well, memes and whatnot that open AI. People are concerned, but also there has been a lot of divine on the demo, because the demo that they showed is like a really cool video. So we'll put everything on the show notes, of course, but the video really looks like I don't know. The guy started doing rock paper scissors or a girl Now we don't know, it's just the hands and then it shows the Gemini supposedly says oh, you're playing rock paper scissors and he's like, wow. But then when they actually go behind the curtains they're saying oh, what am I Like? The prompt is what am I doing? Hint, this is a game and they have three pictures of rock, so it's very different from what it was Like. They have a two paper, really simple stick drawing of cars and one apparently is more aerodynamic than the other.

Speaker 2: 4:42

And on the video they just put both. And then Gemini supposedly says oh, the one of the right is more aerodynamic. But in the behind the scenes they also prompted like oh, which one is more aerodynamic?

Speaker 2: 4:54

So, it's like it kind of takes a bit of the magic away and a lot of people are saying that the demo was faked and even made me think that Google was really. I mean, if you maybe ask, last year you said who's the company of AI, and then a lot of people probably say Google with deep mind and whatnot.

Speaker 1: 5:10

You're Navago.

Speaker 2: 5:11

Exactly so I think. Now, if you probably ask anyone, they're going to say open AI, so also Google had the barred.

Speaker 1: 5:18

This one comes to LLMs right.

Speaker 2: 5:20

Yeah, but I think a lot of people when they think of AI, they think of LLMs today. True, right, but the whole, even the barred fiasco, I think also is really and is Gemini.

Speaker 1: 5:33

Is it public?

Speaker 2: 5:33

yet Did you try it? I didn't try yet. I saw some friends that they were going to try it, but in Europe it wasn't.

Speaker 1: 5:40

Oh.

Speaker 2: 5:40

Europe is not accessible. Yeah, it was not acceptable. They also have like three models, as I understand Like there's the large one, there is also a like I think that's the pro, and then there's also one that is supposed to be more accessible for devices. I also skimmed through some of the details, and they also said that the way they trained was also revolutionary, because I also heard stories that chatjpt is actually very inefficient in terms of training. Yeah, and they also addressed that on the Google side, they actually use the TPUs, which is the Google GPU kind of thing. So they also mentioned this. But, yeah, and also I mentioned the barred fiasco.

Speaker 2: 6:15

I think for people that are not aware, it's like Google released a barred which was a competitor of chatjpt, but even on the opening demo they had a wrong fact there. I think it was like who was?

Speaker 1: 6:28

during a live demo.

Speaker 2: 6:29

Yeah, like it was like the marketing thing, you know, like she wasn't even a live demo but like they were announcing it. And then the demo, like the recorded demo for the marketing was like what was the first telescope to capture Minimich Rambat and it was actually a European thing, but they linked the wrong information and that was on the thing there and everyone like we're very quickly to flag this Painful yeah, and I was like have that one.

Speaker 1: 6:59

Exactly.

Speaker 2: 7:00

So yeah, I don't know. I'm a think the verdict is not out yet, right, I think let's see how Gemini compares to.

Speaker 1: 7:09

Yeah, I think we should try it. I think, like you mentioned, like to me, the benchmarks are always difficult to translate to actual user experience, like we've seen a few times that there were benchmarks that were better than GPT-4 or 4.5. I've never really experienced it as a user.

Speaker 2: 7:32

Yeah, it's very yeah.

Speaker 1: 7:36

And yeah, and let's hope we can try it soon In Europe. We're a bit like the poor nephew in the world. Actually, also today, I tried Meta's new image generation model. I think it's called Meta, I imagine. Did the whole sign up? Yeah, finally signed up. I need to create a Meta account. Didn't have a Meta account, finally signed up. You can't use this in Europe.

Speaker 2: 8:05

Really tricked me to sign up. Got you, got him. If you have a Facebook account, this account is a Meta account.

Speaker 1: 8:12

No, no, really, yeah, you log in with your Facebook account or Instagram account, and then you need to create a Meta account. Oh, that's why, Probably give access to even more data then yeah.

Speaker 2: 8:26

It's like you're key to sign your life rights. Ok, cool, I think. Maybe a cheeky question as well. I think AI, gigpt and whatnot I think everyone's very convinced of its potential. Do you think that the process of making AI models do you think people has figured that out already, or not?

Speaker 1: 8:53

Making AI models, building them? Yes, you're asking the question, like you're expecting it to know.

Speaker 2: 9:04

It's a cheeky question.

Speaker 1: 9:07

ChagGPT is there. Right, Chem and I is there.

Speaker 2: 9:11

But also you could say well, chaggpt it was said that the way that they build the model is very inefficient and also OpenAI, google they have resources that 99% of people do not have, True.

Speaker 1: 9:30

Yeah. If you say it's making it accessible for them, yeah. Audience at large. I don't think so. Yeah, and that's indeed the resources that are needed and the input data that's needed. And maybe resources could be less if you have more performance.

Speaker 2: 9:51

Yeah, actually the ChagGPT, the OpenAI team. I think they've been releasing things very quickly, right, and I think there have been some articles. I need to catch up on that. But the reason I asked is because I also read another article which is a bit Turns out. It was like an ad for a tool kind of, but the title is Navigating the Chaos why you Don't Need Another Amelops Tool.

Speaker 1: 10:15

OK.

Speaker 2: 10:17

So what really caught my eye on that article was a quote that says AI is at an interesting point in time. On the one hand, the world as a whole is convinced of its potential and its applications are already influenced by billions of lives. On the other, the way models are developed and deployed in the industry is often far from ideal. So I think they're tying a bit back to the Amelops story, right, which is something that I still feel like it's a hot topic-ish, right Amelops, which I thought was an interesting way to kind of summarize things. I feel like, on one hand, there are teams that are still figuring out how we can make things fast and iterate. In the article they also do an analogy with Git right. Like if you say, like 20 years ago people were building software, but like they were maybe sending zip files or something, and today you'd be like man, this is so you know.

Speaker 1: 11:08

Yeah, interesting point. I think we're still very much in the transition of looking at all of this as an experimentation, transitioning into like this needs to run in production systems and, of course, the major tech giants like an OpenAI, like Google, like they have this figured out, but I don't think it's super straightforward, like there are still a lot of options. There's no one best practice. Yeah, people say you need to do Amelops. What is Amelops? Also very quickly in the evolving field, and I think the major difference from, let's say, more traditional software engineering is also that you need to think about a lot of things in a more probabilistic manner, like you develop something and you need to test it.

Speaker 2: 11:56

Yeah.

Speaker 1: 11:57

You need to set up this workflow to deploy stuff, but the whole testing, robustness evaluation etc. It's much less deterministic than typical software engineering.

Speaker 2: 12:08

True, true, true. Yeah, it was a let me think a bit like of this whole moment we're living right Like, indeed yeah.

Speaker 1: 12:20

And I think, even if we look at it from a more simpler point of view, where we go to a more managed ecosystem, like, for example, with OpenAI, where you simply use the API to interact with a model, like you still are deploying something that you need to test in a very probabilistic manner. Yeah indeed. Even then, if you can make this statement like it just becomes a software engineering. But the way that you test it is still very much different, right True.

Speaker 2: 12:48

Yeah, I mean, sometimes you want the like. I think, chadgppt, if you put the same prompt twice, you're not going to get twice the same answer, right?

Speaker 1: 12:56

Mm-hmm.

Speaker 2: 12:57

Cool. So there's like it is very, how do you go? I mean, it's the strength but also the weakness, not the weakness, but it's like complexity increases because of it. Right, I agree? And staying a bit on the food for thought corner as we talk about chadgppt real weak is chadgppt weak.

