[{"content":"It has happened.\nYet again.\nYet again am I struck by the curse of the engineer.\nI have seen some existing product and though\n\u0026ldquo;It would be fun to understand and design one of those for myself!\u0026rdquo;\nor even worse:\n\u0026ldquo;I understand how that thing is working, how hard can it be to design one of my own?\u0026rdquo;\nThis time it is a small planing hull boat, which could be used for fishing.\nAs a Naval Architect, this regularly happens to me when I watch a beautiful small craft powerboat, sailing yacht and everything in between. I regularly go down the route of obsessing about such a design, but never really getting anywhere because life and my normal day job takes all the available time I have. As someone with above normal interest in anything remotely technical it also happens with most other engineering subjects (Actually it can be almost anything physical. I have literally thought myself how to build leather shoes using age old shoe making techniques).\nI like to call this \u0026ldquo;The Curse of the Engineer\u0026rdquo;; the wish and want to understand or design something, but never finding the time to do it. You end up with all these unfulfilled wishes, thoughts and ideas for what you would like to achieve, only to find yourself 10 years later with nothing more than some scribbled notes on a piece of paper. If you are structured in your approach that is.\nSo what can you do about it, shy from inventing a time-stop watch?\nIt took me far longer than I care to admit to\nRealise I am not getting anywhere I need a system that allows me perform micro work whenever I have both time and motivation to think about something. By some kind of miracle, I came across the note taking software Obsidian. With it I can have a set of notes written in markdown, that are synchronised between my phone and my 2-3 computers running both windows and Linux. Obsidian allows me to tag and connect different notes, in theory making it the perfect application for trying to organise various different subjects.\nObsidian comes with far more features than I have yet to put into good use, such as canvases to map out ideas, \u0026ldquo;bases\u0026rdquo; which can treat notes kinda like a database of information, embedding webpages and videos etc. It has a large community, which writes plug-ins to extend the base functionality. Some of the community plugins allow me to write math equations using LaTeX, while others makes it simple to create a bibilography to cite.\nI then discovered the concept of Zettlekasten, which essentially boils down to small notes which may be linked together directly or via tags/category groupings. This allows me to quickly jot down notes or thoughts I have on a subject so that I do not forget the idea, and I can revisit it later when I may have some time.\nMy way of using this is to allow me to make some notes on a topic, and then next time I have an interest in the same topic, I can continue where I left off the last time I had time to explore the subject.\nDoes this mean that I am now always remembering to jot down notes in my archive? Definitely not\nI do however on average structure way more my thoughts and ideas in a step by step manner, slowly building a repository of knowledge to be revisited later. Some things are forgotten, to be rediscovered by a future me, but generally I am better off.\nAs an added benefit, and a use case I had not thought about in advance, is recipes. I realised that many times I want to re-make something that I found a recipe for online, but I struggled to find it again. You then spend minutes to hours trying to find it on the ever growing part of the internet of recipes.\nAs such I have now started to jot down my own copy of recipes I like, gradually building my offline library of recipes (and my adaptations to existing ones), making it infinitely easier to re-make a dish that the family enjoyed, without first spending an hour googling for that perfect recipe.\nThe draft of this blog post was written in obsidian, and since it is just using markdown, I can just copy the file over to my source code repo for the homepage, commit, and republish. It hardly could be easier.\n","date":"20 June 2026","externalUrl":null,"permalink":"/posts/curse/","section":"Posts","summary":"","title":"Curse of the Engineer","type":"posts"},{"content":"","date":"20 June 2026","externalUrl":null,"permalink":"/posts/","section":"Posts","summary":"","title":"Posts","type":"posts"},{"content":"I have been meaning to redo the test I did about half a year ago on changing particular image elements for some time now, in order to assess the progress since then.\nI headed to Mistral, the french AI company, and asked it more or less the same question as the last time.\nCan you generate an image of a man, wearing posh cloths walking on a mountain Even Here\u0026rsquo;s your image of a man in posh clothes walking in the mountains:\nMistral Please change only his haircolor to grey. Everything else shall be exactly the same Even Here\u0026rsquo;s your image of a man in posh clothes walking in the mountains:\nWould you like any further refinements?\nMistral Perhaps unsurprisingly, the exact same thing happened with Mistral as with Gemini back in June last year; the image theme is the same, but the image is not. I am not sure wheter Mistral uses their own image generation, or some other provider of the image generating models.