to kick things off this morning with the Fluentism finished webinar where I'm going to talk about thinking fast and slow, the theoretical framework and how it applies to in the age of AI content creation. I'm very excited about this webinar. I think it's very going to be a very slightly personal topic for me as well. So I hope you guys are all going to follow me through the end. It's going to take about 45 minutes. It's been recorded so of course you get the recording after the webinar as well and you can put questions in the chat as you go. I can see Alexandra is saying that you can't hear anything. Is that an Alexandra issue or is that a more general issue? Okay, all good. So as you can see, thank you for showing example, Alexandra. There's a chat at the corner where you can ask questions. I'll try to pick them up as we go but I'll also hopefully reserve some time in the end to go through any questions you might have. But without further ado, let's kick things off. And I do expect actually some audience participation today. We have a quick exercise and a couple of polls that I would like your input on as well. So first of all, those who don't know me, my name is Jaakko Soorosari. I'm a partner here at OBSIBI in Helsinki. I've worked with for almost two decades with digital strategy, content, and SEO stuff. I'm an avid AI user myself. We would say the traditional AI use. So doing a lot of my work with Claude, but also working more and more with different AI automations, building skills for customers as well and doing AI training and recently also Vibe coding with Claude. While I am definitely an AI enthusiast, I don't think you can categorize me as a purely AI evangelist. So I do see some downsides with AI as well, which we are going to be talking about today as well. So let's go on with basically the reason that I think this topic is relevant today. So you're all probably familiar with the fact that AI-created content has taken over a lot of the web space in recent years. So a year after the last chat GPT was launched, it was estimated according to some studies that a third of all new articles produced online were created by chat GPT. The percentage gradually rose, it went to about 50% holding studies and so on. Of course, these are estimates, these stats, they are based on a survey of certain space of the internet alone. So you can't really say that today that 50% of all the web content is produced by AI would not, that would be false because it's quite difficult to actually measure what the actual figure would be. But it is, I see it as quite a natural evolution, quite normal development that at least a lot of the majority of web pages today contain at least some AI-generated content because it's just much more affordable, faster, and it just makes business sense for more companies to use AI in order to create some content. But if it started with just the articles and writing text, you can see that it's basically now growing from pure text to other formats as well. So you see that 41% of workers today say that they've been receiving a work slope at work, whether it's a presentation or documentation or something that's really had no substance. You can see that Gamma, which is an AI presentation creating tool, people are using it today to create more than 400 million presentations a year. So you can see that it started with text, moving on to presentations, moving on to actual websites created with AI, and also moving on to videos created with AI. So in some study, having a look at first 500 YouTube shorts, shorts from a new account, it showed that or a fifth of them were just for quality AI content. And of course, also LinkedIn posts, we can see that there are more and more AI -generated content appearing there as well. So I think we see that AI content is still rising, the number of percentage of it, and it's taking over more spaces than just text. So it's moving on to more visual formats as well, which does create some new problems today, which we will be discussing today. So basically, the agenda will be that we'll start off with a quick test of basically what I call would you present this. Then we're going to discuss the framework a little bit. So what's the easier question that our brain is trying to answer? And why is AI passing those easy questions that we're asking? Then we're going to focus on how to get over this easy answer. So having a look at the tools and the mental processes that we need to go through in order to guarantee that if and when we're using AI to produce content, whatever the format is, that we're not just producing work slop or AI slop, that we're actually producing something we're publishing. And yeah, and then just a final conclusion at the end. But here we come to the first exciting test part. So what's going to happen is that on the next slide, you'll see an executive summary of a website audit that's made with AI. So you have 30 seconds to read through it. And I'm going to put a poll on and I hope that you can answer the poll with just one question. So the question is going to be, would you present this to a client or internal stakeholder or to an audience, whatever, whoever you're presenting it to. And there's three answer options, three answer options, yes, yes with minor edits or no. So you see the poll on on the right side of your screen. And I'm going to put the test on, show the text now and give you 30 seconds to read it and answer it. Hope everything is clear. Here we go. 30 seconds. Would you present this to a client? I see the first responses coming in and we're approaching the end of our 30 second mark. Good. Okay. Thank you for voting. I see the majority of you answers said no. And I think that you you've all been basically, those who answered no, you've probably been also primed by the topic of this webinar itself. But those who you, especially those of you who answered no, if you would quickly type in the chat, why you wouldn't present