Speaker 1: 13:19

Yeah, it's like it goes so quickly that there is so little else to be able to speak of these days. Yeah right, it's a strict. I'm not sure how I feel about it. Yeah, me neither. It's difficult to balance.

Speaker 2: 13:35

Because I feel like if you don't talk about it, then you definitely falling behind.

Speaker 1: 13:39

Yeah, but I just feel like I just, I'm just but I'm also more and more using it myself day to day in almost well, very quickly getting into different types of tasks that I do much more than a year ago, yeah, so I'm also excited about it and if there is something new is out there and the way the multimodal approach it's just chadgppt now like it's works.

Speaker 2: 14:09

So nicely all together.

Speaker 1: 14:10

And it's also the whole ecosystem around it to integrate something like we want to expose chadgppt assistant via Zapier in our Slack channel. Yeah, it becomes so simple to do these things, which I think is super cool.

Speaker 2: 14:26

Yeah, indeed, I think it's for me. Someone that contributes to building these things, like these systems. It's for me. It's interesting because indeed, like as a user, I get very excited, but as a developer it doesn't excite me as much.

Speaker 1: 14:44

In what sense?

Speaker 2: 14:45

In the sense that, like to me the work that needs to be done to deploy something like that, I feel like there's a well-worn path.

Speaker 1: 14:59

Like it's, not like it's work.

Speaker 2: 15:00

You need to do it, but it's not like you have a problem, it's like man, how am I going to solve this, you know? So for me as a developer it's not the most. I mean, of course, I think the application side of the impact motivates me as well, but it's like in terms of problem solving, I don't find it as as enticing.

Speaker 1: 15:23

Yep, I understand like it's a different, where you would, two years ago, whipped up something in PyTorch.

Speaker 2: 15:34

Yeah, you know you had like this. You had more like open questions like, oh, is this like that, this and that? But now it's like, yeah, you plug this here, you have the authentication there, so wrong users don't have access to the wrong data. And then you have this database there, you have this here, you make this API call. So I mean, it's very personal, right, but I had this duality within you know, it's like one hand, I think it's really cool that it can do this, but on the other hand, it's like I don't want to be the one to do this. Yeah, and ChatchPT? We're still learning things about ChatchPT. I think apparently, chatchpt performs better on Julia than Python in our for large language model code generation. Why is that, bart? See how sweet that was.

Speaker 1: 16:19

That's really sweet. That's a blog post I came about a few weeks ago by Christopher Raquelkas.

Speaker 2: 16:33

Do you think you put your name in that last name?

Speaker 1: 16:35

It could very well be, and he made a comparison across 10 programming languages. He used ChatchPT 3.5. Okay, so I'm not sure how much all of this would hold in the current, so 2023. Yeah Well, that's true. It shows how quickly this goes.

Speaker 2: 16:55

Actually like maybe a quick side note. I Google today ChatchPT was first released in November 30, 2022. So it's been like one year. It's crazy.

Speaker 1: 17:06

It's crazy, right, it's really crazy. Okay, and the findings? So he had a bunch of tests, a bunch of assignments like, for example, please do a print statement of Hello World. Like so very simple to more complex stuff. Give these assignments to ChatchPT 3.5 and say now, built this code, julia built this code in C++, built this code in Python, etc. Etc. And the conclusion is is that, julia, it's apparently easier for ChatchPT to generate correct Julia code than it is for ChatchPT to generate correct Python code.

Speaker 1: 17:54

And the Christopher hypothesizes a bit on the why. I think that is an interesting one. Like I think we typically see Python as like a language in where we want to do it, the Pythonic way, which more or less is let's keep it to kiss, very simple, easy to understand, etc. But maybe it's not always the case. Like he gives some examples, he gives a few, but one that comes to mind is like do you want to create a 2D array? Basically you need to define, like what are the dimensions? Like see the bit as the how many rows, how many columns?

Speaker 1: 18:40

In Julia it's like I don't know the exact naming, just check it out. I think it's if you want to do an array of zeros, then the command is zeros, you open parenthesis, 2,2, you close parenthesis. Sounds logical In Python. You use numby, probably for this NP dot. Zeros, you open parenthesis, you open another parenthesis, 2.2, and then you close to parenthesis. So it expects as an argument a tuple, which I don't think I've ever thought about like. This is weird. But if you need to generate this code, like if you try to take that vantage point, I need to generate something that I'm not used to Like. The Julia one is much more easy to reason about. Like why would it have to be a tuple? The Julia is it like, built in? I?

Speaker 2: 19:33

think it's built in. Yeah, yeah, because that's the other thing. Yeah, true, like you're actually writing numpy code.

Speaker 1: 19:40

Yeah also, and they give a few of these examples where, like the statements, basically in order syntax in Julia from look easier on paper than on Python, then the Python alternatives, and I use it a little bit as a to hypothesize on this, and I see he's using my lab here as well.

Speaker 2: 20:04

Comparison also to my lab a bit, yeah, yeah.

Speaker 1: 20:07

Interesting. I think it's an easier. An interesting one, because we typically see Python as, like the very intuitive language.

Speaker 2: 20:14

Yeah, I also. For me it's also a bit surprising because Python is the language of ML, of machine learning, so I would expect that the people that built it, they may have a motivation to make sure that the LLM performs on Python. Also, maybe there's probably more training data in Python because, it's true, it's very popular language.

Speaker 1: 20:39

But again, like it's 3.5, right, yeah. Yeah, I think what you see now in the latest version is that, especially for Python, like it generates the codes, that also tries to execute it.

Speaker 2: 20:48

Yeah, indeed, so you have this feedback loop immediately.

Speaker 1: 20:51

Yeah, it doesn't work, let's let you try to adjust it.

Speaker 2: 20:54

Can you just do that with Python? Can you do this with other languages?

Speaker 1: 20:57

It's a good question. I only try to with Python. I think it's the code interpreter only supports Python, but I'm not sure.

Speaker 2: 21:02

But it's actually super powerful.

Speaker 1: 21:03

Yeah, and actually what it's used for as well, and I think that so it used to be not that long ago that mathematics was horrible in Cheshire PD. When you ask it to calculate something now it actually generates Python code.

Speaker 2: 21:16

Ah, that's clever.

Speaker 1: 21:18

So it's much less of the hallucination. Yeah, it's clear Python code and I've actually since then. I've never had it give a wrong answer, Really yeah.

Speaker 2: 21:26

That's really cool. That's very clever as well.

Speaker 1: 21:29

Very clever, isn't it?

Speaker 2: 21:30

Yeah, indeed, and you always have the feedback right there, like if you write some code that doesn't run, yeah. Or I think I think I even saw an example that they said, like write this function, write 10 tests that pass, and then it would create a test and one test will fail and they will kind of reason why and is like goes back and forth is like Super cool. Do you ever feel scared for your job, art? Or you feel like a 10x developer? I feel like that's one of the two you have to know. It's like it's not normal. You're there, I think, for me, for my mortal ladder yeah.

Speaker 1: 22:01

I can become super efficient yeah okay. That is.

Speaker 2: 22:04

I've also a bit found my workflow Nice, I mean like yeah, but I think also that's a good point Like there's I mean there's co-pilot, that is right there. There's chat, gpt, there's also other. I mean there's even a. There's an AI for that website. So there's an AI for a lot of stuff, and I think finding your groove is a. I think it's takes time. Actually, I think Simon will listen.