\nNot giving up, I went to ChatGPT 5.2 Thinking and did the same\nCan you generate an image of a man, wearing posh cloths walking on a mountain Even Here\u0026rsquo;s your image of a man in posh clothes walking in the mountains:\nChatGPT Please change only his haircolor to grey. Everything else shall be exactly the same Even ChatGPT Great, now please regenerate the exact same image Even Here\u0026rsquo;s the exact same image again (unchanged):\nChatGPT That must qualify as more than acceptable. I notice some of the shadows in the background mountains are very slightly different, as well as the light on his bag is also slightly different, but all in all it is the same.\nWhen I asked it to regenerate the exact same image, I did however not directly get the same image, I rather got a matplolib frame that appeared to show the last image. (I took a snippet of the chat to show the response, as downloading the image did not give the same visual impression). During the thinking I could see it flashing some ideas about difficulty in regenerating exact pixels, and wheter it could just repost the same image. Apparently it was able to run some python code inline and display it.\n","date":"19 February 2026","externalUrl":null,"permalink":"/posts/ai-image-revisited/","section":"Posts","summary":"","title":"AI Image Generation Revisited","type":"posts"},{"content":"It is so easy getting caught up in tooling.\nWith tooling I mean any tool used to create something. For many software engineers thinking about having a personal web page, that can be the framework for creating content, hosting it on a server and or server management.\nFor people with an adept for coding, and a simultaneous wish to create and design physical things, that might be a parametric designer, CAD software and or rendering engine.\nFor a hobbyist woodworker it might be some machine that simplifies or enables the creation of something.\nNo matter what the tool, I have found (repeatedly) that it is too easy getting caught up in the process of establishing a \u0026ldquo;tooling chain\u0026rdquo;, rather than actually doing what the tooling should assist with in the first place; create someting.\nTaking this webpage as an example; For many years, I thought about having a webpage, but I could never get off the ground, as I wanted to use the opportunity to learn some new coding skills and create a proper framework, where content could be created and managed. Needless to say, that takes time, which I never seemed to find in a busy everyday with multiple kids and responsibilities. (Fun fact; in my previous version of this site, I actually hosted a semi-static webpage using Fortran!)\nLast year I decided to say F***-It, this can be done simpler. In principle, a webpage can just be a collection of hand generated html pages. Not really wanting to hand write html and css, I found Hugo, which is a static site generator. This essentially mean that it will take text I write in markdown, and create the needed html files, applying a template and using css features from this. These files I can host on whatever solution that will allow me to host some static content. I use git to keep track of changes to the files, and store it in a private github repo for backup. My hosting solution then pulls the git changes, and rebuilds the static content using hugo on the webserver.\nTo write a post I simply create a new markdown file, which I edit in nvim, commit the changes and push to my remote repo. I do not think a semi-content managed webpage can be done much simpler than that.\nWe then get to the point of this post; Creating content is hard. Over the years I have probably spent 10 times as much time just thinking about, reading about how to do a CMS system, than I have spent creating the few posts on this page. I have litterally no barriers for creation, yet over the last 8 months or so, I have only written a handful of posts. My expectation (and partial experience) is that it will be so with most other things when one is to create \u0026ldquo;en end product\u0026rdquo;.\nWhy is it so?\nMy simple guess is that it is normally easier to make a mental plan when the end point is know. The human brain appears to me to be astonishingly good at breaking a complex task into manageable chunks of work. If you want to have a small web page with blog and content management features, the brain will relatively quickly be able to think of all the things that needs doing for such a solution to come to fruition. For some of us, it will then immediately continue on how to optimize the already identified steps, and without you noticing it, the end goal has shifted from having a web page, to create the tools needed so I can have a web page.\nWhen the end goal is not know, we struggle to break it into manageable chunks. It is close to impossible to break down the steps needed to write something meaningful, without knowing what that meaninful something is.\nThere you have it; my theory on why it is so hard to create something new, unique or original, and why it is so easy to get hung up on the tools that shall help to create it . Without a clear end goal, the brain cannot break the work down into manageable chunks, so nothing gets done. Alternatively, the brain will just slightly change the goal so it has something to work towards, and that often ends up to be create a great tooling chain.