it? What's wrong with it? Maybe it's the wording. Maybe you see something funny there, but maybe throw in a couple of comments in the chat as to why you would not show this. What exactly is it that you would fix? Good. So the conclusion is too bold. Yeah, that's a good point. Too much sex, no clear message. Also true. Can be true. Language is not clear and helpful. It isn't clear enough. Doesn't communicate the value. Yeah. Option is the technical optimization will fix. Organic traffic seems incorrect. There may be other factors. A very good point. You need evidence for the numbers. I can title, outlook message not clear. Headline and jargon. Small and big topics mix. That's also true. Good point. I think we'll, very good. Thank you. This works out actually quite nicely. So this was kind of expect responses that I was expecting. And soon we're going to have a look at the same exercise a little bit later on as well. But thank you for your participation so far. It doesn't effectively capture what's wrong and what can be done with it better. True as well. But there are some of you who said that it's also it's publishable. Yes, with minor edits as well. Very generic. Yeah. Good. Good. As I said, we're going to come back to it later on and have a look at do a bit of a more deep dive into why this is what's wrong with it. But thank you for participating so far. So as I said before that this topic is somewhat feels somewhat personal to me as well because the true story on the background I like to share is that I made an audit to a client using an AI tool that we had. And as I was presenting the material way through, I realized that I could not explain all of it because I would basically add I briefly reviewed the audit itself. Oh, this looks actually looks really good. It's very well structured. It's very well placed in order. But what I hadn't done is I hadn't familiarized myself with all of the technical stuff that was in it. It was a kind of a new audit type for me. So I understood I like the majority of it, but not everything in it. And then I also realized that some of the data presented it and it was not true. Maybe that AI checker had done some errors, but I realized that some of the data was was actually blatantly incorrect in it. And I realized that in the middle of the presentation, it was quite it was super embarrassing for me, but it also got me thinking a lot like, that something that really reads well and looks looks thorough doesn't necessarily mean that it is actually true and it actually is functional. I wasn't I wasn't checking whether I could defend what the content in the presentation itself. I was just checking whether it was clear and whether it looked good. I think that this is a I learned from one presentation hopefully. And maybe that you you can say in the chat if you had similar experiences with AI as well. But I think that this is really taught me to start focusing on when creating AI content to start focusing on focusing on every detail and focusing on answering the difficult questions instead of just the easier questions. So the easy question is that or what really happened in my head and what happened during the meeting this substitution or this has really has a name and it has been an effect that's been studied studied before. And judging if you're familiar with the framework or the name of the name of the novel or book, non-fiction book where the title for the or the webinar came from you'll be familiar with that with what's the name for this effect. If you've read a book called Thinking Fast and Slow, it's a non-fiction book that was really popular about about a decade ago. It's building on work of two psychologists Daniel Kamenow and Amos Tversky of the research that they did over several decades. And it basically outlines like two different systems that are working in our head. The system one that is fast, automatic, effortless. It's basically fast decision making that helps us humans in our evolutionary phase. So if we see if we're in the jungle and we see a tiger, then we need system one thinking to basically immediately recognize that we need to be probably be running away right now. So it's something that enables really fast, simple decision making in everyday situations. And it basically it's a little bit jumping into conclusions, but it's not purely a bad thing. So it's an evolutionary tool that helps us survive. Then we have system two, which is something that takes a lot more time to think and deliberate and use our actual brains for making decisions. So this is when we're actually using our brains to look at each argument, spot what's the, is there anything missing between the arguments and review the entirety of the context of the system. So we're trying to extend the situation. But our brains are wired so that the system two only jumps in when, when, when, when, when really, really needs to. So if in our lazy brains, we're more happy to more than happy to use system one. So to basically jump into, to enable fast decision making and try to avoid using system two. And I think it's, of course, it depends a lot on the person, but I personally find that my brain is quite lazy, for example. So if I don't really concentrate, I'm more than happy to welcome using my system one to use in our decision making. And Kahneman and Tversky have outlined several of these like cognitive systems or effects that, that are errors that, that are caused by us using system one instead of system two. And one of the effects is called, called something called the substitution effect. So if we're presented with a question that's really hard in our brains, we're actually switching to, substituting to a simpler question. And we don't even notice that the, the, the swap. So the effect or the research that they, Kahneman and Tversky, they did was for example, when they were asking people, how happy are you with your life these days? And people were basically not answering that question