Speaker 2: 22:26

I think I heard him on a podcast saying that. I think he really jumped onto these things very early, right, and I think I know I feel like I understand better, like in a more personal sense, because you're saying, if you're not going to use this, you're going to fall behind, you're just going to try. He was also saying like be very methodical about it. You know, like it's not just use it, but say, okay, I'm going to use it for this, and then almost like write down what is it good for? You know, like if you want to make a blog post, it's good if you put the bullet points that you wanted to say, right, or instead of just saying like, create something. You know. So it's like be very methodical. And I think, yeah, you can definitely be more.

Speaker 1: 23:02

Yeah, I typically use myself, if I start a coding project from scratch, that I say in chat GPT like create this set of functions for me, that this is the functionality I expect it will like generate like a first attempt at it, and then I work on that in VS code with the co pilot. Yeah, it's a bit the workflow, yeah.

Speaker 2: 23:28

Yeah, because I think you get like the bulk generation. Yeah, exactly, they just have the little in between. That's pretty cool. That's pretty cool. And what do you coach when you're bots? Maybe X bots?

Speaker 1: 23:44

X bots.

Speaker 2: 23:44

That was a difficult segue. I do my best. It doesn't always work, but you know what about it? Apparently, there was a new about X, x, even saying X sounds weird. I feel like now, if someone is listening to this, you know they don't even know. Yeah, like if you just say, oh, you built a bot on X, I think people are gonna take a lot realize I'm talking about Twitter.

Speaker 1: 24:09

Yeah, and X is so hard to Google. Yeah, like, if, like, if this issue with X, like you never get anything that's related to how to do this on X.

Speaker 1: 24:19

Yeah, yeah, just a piece of news I saw floating by, I think, yesterday. Somewhere yesterday evening X released their own chat, gpt ish version, basically a chatbot which is called grog, which apparently is considered by their users and I woke, not sure what to make of that. There was some rumor going around that Elon was tired of the politically correct chatbots everywhere. It's anti woke, it's very right leaning, apparently a bit racist. So yeah, there you have it. Not sure what, how long this whole Twitter X, whatever saga is going to continue, but I'm a bit tired of it.

Speaker 2: 25:23

Yeah, me too.

Speaker 1: 25:26

My main issue is because I've tried to go away from it a lot of times to blue sky, to Macedon Everybody is still there, right? So everybody also keeps their account. And then people try to go away, but they keep their account and I say, oh yeah, but everybody's still there, so let's go back. I like it's a vicious circle, right. Like I don't know what needs to happen for people to actually leave the platform.

Speaker 2: 25:50

Well, yeah, I think, the Macedon versus Twitter X, I think.

Speaker 1: 25:55

I think Macedon is too difficult. Macedon is really for a very technical audience. Yeah, I think so.

Speaker 2: 26:00

I think so, but even like some things, like when I went to Macedon, first thing that I noticed is it's very chronological.

Speaker 1: 26:08

Very chronological. Yeah, true.

Speaker 2: 26:10

Which is like most of the times that people are like oh, there's no recommendation, so it's very biased, there's no algorithm really manipulating or leaning this and that, but I realized that I like the recommendation. I agree I think it's a bit I don't want. Of course, like I'm just being difficult, I feel right because I don't want the recommendations to alienate me and I want to have a clear depiction of reality, like what is actually talking about, what is actually you know because in X today that's also the opposite.

Speaker 1: 26:41

Like I get notifications on my phone of X of people I don't follow at all, like extreme right leaning Dutch politicians, I don't understand why I get notifications. Actually, I actually disabled them now. But that's the. That's the alternative. Like that is yeah it's like pushing a certain message.

Speaker 2: 27:01

Yeah, and I think that was even in Facebook. No, maybe last year that they there was a whistle blower, that was a Facebook or meta research. They found out that, like, the way that the algorithm works is really to push these very controversial things are very, you know, extremist Because that's what sells. But then there was a lot of impact on people's well being and Facebook or meta was aware of that but they didn't want to do anything.

Speaker 1: 27:27

That was like there was a whole big thing about it, right, but I think it's like uh yeah, extremist Funny thing on Macedon I see there and as a bubble I've never experienced like, but people on Macedon, for some reason, I frequently come across people that are still very much in the in their mindset in the COVID pandemic. Like, oh yeah, but I went there and no one was wearing masks and I was trying to stay healthy. And I like, yeah, okay, but like do you know what year it is? I've only encountered this massive one, but like, not once, Like every day, like I see someone and where are they from actually, Do you know?

Speaker 2: 28:04

I should check. I should check Because actually I was. I heard the PyCon US. Is they're actually going to require masks?

Speaker 1: 28:11

Okay, okay.

Speaker 2: 28:13

But it's a really Okay. My PyCon US is going to be next year in May and I think they're going to require masks. I don't know if it's going to be everywhere.

Speaker 1: 28:20

I'm actually also on the foster dom. Yeah so you're very much an attack audience.

Speaker 2: 28:25

Yeah, probably US. So that's always wondering because is it still a thing there?

Speaker 1: 28:30

in the US.

Speaker 2: 28:32

No, I don't. I mean, I don't know right, but I only heard that. I never heard that COVID is going up in the US or anything like this. I just heard that PyCon US, they're gonna have, they're gonna require masks. Okay, At least in some parts. I don't know if they changed to put interesting yeah, yeah, yeah, yeah, yeah, it's another. I'm trying to say too much there, but I think it's been a bit in the past, right Like it's true.

Speaker 2: 29:00

So, all right, something a bit lighter maybe. Yeah, so came across this. I thought it was a interesting article Six tiny wording tweaks to level up your communication as a software engineer Wow.

Speaker 1: 29:18

Yeah, I think I'm wondering what they're gonna know, but I think it's interesting.

Speaker 2: 29:24

I think as time passes I kind of see how the hard part of software is people. It's not the software. Yeah, you know, like, and I think the way I'm trying to emphasize more and more to myself the way you communicate things, okay, right, even in silly examples, like you submit a poll request, how pedantic should it be? You know like, oh, you didn't put a period at the end of the doc string. Should I bother you with this? Right, I know there are some tools that you can kind of automate, which I think helps, but, like, I think the way you communicate with people makes a big difference on the success or not of your project.

Speaker 1: 30:04

Okay.

Speaker 2: 30:06

So I was. I mean, I try to pay attention to these things myself, but I think whenever something like this comes up, I also take a look and I want to hear your opinions on this. So first one for example, use would you be open to, instead of can you, when you want seamless to see, when you want to seem less commanding, but still need to lead that to a yes. So I guess it's like saying I, if I want you to do something, I would say hey. Instead of saying hey, bart, can you do this, I'll say would you be open to doing this? And his reasoning, I think, in the article, is because people, they like to think of themselves as open minded. So if I say, would you be open to people more inclined, required to say, inclined to say yes, okay, yeah, is this something you can relate to at all?

Speaker 1: 30:55

I can definitely relate to it. I think it also, I think you. There's not necessarily a problem in saying can you? But then if you want to be sure that it doesn't come across, across this command, and you like, you need to know the person, like the person needs to know you and like you, can't. You don't have this risk with you.

Speaker 2: 31:11

Be open to indeed, I think it's more soft, like it's not like a very direct confrontation thing, I think it also relates a lot, if you like, the color insights right, like not put their correctness. Second one is adding because to any request, which I think this also just makes sense. I think he, he, there was an experiment from psychologist and then she was saying at the library say hey, excuse me, I have to make a copy of five pages. Can I use the machine because I'm in a rush? And just adding this because I'm in a rush, people way more willing to let, which I guess. To me this is just common sense.

Speaker 2: 31:46

I think is, even if you go like I don't know, fast for help and you say I tried this, this and this, people are way more willing to help you if you show you've done your homework, if you're sure that you're have a reasoning behind. You thought this through, you know you're not just delegating work. So I think this kind of links somehow.

Speaker 1: 32:03

So it's logical to make it too extreme. Yeah, when you go to the McDrive, on the order some fries, because Would you be open to yeah?

Speaker 2: 32:16

Third one saying use can we instead of can you. I think this actually a good one. I think it's like we're team right. So can we make this better? Can we improve this? Can we make sure the code is clean there? I think this is a good one. I think kind of puts frames the message correctly in the sense that we're all we're trying to do this right. I'm not, you know, like it's not your fault. Is that? Can we do better?

Speaker 1: 32:37

But what if you say, let's say you want me to do something, can we eat a banana?

Speaker 2: 32:44

and I say, yeah, we can just start on one end of the other.

Speaker 1: 32:51

Yeah, we can, but it's not clear to me that you're asking me to get that right.

Speaker 2: 32:56

But I feel like there's a commonality that I think the less direct you are, like the wreckness also brings clarity.

Speaker 1: 33:04

Yeah, exactly, yeah right. I think there's a bit of a balance you need to strike.

Speaker 2: 33:11

But for example I think the example they give is can we rename this to match the pattern of the other file? So if you're naming variables, you can say can we rename this?

Speaker 1: 33:18

Okay, yeah. I think if it's depending on the Right, like to me it comes across like let's have a discussion on an interesting topic but actually mean like yeah, no, you need to do it.

Speaker 2: 33:29

Yeah, I think you're also. You're a very direct person. So I think it's like if, if someone is not direct to you, they're trying to like, you're gonna be like what are we doing? Here, Fourth is adding a. What do you think at the end of your suggestion? Actually, I do that quite a lot, but I think it's also. I think it's also a way for me to if this goes wrong is not just me right like you're, so agree with it.

Speaker 2: 33:58

You're building your safe, just like it was not just my idea or like you finished, you make this big statement and then you go oh, but that's just my opinion, you know, it's like it's classic, you know.

Speaker 1: 34:11

But I think that that makes a lot of sense. Like you want to Get a bit of alignment within a group, like do we all agree or not?

Speaker 2: 34:20

and I think, sometimes Voice in this explicitly that I'm like, I'm positioning myself like it's not my decision. I want us to agree. What do you think?

Speaker 1: 34:29

you're open for people explicitly and I think, for people that Are less direct or more trouble voicing their opinion like it's a, it's an opening that they can use. I agree, I agree.

Speaker 2: 34:42

Number five Quote quotes it seems like for stalemates in conversation, so what they're saying, there is a parent, apparently there's a technique called labeling, but the idea is, whenever you feel like the conversation stops, you just State something like it seems like I need to describe your point of view so, for example, so it seems like it would be impossible to make a change due to the way we do this. So like we're discussing something, conversation kind of gets stuck and then you just label it and Maybe that goes against someone's perspective and it's like oh no, actually it's not impossible, you can do this in this, in this. So I think the I don't do this necessarily for stalemates, but I want to always I think it's helpful to me Say I'm going to try to repeat what you said with my words, to make sure I understand. And I think sometimes it also is good to always frame it in my way, because if I think it's a bad idea, for example, I can say Let me make sure I follow you. You're trying that we should, you know, run with scissors in our hands, you know, like whatever, like yes, it's a bad idea, right? So I think it's like it's a way to make sure that I understand before criticize, but at the same time, if I see potential downsides or disagree with it somehow You're kind of hinting it already. Yeah, you say, and also I think sometimes I had a lot of discussions that in the end we didn't get to a consensus because the definitions we had were wrong.

Speaker 2: 36:20

Right, so I think sometimes we stating with your own words, I think it helps make sure we're on the same page Makes sense Changing the order of. But so, for example, this is number six, by the way. I see what you're saying, but in my experience the other way has worked best. And if you flip it, in my experience the other way has worked best. But I see what you're saying and I think his point and it's something that actually I learned I had a class about. I remember hearing something like this in class in Brazil when I was in high school. Okay, long time ago. They like, if you add the sentence that comes after, gains more emphasis, right, so just flipping the order of things can make a difference on the message perceived.

Speaker 1: 37:01

And that's interesting. So one I don't think I'm.

Speaker 2: 37:05

I think it's hard to do it in conversations. I think if you're writing an email or something that you can read. I think it's easier to plan it through. But what do you think? There is anything you agree or disagree with? No, I think they're nice tips and tricks, and do you have any tips or tricks you add to these?

Speaker 1: 37:27

I feel like I should have thought about this a bit more.

Speaker 2: 37:31

You can just say no.

Speaker 1: 37:34

I think communication wise We've had a few examples here like a stalemate. I think if you're working on a project, you need to make a big decision. I think you need to try to get everybody on board before the moment is there to make a decision in group.

Speaker 2: 37:56

True.

Speaker 1: 37:58

Because if you have to make a big decision and there are a lot of different viewpoints, it's very hard to get to a single vision. Or you have to force a vision and then people are unhappy.

Speaker 2: 38:13

Yeah, true, I mean a lot of times if you force a vision, exactly.

Speaker 1: 38:21

You can say we're going to do that, but then people don't share the vision.

Speaker 2: 38:25

Yeah, but I think sometimes, if you build a tool, this is going to solve all our problems and everyone's going to use it, but we cannot force people to use this stuff, right? So I think that's also something else. But yeah, I thought it was interesting.

Speaker 1: 38:38

True.

Speaker 2: 38:40

Now moving more to Miscellaneous things. Sq White Quite new but still under development.

Speaker 1: 38:54

Still under active development.

Speaker 2: 38:56

What's the latest?

Speaker 1: 38:57

The latest is that they have a JSON B type. Which is basically JSON in binary format. Bit more efficient to store.

Speaker 2: 39:08

Like if you read it for you as a user. It's really just a storage layer.

Speaker 1: 39:16

So before if you would want to store JSON data, you would store it in a text field, basically Like a string and address JSON B JSON B is cool. I think JSON B is cool, like it gives you a bit of flexibility, but still B in a SQL setting.

Speaker 2: 39:40

But JSON B is like it gives, like it understands JSON as well.

Speaker 1: 39:45

It's JSON. As an end user you can see this JSON but, like you, have this flexibility, typically in no SQL databases which is more or less a JSON-ish structure, often Like the entry is a JSON-ish structure. I very much tend to stay away from no SQL databases, very much prefer SQL. But JSON B gives you a bit of wiggle room in that.

Speaker 2: 40:12

Yeah, but I love the databases support. That know, like I think Postgres has a JSON thing.

Speaker 1: 40:16

I love databases. Snowflake has a variant or something.

Speaker 2: 40:20

I think a lot of them have it.

Speaker 1: 40:22

And what they also did is that they made the migration part super easy.

Speaker 2: 40:28

SQLite.

Speaker 1: 40:29

Yeah, but like super easy, like if you already were storing JSON in a text field, you can just change the type of the field and it will automatically be transferred to JSON B. That's really cool. It's really cool Like super minimal effort. And I just wish because the announcement, I think, was yesterday just wish it was done one week earlier. What did you do this last week, bart? So I maintained this very minimal schedule thingy.

Speaker 2: 41:01

Check it out.

Speaker 1: 41:03

We put in the show notes and we actually moved to SQLite storage. Okay, before it was, logs were stored on the files and now we're using SQLite. We won't go into the argumentation. I think we have good argumentation, but the logs are actually in JSON format and I was thinking ah, penny, that's not like there's not a JSON B, because then we can actually query on. So it's stored in the text Text field is still need to migrate.

Speaker 2: 41:36

But you said migration is easy.