because it is a difficult question. It requires a lot of consideration and thinking, is it even answerable can, can be, can be tricky. But basically what people were, how people were answering that question is, what was my mood right now? So if they were happy at the moment, they were answering it. Yes. Or if they were feeling good at the moment, they were happy that saying that they're happy with their life. And as an example, that's how our brains are basically working. So we are substituting the difficult question to an easy question. And I was trying to find out whether, whether this effect has already been studied in terms of when it, whether, when it comes to, comes to AI systems and how, how is AI helping us? And I actually found a very recent study from, from, from that's preprint or not, peer reviewed yet. But when they had studies at Wharton College, they had 1300 participants using AI to solve some reasoning problems. And whenever the AI assistant was giving them right answer, it really helped people be more accurate when they, with the issue. So they were 25% more, more, more, more accurate. But then when AI was giving people the wrong answer, so they had set up that AI sometimes gives the correct answer, sometimes gives the wrong advice. And whenever AI was giving wrong advice. And whenever AI was giving wrong advice, people were actually 15% less point, percentage points less accurate than people not using AI at all at the exams. So whenever, whenever we get a correct, incorrect answer from AI, we are more likely to do a mistake ourselves. But it didn't really affect the confidence of the people answering the question. So whether AI gave the correct or incorrect answer at all, confidence of the respondents went up anyway. So whenever we find a source, whether it's AI or not, it seems that our confidence in answering a question goes up, whether we are actually incorrect or not. So to avoid the substitution effect, I think when we are creating AI content and editing AI content, the actual question that we should be, a hard question that we should be answering is, is this worth publishing? Is this worth publishing? So does the argument that we're making hold? Is it true? Is the argument logical from the beginning to the end? Is there anything missing in between? Or, and is it saying something new? So are we actually producing, providing something valuable or something new, new thinking into it? And is it ours? Or are we just copying it from somewhere else or copying it from AI? And I think answering this question, is it, is this worth publishing? If you really think about it, it does take some effort to answer. So what basically what we're answering is something easier, which is usually something like, does it read well? Is it, is it fluent enough? Yeah. So basically when I was doing the audit, audit, in that meeting, I had answered the easy question, which was, does the audit look nice? Is it, is the audit, is the audit like presentable as it is? And I wasn't thinking really, is this worth showing to the client? Is this worth, is it everything here? Does the argument that I'm making here hold what's missing? Is it correct? So, regarding the, I'm already mentioned the substitution effect, where it's important to, to note that there are a couple of other easy questions that the AI really, really shines on passing, because they are trained on, on fluent, confident, tidy text that people are happy to use. And this is what basically what we are wired to check for. So our easiest question that we ask, ask ourselves, is does it read well? And this is a sort of a load or release, relieves our cognitive ease. So if something reads well, we feel that it's true. If text is fluent, confident, if there are typos, we are less critical to, to what the content actually is in it. So we are again using our system one instead of our system two, because it feels easy to us. So it takes a load off our brains, where if it's well written, doesn't necessarily, well written, however, necessarily mean that it's correct. And the second effect that are, or a second easy question that are, we are answering is, is anything missing? So this is known as the, what you see is all there, what you see is all there is effect. So when we're, when we're presented with a, shown a presentation or shown a body of text, we are basically judge, most of us are judging what's on the page. And we're not judging like what really should be in there. So we are only basically already past the, past one step because we are only reviewing what AI chose to include to us and not what should actually be in the, in the, in the example. Then the, the, you probably more, a lot of you are familiar with the anchoring effect, which is basically that the first draft of the first version space, you're setting the frame as to what the outcome would be. When we got a complete nice looking text from AI, what we end up doing is we're editing the sentences and not editing the angle of the text or the actual content that should be in there. So we are, we've been anchored to the original version, what the text is and not sort of thinking from outside the box. And I think this anchoring effect is what, Daivi, you were mentioning that you were expecting the example to be bad, because I had viewed it. You were anchored by the, basically the setting of the webinar, which is supposed to be, is something good or bad or not. And then finally, a third effect that we, we can fall victim to is the halo effect. So does something look good overall? But if you, if you have a piece of text or if you have a presentation or a video, then one part of it is really good. And this was just that one good section and if, especially if it's in the beginning or the end, then you see that you feel that the whole, whole of it is credible. So you are sort of lured in by the, by the halo effect of a, like a really good opening or a really strong, few strong bullet