Speaker 1: 41:37

Migration is easy. That's really cool, that's really cool.

Speaker 2: 41:41

Yeah, you know. So I heard like there's SQL, then there was the no SQL stuff and now it's kind of like mixed right, because you still have the JSON B type for SQLite but Postgres has their own thing. No flakes, you can have some unstructured within your structured stuff.

Speaker 1: 42:01

Yeah, and I think you need to. You should not abuse it Like you need to have very good arguments for it. I think that is an important one, but could, for example, an argument be is let's say you have this table where you land raw data from a third party application and the schema is not 100% stable. It's a third party API that you adjust to form in JSON format. You're going to land it in a JSON B column and then you're going to extract that in a data phase, like these type of things.

Speaker 2: 42:33

Yeah, I did something for a toy project thingy. It was streaming data from Reddit to snowflake and I used like the manage Kafka conflict, yeah, and yeah, like it landed, like it's just JSON. It was two columns, but it was just JSON. But with snowflake was super easy. Okay, just take this key, this key. Yeah, I definitely see the benefit.

Speaker 1: 42:56

Actually haven't really checked out the if they already have a query format for it. So the, the querying and JSON B and Postgres, for example, super nice, oh yeah.

Speaker 2: 43:05

And also I was hearing that they had that, and I think now with the vector databases is a similar story. Like now vector databases are super high because of the whole AI thing and now we have just vector databases like quadrant and whatnot, but there are already some like I think there's like a vector plugin for Postgres.

Speaker 1: 43:25

Yeah, it's a big community.

Speaker 2: 43:26

And I think they're saying like yeah, like every time they try to do something like super groundbreaking, there's like completely different, and then it just kind of goes back to the same. You know, it's like it just takes some time.

Speaker 1: 43:36

But I think 10 years from now, everything burns down. The only things in terms of databases remaining are Postgres and SQLite.

Speaker 2: 43:46

Yeah, actually, yeah, sqlite is everywhere too.

Speaker 1: 43:51

No, they beat all the evolution. So all the peaks and the valley is like they're still alive, right, they're still growing, so it's amazing.

Speaker 2: 43:58

Still there. Indeed, indeed, indeed. All right, what else do you have on the MISC, the miscellaneous corner? I see here wizard designs. I'm not sure what to make with this, yeah.

Speaker 1: 44:10

I actually came across a blog post and I'm not going to go into into much detail by Julia Evans on hit branches and it's I think it's a very good example of how you, how you explain difficult concepts. I think, from that point of roots, are interesting to read blog post and what she explains to them there. Like you, have an intuition about a hit branch, but it's actually for a lot of these things, not how it actually works and that's sometimes causes the challenges in your hit workflow. That your typical thing. I don't know how to solve this animal. Let's Google this and place in like five hits commands to fix it, Like these type of things. Copy paste stuff, right.

Speaker 2: 44:58

Yeah.

Speaker 1: 44:59

And an example that she mentioned there is like a branch, like we have this idea that we have, like our commits are like a timeline and suddenly we branch off that timeline, but a branch is actually a full history. So, at that point you have like two timelines.

Speaker 1: 45:14

Yeah, I see what you're saying so in like these, these things she challenges a bit in a very visual way in her blog post. I think it's very interesting. We're going to link the show notes. And then I would started looking a bit further on her website for Julia Evans and she is the I think it's called, of wizard science, which is as very cool cheat sheets on a lot of different topics Like, for example, how integers and float works, bite sized command line and very nicely drawn, very comedic, a bit tongue in cheek, a bit geeky. Oh nice, it's really cool. Really liked it. I'll check it out. I'm going to share the link I actually show. She also made a few posters, so the post I showed you we're going to print before we start recording from.

Speaker 2: 46:02

There.

Speaker 1: 46:04

There was no way to purchase it Really. She just shared the PDFs on the website and I'm like but I want to pay for this. Yeah right, this is cool. So we're going to print them out. Maybe we can actually put them here. Yeah, so if you're not on the live stream yet, so shout out to Julia.

Speaker 2: 46:21

Evans, yes, and be know that we're trying to pay you, but you don't want to take our money.

Speaker 1: 46:26

Okay.

Speaker 2: 46:29

All right, all right, all right Um.

Speaker 1: 46:37

Sports Illustrated yeah, Sports Illustrated, I think, got a bit of a flak. They had a few articles, people from very, I think, random-ish-named authors. People did a bit of digging and they couldn't find any LinkedIn or any profile on these journalists, which is a bit weird. At some point, I think they found the portrait picture on some stock photo website and apparently it was. That was. Well, that is a bit what the community thinks is fully AI-generated stuff.

Speaker 2: 47:18

Really, can you prove that?

Speaker 1: 47:20

Well, no, I don't think so. But Sports Illustrating is saying it was just the third party that placed this and we gave it a fake name. But it's real, but they did remove it. So I mean, there must be something fishy.

Speaker 2: 47:34

I'm not doing anything wrong. Where is it? Oh, I don't know it just left.

Speaker 1: 47:39

So yeah, but I'm actually To me it's an interesting case and it's good that this got flak, but I'm, in this day and age, like it starts to get a bit fuzzy, like what is authentic and what is correct and not fischly generated and what is just fake and generated. Like this is a thin line. We're threading.

Speaker 2: 48:04

And would you be comfortable if there is AI-assisted content? So, for example, someone used JetGPT. They created like a block of text and then they went and they edited that. Is this something that, for you, is like that's fine, that's just a new reality, or do you think this is a bit cheating? When you read an article from someone with their name on it, you expect the words to be poosed by them.

Speaker 1: 48:28

Well, I do it myself as well, so it would be a bit.

Speaker 2: 48:31

I was gonna say nothing, but you know.

Speaker 1: 48:33

It would be a bit If I would challenge that. It would be a bit weird, but I don't know. I don't know, to be honest, what about my stance on this. So I do sometimes, but it's very much hobby, very now, and then some short stories, and there, when I start to write something it's typical of fiction I tend to stay away from it because it feels like, yeah, why am I doing it then?

Speaker 2: 49:07

Yeah, but I think you said it's a hobby, right it's?

Speaker 1: 49:09

a hobby.

Speaker 2: 49:10

It feels like a profession that you need to be productive.

Speaker 1: 49:12

Yeah, yeah, yeah, yeah, then I would probably see it as a tool. If I would be a journalist, I would probably see it as a tool.

Speaker 2: 49:19

I think I saw it. I didn't link it. I need to probably look into it, but there are tools. Apparently people are creating tools to help people. So I think you copy-paste GPT code and then you keep track of how many things you changed and how many things actually changed.

Speaker 1: 49:32

Yeah, that was IA Writer yeah, it's like a minimal distraction free writer and they now have an option to copy to paste in text and that you can basically text like ah, this was generated so that, as a writer, you still know like what did I write and what was actually.

Speaker 2: 49:47

Yeah, which is an interesting it feels more honest to me somehow. It's like I mean, if you're using this, it's like I think, yeah, you're using it as a tool. You're not trying to take credit. Maybe we need to also.

Speaker 1: 49:58

And, but also as a writer, I think, if you want to assert a certain tone of voice, if you visually see okay, this was generated, this was my own, so the generated one I need to be, pay a bit more attention. If you see it as a tool, like it also helps to know which parts were generated by this tool, which maybe I need to double check for facts and this episode.

Speaker 2: 50:18

Okay, maybe I take another turn here. What about co-pilot? Do you think we should worry about that If it's a AI generated code? Because I guess you could also argue that code has a bit of the flavor of the author right, like because I feel like for code we never think of that right. It's like the ChatchaPT is there, it works fine, whatever, that's true like the tone of voice. But it's like if you read a piece of code that was completely AI generated but someone had their commit named there.

Speaker 1: 50:48

You know, Interesting one? Yeah, I am, but tone of voice. You notice very much when I say summarize this for LinkedIn and if I don't add anything about tone of voice, it will do like, I think, very American style, we are super excited. Exclamation mark. This is amazing Exclamation mark. Here are five emojis, Exclamation mark. And then I need to say, okay, this summer Tone down a bit, keep it a bit level-headed, keep it a bit serious in Dutch. And then the next version is already closer to what I would like, right.

Speaker 2: 51:20

But indeed, like I'm just thinking of stereotypes now you know, I'm Dutch and it's like, yeah, that's exactly what I wanted.

Speaker 1: 51:30

But the thing is that in natural language you immediately notice it. Yeah.

Speaker 2: 51:36

Well, yeah, so there's the clear does it work or not? Yeah Right, it's a very clear test Does it pass or not pass, right?

Speaker 1: 51:42

But also some best spread, like, let's say, take type hinting in Python, for example. True, this type of things yeah.

Speaker 2: 51:55

Yeah, but you know what I'm saying. Like sometimes you re-code and you're like you can tell this person Like there is a.

Speaker 1: 52:02

Yeah, so why I'm just thinking about it now? Like I said before, it's sometimes these bulk generations in Chet-GPT, yep, and then you get these in Python like these functions. Within the function, the actual content is like one line, and I'm gonna If it would have been me, I wouldn't have created a separate function for this, probably.

Speaker 2: 52:20

Exactly yeah.

Speaker 1: 52:21

Like these type of things, like where you see like there is a difference then versus how you would have done it, maybe functionality, why is the same? Yeah?

Speaker 2: 52:28

But I think if you're like for more flexible like I don't know if you're on Python and using Pundas and there are like three different ways of doing the same thing, you know like you can chain everything and you can tell oh, this guy probably was I don't know Dorian or you or whoever, because I know that someone that cares about this and then you have the classic Jupyter-like code. You know that you just like you do some stuff, you change, you have these huge functions. You know like there is a style to it as well.

Speaker 1: 52:52

That's true, yeah.

Speaker 2: 52:53

Right, I think there are some languages or for us, that sometimes it nudges you more to one way of doing things than the other. And there's like Python, I think, is fairly flexible and I think like if you go for Pundas it's very flexible. And I think there, because there are different ways of doing things, you do get a lot of very different code styles, which makes a difference in debugging, makes a difference in this and that right.

Speaker 1: 53:15

True. So the difference is, of course, with natural languages that, like, functionality-wise, the outcome is the same. True, right, true. Well, with natural language the outcome is not the same. Like your interpretation of a message changes. Yes, definitely Because of the tone of voice.

Speaker 2: 53:31

Yeah, I agree. I think it's like I can write tests and if all my tests pass, I don't ever need to look at it. I think the only time when it matters is when you're debugging right when you actually have to read it when you need to maintain it, when you need to work on it. But I'm also wondering, right like because we're saying now that if AI is a tool for writers and we say no, that's cheating. But if you say that Copilot is a tool for developers, but that's not cheating, Well is it?

Speaker 1: 53:54

cheating? That's a question, a cool question, I think from depends on the person's point of view.

Speaker 2: 53:59

Yeah, that also depends.

Speaker 1: 54:00

Like for me when I had you as a hobby. Like then it feels like cheating From the moment it's your job. Maybe it's not.

Speaker 2: 54:08

Yeah, because I think the difference there is that at the job, you need to be productive, right, like they're paying you to be productive, right. I think it's the fact that the productivity aspect makes a big difference.

Speaker 1: 54:18

Yeah, true, true.

Speaker 2: 54:20

It's just some more food for thought.

Speaker 1: 54:23

you know there's a lot of food for thought.

Speaker 2: 54:25

Yeah right, that's what happens with just you and me. We just have a lot of size all day. We don't have content, but it's just cool and I put this one down for you. I thought it was going to have a special place in your heart.

Speaker 1: 54:39

I've seen it as well. You're talking about MonoSpace.

Speaker 2: 54:43

Yes, I am.

Speaker 1: 54:45

It has a tagline and it knows the super family fonts for code. Wow, it's from a set of fonts from YouTube. I think they felt a bit I think they were later than Vercel right, With the guys font, guys Sans and guys Mono, I think so and they were thinking, oh, how Vercel is doing this. Like, oh, wow, the community is so happy with this, we also need a set of fonts, and they yeah, it's nice that they do it.

Speaker 2: 55:22

But what is the premise? What does it mean to be a super set? No, super family.

Speaker 1: 55:31

That's a good question. I don't know what their definition of it is. I think what they try to do is that they made so they have five fonts, a lot of different styles. So they have Sans fonts, a Serif font, they have something that looks like handwriting, something that looks a bit like mechanical, a human is font. So different styles, five different styles, and all of them are very. You see, they have a lot of modern fonts. They're very variable in a sense that they have a lot of weights. So the weight of a font, like, is it bold, is it black, is it extra black?

Speaker 2: 56:07

These things.

Speaker 1: 56:08

So it's very. You can do a lot of things with one font. Basically, they don't excite me.

Speaker 2: 56:15

No, no, no, no. I brought this just to make you happy. I thought you were going to be off the road.

Speaker 1: 56:21

But I was happy about the news. But then when I look at the fonts and I can't really explain why it doesn't excite me.

Speaker 2: 56:29

Another mysterious of Bart. Yeah, but I saw also one thing they proposed actually I read it somewhere that what if within your code you have different fonts for different things? Yeah, that's true. Yeah, yeah, what do you think of that? Do you have more handwritten stuff for comments, doc strings or more official or whatever? Does that really thing that?

Speaker 1: 56:51

I don't know. I've never had the feeling, never thought about this is something that I need in my life. And when I look at their example, I still don't think this is something that I need in my life. What to me, as that functionality very much is like syntax highlighting yeah.

Speaker 1: 57:10

Like to me, that is enough. Well, syntax highlighting makes a big difference. Right To me it's also like when it comes to typography and fonts. Like less is more, Like you don't want five different fonts in one string. You want like clean, A little bit clean and sleek, and sometimes be a bit geeky when you're coding, but like not five fonts at the same time Too much. Here's a comment Get a live. Here's an if statement. That's that font. There's a function declaration. That's the data font.

Speaker 2: 57:39

What if they had like, the copilot code is a different font? That could be interesting. That could be interesting. That's a good point, because there are, like the syntax highlighting stops because that's an interesting point.

Speaker 1: 57:50

That's there. I follow you.

Speaker 2: 57:51

Because it's like it's really different tones of the. You know like you wrote this copilot yeah.

Speaker 1: 57:56

And also because, like there, you reach a limits of syntax, highlighting because, like it just, it's also code Right, like the only difference is like what does the source? That's an interesting yeah.

Speaker 2: 58:08

But then the issue is like if you save that, if you save a file and open it, you also need a way, then you need metadata to keep track. Another one Another one Like DJ Kelly.

Speaker 1: 58:23

I was thinking about Queen, oh, okay.

Speaker 2: 58:26

See, that's another one.

Speaker 1: 58:28

Another one bites the dust.

Speaker 2: 58:29

Yeah, true. No, you just age yourself part, you know, just dated yourself Sorry. Sorry, another one. Okay, moving on quickly. Gpt coach. This is a new one, this unique state of this one in this wasn't here when we started.