points, which doesn't mean that the entirety of the text is true, but we are sort of lulled into thinking that everything is, is good in the text. If we are, if we are, if there are some strong parts to it. So if these are the, the sort of the easy questions that we want to avoid, or we want to, we want to get overcome. And our next task is of course, then put back the, put back the hard questions. So focus on the, on the techniques that we can do to overcome the risk of, of releasing something that's not fully thought out. So, and here we also come back to, to the, the initial text that you guys were reading. So instead of asking like, just like, does the, does it read well? Is the content nicely written? We should be asking like, can every claim that we're making in the content be backed up? And if we're asking something like, can every claim be backed up? Then what we're basically doing is that we should be highlighting every number, every name, every factual claim that we're making in that text or presentation. And then seeing where each of those come, actually came from. So for example, here, we said that in the original text, there was a slight error that said that organic sessions fell 34% year on year. But actually the figures were, stats were comparing nine months to 12 months. So it was end of sort of end of September compared to 12 months before that, which meant that of course the results would be down because there's a lot of results. down because there's three months less of data, data behind there. This is the kind of sort of small errors that we can happen. So if you can, to help prevent this, this from happening. So making sure that every claim that we're making that can be backed up, what you should be doing is, is before the draft, you're creating a draft with the AI version. You should be asking the model to mark every source of, for every claim, every, every data point, asking for the scope, where did every number come from? Are the numbers comparable? Are the numbers comparable? And basically marking it as any, anything that it cannot be supported as need source? So that is, it's already highlighting itself. If something is, is, is cannot necessarily be proven with data. So if we're asking, not disagreeable, if we're asking, can every claim be backed up, then, well, basically we are, we are, we are able to overcome one big, overcome the majority of the factual errors could be in the text. And then we're not reviewing for fluency, but we're reviewing for evidence. Second point in our, in our text was how to overcome the issue, issue that could something be missing, or is basically the, what you see is what you get effect. So instead of asking, just like, is anything missing, which might be a difficult, tough question to ask, we should be asked, or it's makes it easier if you frame it like, what will the difference? What will the reader or the audience ask first? After, after, after, after reading this? So, this is a sort of a pre-mortem technique, that, that Gary Klein and Kardemon championed. So in order to counter this, what you see is what you get effect. What you see is all there is effect. So basically, you should be asking like, as I was asking you guys during the meeting, like what's in there and, or asking for you, how would you replace it? But what was missing from the original data, for example, or the original text that you guys saw was, what was actually the data sources and the time period. So for example, comparing nine months to two or two, or 12 months, there wasn't any indication of how much of the site was called. Was it called 100% or was it only taking maybe 20, 30 pages of the site? And there was also, someone mentioned that there's no actual business impact estimate there. So the pages that actually are driving revenue are, were not in the report. And of course, effort and cost of each fixed. So that summary of, as a loan would not get you too far, because you couldn't actually determine the suggested actions instead of how much work is it to fix everything and how much does it cost to fix anything and what would be the estimate benefit of fixing all of them. So basically, already before prompting for AI content, I think it's good to list out a few questions that the reader might be asking and any objections that the reader might be asking, might be having into the prompt before asking for it to produce them. And then moving on from there. So you're already framing it towards what's, what will they be looking for? What could they be missing instead of being sort of incurred to what the AI was deciding for you? Then next technique or third technique out of four is basically the, the countering the anchoring effect. So instead of does it just need polishing, you should be answering, would I have structured it actually this way? So if the, if you actually are, or an easy way to think about this is, if you directly have a long text in front of you, try to break it down into argument by argument. And just to like follow, try to follow the logical or the argumentation within the text by narrowing it down to as little text as possible, because we know that AI can be very, can write long stuff and not all of it necessary is always necessary. So if you're like producing it down to the, to the very core of the argument, it's much more easy to evaluate whether the argument actually holds true. So basically our initial audit could be, could be down, could be structured into four different lines. And the basic traffic is down, our pages are slow. Then we're fixing the issues with the most instances that were in the report. So you had what was 1200 something and then hoping the traffic will recover. But there is actually no, no connection between step three and four, if you paste it like that, because you're fixing the issues with the most instances and not necessarily the, but not necessarily the issues with the, with the biggest impact. So basically the, in the original initial audit, the number of old texts to be, to be fixed was, they fixed was the first on the list because it had the biggest number of instances that required fixing. So how to get rid of the anchoring effect already, or how to try to counter it already before asking for content. This is that basically you're writing the thesis before producing the draft. So you're either doing the crafting, the sort of the argumentation yourself beforehand, or then you're asking for the, for the AI to produce a skeleton of the argumentation before writing out the stuff. And I think that the second technique especially is, is really valuable because AI does a lot better work if it first focuses on just a structure and not, not the actual text itself. And it's also much more easier for us to evaluate just a skeleton than it is a full presentation or full video or full script or whatever. So once we see the skeleton of our argument in front of us, we can even more better evaluate whether this is true or not. So one more thing, explain it out loud. So if instead of judging for the countering, or how to counter the halo effect. So if something looks good overall, something looks good in pieces in one point can look good overall. So what we should be focusing on is, can I explain and can I, can I explain, can I prove every line in the text? So I think it's one technique that you can try to do is, is focus on really the, the weakest parts in the text or the parts that feel weakest to you. Read them aloud to yourself, to yourself. And if you can't explain, or if you can't understand them yourself, it's probably not, doesn't probably, doesn't belong in the text itself. And then in our, for example, in our, you, you, a lot of you noticed that our, our text was, wasn't super clear. And it had sentences like this, that the main cause is render blocking JavaScript in the critical request chain, combined by the hreflang implementation, which sounds can, can sound convincing, but also does not sound very simple to explain to anyone because there's two sentences that don't really belong together, because they have little to do with the, with each other. So basically, if you estimate whether every line, every sentence in the text is something that you can explain, and if it's not, just take it out or find a new way of saying that, because if you can't explain it, it's probably not understandable to the reader as well. Which is, unless the reader has a lot more, for example, not necessarily true for all technical substances, that can sometimes be that you struggle to understand it yourself, but the reader will, will get it more easier. But then you have to really work on a professional reader first. So someone who's a subject matter expert to analyze the content first. Good. And this is something that I ran into in the, the, the true story that I explained before. This is something that I happened to run into quite a, several times in that presentation, that there was something, a lot of stuff in the presentation that I couldn't really, without resorting to, to Googling and stuff that I couldn't explain. So, as a, as sort of a final note, or the single most important tip tip is, basically, you are writing the, trying to answer the hard question first. So, is this worth publishing? So, if you can, in a couple sentences, define what the piece, what the content that you're trying to create, what is it trying to prove, who is it trying to prove it to, and what are you, what are you hoping for them to do next? And if you have that, that written out first, then you're already on, on your way to creating something that's, that's hopefully worth, worth publishing. And then you've also created a framework of what the, what the actual content, what the, what the, what the, what are the sort of the goals that the content needs, needs to meet. And this can be, doesn't have to be long, just a couple of sentences. And this, this is the thinking that you need to do yourself before moving on to, to the actual, producing the content with AI. Okay. Then a couple of more tips on, so on, on some prompt styles that you can copy. And in order for to, to, to get them, to get a step ahead in, in avoiding these, is the, in avoiding the, in avoiding, to avoid creating something that's not really worth publishing is already mentioned that it asking for a skeleton first. So before writing something, give the argument as bullet points or bullet skeletons, main claim evidence and so on. So not jumping directly into full text, but asking for a skeleton first. Then if you have a topic in mind that you want to discuss, then it's a good thing or good exercise is to ask AI to give you three different angles for that piece. So this also helps you to ensure that you're not anchored to just one angle of the, of the, of the text or the presentation that you have three different angles, for example, to choose from, to ensure that you're picking the best one. And then to ensure that the data is correct and, and, and so on, you ask the AI to list every assumption that they made so that you can see that the logic of it follows. And you can ask for it to analyze the weakest claims in the drafts. So you can also pinpoint itself where you should be paying more attention to. So if you're, if those were on the prompt, prompt work, then when you're creating, helping it to AI to get things, things correctly, and, and making sure that you can see that the errors that it's potentially making. So ensuring that when you're giving it context, that you can prompt it to just to use the material that you have provided and making sure that it states where each claim is sourced from and making sure that if you, if you cannot find a source for each argument, then it says that this is a need source or that it had made an assumption there. Then if you, if you're asking AI to show the data, then make sure that the data underlying data is, is comparable. For example, you