Speaker 1: 58:47

You were asking me and then you switched the topic. You asked me I would worry working on them last week and what I was actually. What's up? No, it's actually not last week's. Two weeks ago, I made no need to set the price. Next week weekend, next week, with daydruids, we are doing a trail run in the forest here, close by middle of the trail. What I did is I made it a small GPT coach. Okay, that is coaching me every day by saying this is the workout you need to do.

Speaker 1: 59:24

No okay and I'm extremely instructed when it comes to working out. Really, I'm like this is like zero structure. It's like, okay, let's I feel fresh, let's do an interval Okay. But not much more thought than that.

Speaker 2: 59:38

But actually I think it's good. I need more structure. I mean, I don't like probably when we should say we're counting, probably running right.

Speaker 1: 59:44

Yeah, I mean, I mean running.

Speaker 2: 59:46

For me, working out is really just like going to the gym and I think I need a lot of structure. But just because if I don't have structure, I just don't do it.

Speaker 1: 59:53

And that is what GPT coach is now bringing me. So I made like a very small script to do this. I have a prompt that says I'm a runner. I feel a little bit this bit of experience. This is the type of times that I run, this is I have so many days until this trail run and this role and bit of characteristics about the trail run. I inject the workouts that were previously generated by this and then I say generate a new entry, keep this in mind, a bit of a, this style of the type of workouts. I also say, like these days I have a rest day, so keep that in mind, and then adjust the workouts based on how close I come to the, to the, to the actual event. And that's what I've been doing for the last two weeks.

Speaker 2: 1:00:37

And it works well. Do you like the workouts? Does it make sense? I've been working out much harder than I used to do Is that also motivational.

Speaker 1: 1:00:46

It's like get out there, tiger yeah but it's good to get to have something on paper that says, yeah, but you need to do this. And what I'm actually doing is it's entering because I'm wearing, like an aura ring. So what is also a part of the prompt is that it's a very minimal Python script. So I just do input on the terminal and it asks me what is your sleep score from aura?

Speaker 2: 1:01:06

What is your readiness score from?

Speaker 1: 1:01:07

aura and it injects that in the prompt as well. So keep consider these, these readiness and sleep scores from aura.

Speaker 2: 1:01:13

But do you need to add manually, edit or it's also automated?

Speaker 1: 1:01:16

Every morning I run the script and then I manually edit. So that's really cool and it's maybe good to have a recap of that after the event to see what it will do, true.

Speaker 2: 1:01:30

Oh, that's really cool.

Speaker 1: 1:01:32

Are you going?

Speaker 2: 1:01:33

to open source that as well.

Speaker 1: 1:01:37

Well, yeah, it's small enough to open source. When I do something for fun, I'm always a bit like I'm at pooh, but then the code needs to look a bit clearer than this.

Speaker 2: 1:01:48

Just put a bit of these claimers Judge Pt's code. Like don't judge me, but it's like it's because it's a use open AI.

Speaker 1: 1:01:54

Yeah, it's just an API, like the prompt is much larger than the actual code to do this. Yeah, so it's interesting to see how these things evolve.

Speaker 2: 1:02:05

Yeah, that's for the whole prompt engineering thing. Yeah, ok, true, that's actually very cool.

Speaker 1: 1:02:11

Some things that are also just something that I find fun. Like I went for a long run, something like I don't know 18 kilometers, and I forgot to eat before and I was like really like a low blood chocker. I was really feeling bad, like something like 15 kilometers, and I went through it and I felt bad and I was thinking to myself, shit, I need to better do this. I had this in the past as well, but what I did? So I also have a field post-workout feeling. I added to this field.

Speaker 2: 1:02:43

OK.

Speaker 1: 1:02:44

It went OK. But after 15, I had a bit of a low blood chocker, et cetera, et cetera. Please remind me next time when I have a long run. Oh really so in the old workouts and post-workout fields get injected every time into the prompt. Next time a week later I had a long run. It reminded me.

Speaker 2: 1:03:02

Wow, that's cool, that's super cool.

Speaker 1: 1:03:04

And normally if I would have done this before before, just GPT, before LLMs. This is very complex Because I actually tried this a number of times to do this, but it is extremely difficult to say. I did a workout today, an interval training of 8 times 600 meters and I felt like this and it was at that speed. And to encode that in such a way that it makes sense, yeah, yeah, yeah.

Speaker 1: 1:03:34

That checks all the like if this, this, this, then that and that you're heuristic or you're optimization mode and also can understand like this is potentially the impact it can have on the day after. Like it's a very difficult problem to encode if you need to do that in a deterministic manner. Yeah, I tried it a few times and it always gave it up at some point. But, like, LLMs allow you to do this, so naturally.

Speaker 2: 1:03:58

Yeah right, you just do it there. It's like you're talking to a person almost. Yeah, exactly Exactly. You just say, oh yeah this happened, just like a person that is reliable. Huh, doesn't forget. Well, you hope it's reliable, right? Yeah, yeah, it's another discussion for another day. That's pretty cool. That's pretty cool. Let's see how I'm going to ask you this again after next weekend. You said right.

Speaker 1: 1:04:20

Next weekend, yeah.

Speaker 2: 1:04:20

Not this one. No, we're on Friday. I just hope you don't become a dead patient. See what I do there.

Speaker 1: 1:04:29

Well, yeah, I hope that, if you're my doctor, that you're not blocked by the IT security when you need to do an intervention on me.

Speaker 2: 1:04:43

Yeah, we're trying, we're trying. Yeah, we get better Sometimes we go, sometimes we don't, you know, it's just like a.

Speaker 1: 1:04:48

I came across a paper and let's not go too much in depth, but it's an interesting premise, it's a bit like. So. The title is do you want my password, password or that patient? And we all agree that within a health care setting, for example, data security is super important, IT security in general, super important. But from the moment that you have a patient a polytroma patient that just was in an accident, gets rushed in, maybe wasn't announced, gets rushed in bleeding to death on a table, and then you need to log in in four or five screens to get to understand who this patient actually is.

Speaker 2: 1:05:39

To factors indication. Right yeah.

Speaker 1: 1:05:42

I think that is a bit like on the moment that time can literally mean life or death, yeah, or at least have a big impact on someone's life. How important is information security then? I think that is a bit the discussion.

Speaker 2: 1:06:02

I think it's all about yeah, it's when you put your perspective right. Yeah, it's like every woman's privacy, yeah, but if you're going to lose an arm to gain that privacy, then you say maybe I'll let you. Yeah, right, I think we're very relative beings, right? It's like we're worried about. We're worried about what we're going to eat for dinner no, but then there is war zones going on right now. The people are not worried about that. No, right, it's always putting perspectives in, it's true, and what they propose a solution, let's say, to this biometric or something, or it was more like frame-up.

Speaker 1: 1:06:38

No, I think it's more the discussion around the topic. I think that within a clinical setting, we're and I'm speaking then maybe mainly Belgium. I have a bit of a professional background also. I started my professional career in a university hospital. You don't have to go very far back, like I think, even today, if you go to small hospitals in Belgium, there's a lot of things still being done on paper, which means that you can focus on the patient and just write down what you want. Afterwards you take the paper, you take the pen and you write it down. I see.

Speaker 2: 1:07:24

So then you just write it down on paper first and then you put it on the system, or not even. No, it's actually on paper, like just paper, like a paper file. Paper is just like.

Speaker 1: 1:07:32

Literally that still happens today, like luckily no longer the standard, but I think it was definitely the standard 20 years ago and I think in terms of patient contact that's maybe a plus, because now you have to lock so much things in a system that maybe even in front of a patient, that you sort of remember 50% of your task is just logging it into the system. But you also have this thing like what in an emergency setting, like how much is the? How much health important is cybersecurity, when you're in a situation where this might impact someone's life?