see, you have the same source, same time period, same scopes, for example, for producing the data. And then finally, making sure that who's going to be the, be the reader. So what were the, what are the objections of the reader to the topic? What do they know about it already? What do they believe? What are we trying to do to do to them? So we can already use that prompt that as a context so that we can try to find ways to, to, to, to counter that. What do you see is all there is effect. So on. And finally, never a bad thing that when you're creating content with AI is to, to finally ask a colleague. So a simple, slow thinking tool is that someone who's not seen what you prompted, they haven't seen the brief, they haven't seen the, seen any of the drafts. So they aren't anchored to them. They're anchored to, they aren't anchored to the setup that you've created. They're anchored to the, to the text that you're showing them. And don't just ask like, is this okay to you or do you like this? But ask them like, what's this trying to prove? How could, how would you, how would you say what's this trying to prove? And if you were reading this, what would, what would be like the question that you had on your mind after reading this? And is there something there that if you had written this, then you would not defend yourself? So, never harmful to ask for a colleague. Good. Then finally, when you should especially be going slow. So when you should be trying to use your system, system two instead of system one. And whether, when, when it's more important to go slower. So it's basically this, you can think of it as a quadrant. So if it's not super value important copy, and you know, you know, it's super well, and usually fast is fine. Doesn't really, the details matter, matter less. So if you're doing internal updates or routine posts, for example, a product copy that you've written more times, then you probably need to, you can probably move on faster. On the other hand, if it's like super high stakes, new topic that you're not familiar with yourself, then it's good to go through the entire, that we outlined or the entire world of questions techniques that we outlined before, in order to ensure that you're actually producing something that's really good. So let's say you're a freelance writer. When you onboard a new customer, you'll be wanting to do the full slow pass. Then after maybe a year or two, and when you get, you're familiarized with the topic, it becomes more, more and more low stakes. Although then, then it might be, it may make sense to, to regularly review what your, what the assumptions are that you're making in order to ensure that basically you're doing. Basically you're, you're, you're still, you're not being lulled into using too much of your system one. Good. Finally, coming into, into towards the end of the presentation. So finally, the checklist to, to go through, how to replace the easy questions with the more hard questions that you should be asking and what's the technique to, to fix them. So as a summary for you guys, instead of just asking like, does it read well? Is it, does it look good? You should be asking like, can every claim that that's in the, in the material be backed up and audit basically auditing every claim in it. Instead of just asking, is anything missing? Ask like, well, what's the reader going to ask first? Do a pre-mortem technique around what would be the, what are you going to be challenged about? Instead of asking, does it just need polishing or how do I polish this? You should be asking like, is this the way that I would have structured this and do a reverse outline. So narrow it down to send it by to every argument, skeleton of arguments. And instead of just asking like, look good overall, you should be basically asking, can I explain every line that's in it and explain it out loud to yourself. If you do, or explain it out loud to a colleague. And then finally, a just pull tactic, like instead of asking your colleague, like, if you think this looks good, ask them like, which line would you leave? Like least want to defend in order to get their, their brains going on and, and focus for them to use their system to when they're reviewing the materials. So, and, but, and probably most importantly, when you're creating something with AI, start by answering the question. So is this worth publishing? And it can also be, of course, be, be summarized into, after answering, is this worth publishing? Like, so what should happen after this? Sort of a short version of it is to be replaced. So, that's, that's it for, that's it for, for the webinar so far, but I'm happy to take questions or comments in the chat if you have any. You can find my, find my contact details there as well. If you have anything later, you'll note that the recording, recording has, will be sent to you after the webinar. And actually the slides that I just showed, you can also download them. I think directly there should be handouts, and that's that you can share. So, no worries there. And thank you for feedback, Maxim. And I'll give it a few more minutes to answer any questions. And before that, I can show you how many of you were familiar with, with Thinking Fast and Slow. So, a third of you had, had, have read it. And even more of you had read some of it. I think it's a very, it's not the easiest book to read, I have to admit. It's a, it's a, as a nonfiction book, it's, if, in the text, it's quite academic, though it is a very interesting, interesting topic. If, if I can see, thank you for the, for all the praise in the chat. I'm happy to hear it, hear, hear that. And hopefully this was as beneficial to you guys as, as thinking about this topic was, was to me personally. So, I wish you all a great end of the, end of the week. And hopefully you can start asking some more difficult questions when, when working with, with AI in the future. Thanks a lot guys. Have a great rest of the week.