Speaker 2: 1:08:09

Yeah, that's true, but then also for so to be a bit of devil's advocate here. If you're in an emergency situation 20, 30 years ago someone is bleeding they are going to be like running with a piece of paper. Interest right, like the critical situation is, the paper story is not there.

Speaker 1: 1:08:28

But today, like all of the medical devices that you need, to do whatever right, yeah, yeah that's true. Like you need to lock them into the system, you need to authenticate.

Speaker 2: 1:08:36

You need to like, you need to do a lot of things. You need to take this, then you need to put it there to show that you're authorized to take this, exactly, exactly, exactly.

Speaker 1: 1:08:43

You can't just take it out of the cupboard like, yeah, I see what you're saying. I see what you're saying and I mean it's an interesting thought experiment.

Speaker 2: 1:08:48

No, I agree, I agree, I think yeah.

Speaker 1: 1:08:53

I think if it were me and I was about to lose my arm or my password was published publicly on the internet, I'd go for the password publicly.

Speaker 2: 1:09:06

I probably would do the same Me too, probably. That's yeah, yeah, that's true, that's true. I think it's all a matter of perspective, right? And I think maybe the people that design these systems, they're not thinking of these scenarios, right? So I think it's also true, yeah true For the doneness All right.

Speaker 2: 1:09:26

So I think these are all, if you're still with us, thanks for staying with us. I guess it's a long one. It was a long one, it was fun, I thought it was fun, but the fun doesn't stop yet we still have the quote or no quote. Hit it All right. So last time, actually, the winner was Paolo. That was not there, so he won again.

Speaker 1: 1:09:50

I'm president. So two times ago Paolo won the challenge.

Speaker 2: 1:09:54

Yes, Well, he won the well, I was the one putting the quotes and he got me, he got you.

Speaker 1: 1:10:01

So it was on Paolo then to make another one.

Speaker 2: 1:10:05

And then Paolo made the quotes and for the first time ever the imposter won. So I've asked him to come up with more quotes and he kindly did that. I haven't read the answer yet, so I'll just read. So the famous person is Diogenes I don't know how to say in English actually. But I need to say in Portuguese Diogenes, it's like the Greek philosopher.

Speaker 1: 1:10:30

Yeah, no, ala-mahar.

Speaker 2: 1:10:38

So this are the quotes. So first one is the only true wisdom is in knowing that the finest couch is in the grass under the open sky. That is the first one. Second quote in a rich man's house there is no place to spit but his face. And the last one is seeking truth in a society of liars is like searching for a needle not in a haystack but in a haystack of needles. So last time there was a bit of a confusion. He actually gave me two real ones and one fake Now last time. So this time I made sure he said there are two fake and one real. I'll let you. You were sick, you were off, you were out of the game for a while. I'll let you go.

Speaker 1: 1:11:29

First part I think the couch is real.

Speaker 2: 1:11:32

You think the couch one is real? You see, I don't think so, because he's like a Greek. This couch has even existed back then.

Speaker 1: 1:11:41

Maybe it was. He called it a bench and then it got translated over time to a couch. I think it's made up. Were there no couches in the Greek era? I don't know, but if I had to guess right Would have been an uncomfortable time to be alive. Yes, so that's yeah, a lot of time it's like, and also I think, the whole my reasoning was like this is too random to be, like this must have been chosen because it looks so much as something generated.

Speaker 2: 1:12:15

But then let's say but you guessed the true quote right, there are two fake and one real.

Speaker 1: 1:12:20

Well, yeah, by saying the real one. So the real one is the first one. Well, you're now challenging me, yeah no, no, no no coaches.

Speaker 2: 1:12:27

I just I, you're locked in. There is no, I'm the, I'm the host.

Speaker 1: 1:12:30

Okay, okay, okay, there's no going back. I just want to tell you probably, probably wrong.

Speaker 2: 1:12:36

I'll be wrong, but we'll see. We'll see. You can always go Okay. So I think the first one is not true because of the couch, and the last one is like the whole thing of like searching and even haste, which feels something very modern Right, like the whole A haystack. I think that was there in the ancient Greek, but do you think they said it in Greece? It's like finding a needle in a haystack.

Speaker 1: 1:12:57

Well, maybe, well, I I don't assume needles were were a thing, right.

Speaker 2: 1:13:01

Yeah, but to me it's very. I mean, I think it's something I don't know, but if needles were a thing. I don't feel like coaches.

Speaker 1: 1:13:06

Couches wouldn't be far off right.

Speaker 2: 1:13:10

That is true, you got it. Don't comment this.

Speaker 1: 1:13:13

Like you need needles for the.

Speaker 2: 1:13:15

Yeah, if you have needles, you have fabric. Yeah, you can have couches, right. I mean, they never have fabric right, Like they would wear the little curtains.

Speaker 1: 1:13:23

And if you say it's not possible in that era, then it needs to be the other one. But yeah, you chose, you chose.

Speaker 2: 1:13:29

No, no, no, no, no no, I said that the needle in a haystack is not correct because I think it's a very modern saying. So, the one I think is actually true is in a rich man's house there is no place to spit by his face. It all sounds very random, but it sounds a bit like a very aggressive, like it was the buzz of the guy. So aggressive. There were Greeks, there were conquering places and stuff right.

Speaker 1: 1:13:50

But what's with the spitting in a rich guy's face, Like why would you immediately go to spit? Like the rich guy comes in and says ah, Marilo, let's have a coffee. Sounds a bit like we're missing a step here.

Speaker 2: 1:14:05

Yeah, that's true. Maybe, I don't know. People are really good at this one. You got to give. Take our heads.

Speaker 1: 1:14:13

But so you're saying the rich ones, the spit you challenge me?

Speaker 2: 1:14:17

Yeah, I thought that's the one I chose, but you're challenging me. But if I need to abide by my own rules, right as an honest man. So he also sent me an email with the actual, actual answer. So let's see, let's see, let's see. Ah, ah, what is it? The correct one is in a rich man's house, there's no place to spit, but it's face Woohoo.

Speaker 1: 1:14:47

I'll do an applause for you, thanks.

Speaker 2: 1:14:53

Thank you bro.

Speaker 1: 1:14:54

Congrats, congrats.

Speaker 2: 1:14:55

Thanks. That means I guess next time I'm in charge of the, I'm the imposter, next time You're the imposter, all right. Looking forward to it. Is that it? Is that a show? It is a show. Thanks, it was fun. Thanks, I think so too. Well, I think for us, right, we should be the last ones, but I thought it was fun. Hope everyone else had fun, hope next. I think so. This time we're not live streaming. Next time we'll give it a whirl.

Speaker 1: 1:15:21

We'll be very anxious though.

Speaker 2: 1:15:23

Yeah, right, I'm going to get a haircut. Yeah, I'm going to shave Scarra. You know yeah.

Speaker 1: 1:15:29

Just like we're going to have a whole Need a more decent like, not a sweater like shirt right, yeah, but more of the like, really like you know, with the stuffing, so it looks like you're buffed. I do need a second akin. Then I think, right, I think when you live stream like one, I think it's not enough.

Speaker 2: 1:15:43

I think they're not enough in the world. That's true, I think, the more yeah, and then we can also decorate.

Speaker 1: 1:15:48

I think if I would have a clone machine like it, would be on my shortlist. Yeah, for sure, for sure.

Speaker 2: 1:15:55

Yeah, me too. Well, no, my partner, my partner would be the first one Nice one Nice save, nice save. Thanks, pretty good at it, all right y'all. Thank you, ciao, ciao, questioned Notwithstanding, that's you Left? Yes, right, yes.