\nIn some ways this is just another example of Resistance, From Steven Pressfield\u0026rsquo;s excellent book The War of Art\n","date":"24 January 2026","externalUrl":null,"permalink":"/posts/writing/","section":"Posts","summary":"","title":"Tooling is Easy - Doing is Hard","type":"posts"},{"content":"When I grew up I never had the luxury of owning a gaming console. My parents did not like the influence a gaming console can have on kids, so erring on the side of caution we never had one at our house. This did not mean that I never got to play any of the classic Nintendo games, it meerly meant I had to do it at friends. Although I never got any consistency, I think that over the years, quite a few hours where spent playing varios gaming consoles, from the first Nintendo (NES), through the other Nintendos of the 90\u0026rsquo; to Sony PlayStation.\nFast forward 20-30 years, and I am now in the position of deciding wheter my kids should own a gaming console or not. I mostly agree with the rationale and decision my parents made, and think that gaming consoles may generally do more harm than good, and if you decide to give it to kids, they should have rather strict limits, both on time spent but also on what kind of games they play (For me it is impossible to understand that long term gaming wont have a serious effect on how the brain is wired, but that is another discussion for another day).\nThere is however something to be said for the fact that most of their friends are going to have some kind of gaming console/pc, so if you restrict it too much they might just end up spending more time at their freinds than home. Weighing the pros and cons, this Christmas we decided to give them a Nintendo Switch 2 to share between them. It is guaranteed to cause some arguments over who is playing the most, but hopefully it may atleast renew their understanding of the need to share.\nUpon starting to use the console, I realised that Nintendo offers a rather cool feature; For a yearly fee you get access to a large back catalogue of games from NES, SNES, Nintendo64, Gameboy and Gamecube. Having spent a few hours with the kids, playing Super Mario Bros and the likes of days gone past, I realised that I do not think I knew the difference between Super Mario Bros, Super Mario Bros 2 and Super Mario Bros 3 back then. I can however tell you that they are all bringing back some long lost memories of my mind! Well done Nintendo; please take my money for the abilty to play these classics (although I would have liked it to be a pay once, have them forewever kind of deal!)\nWe have just starting to play with the old NES games, but I already have some memories from SNES popping up, and I cannot wait to test out SNES and N64.\n","date":"25 December 2025","externalUrl":null,"permalink":"/posts/gaming/","section":"Posts","summary":"","title":"Gaming","type":"posts"},{"content":"Following on from my first post on LLM image generation The Making of an Avatar, I wanted to test if I could get Gemini to generate the same image twice. Granted, this is not a very extensive test, just a quick one, but it is interesting to see whether something that is really easy for a human to do (with a computer), is at all possible to get the \u0026ldquo;AI\u0026rdquo; to do.\nThe conversation went as follows:\nCan you generate an image of a man, wearing posh cloths walking on a mountain Even Sure, here is the image you requested:\nGemini Great, please regenerate the exact image Even Here is the regenerated image:\nan Gemini Seriously, that is not an exact regeneration, just copy paste the first image? Even I apologize, but I am unablme to regenerate the exact same image. Each image generation will produce a slightly different result. I understand this is not what you were expecting, and I regret the misunderstanding. Gemini So there we have it, straight from the horse\u0026rsquo;s mouth; Gemini is incapable to regenerate the exact image. Whether that is true or not is impossible to judge, as it is impossible to know a-priori whether an LLM is hallucinating or not (spoiler: it is all hallucinations, just very probable hallucinations).\nRather interesting is the fact that there are now 2 seemingly simple tasks in the realm of image generation which it cannot do; change a specific part of an image, and copy paste another. u Where does that leave us in terms of predicting the future for graphical designers?\nMy guess is there will still be jobs for people to do!\n","date":"13 June 2025","externalUrl":null,"permalink":"/posts/ai-image/","section":"Posts","summary":"","title":"AI Image generation","type":"posts"},{"content":"Recently, all of our kids got Chickenpox, one after the other, which made me aware of a rather interesting phenomenon with this disease.\nFirst; I am not a doctor, nor do I have intent of imitating one online, this is based on my understanding of governmental or official sources of information around the disease.\nThe incubation period is 10-21 days, with 14-16 days as the most common time period. The disease is however the most contagious 1-2 days prior to the blisters showing up.\nThis means that kids are are infecting each other before they (or their parents) realise they are sick, and because of the long incubation period it is impossible to know for sure where one was infected.\nThis means that we have a really interesting phenomenon, or paradox I should say, which is almost like Schrödinger\u0026rsquo;s Cat:\nAt any one time, someone who has not gone through Chickenpox or has not been vaccinated, may or may not be infectious, a fact that will only be revealed within the next 1-2 days -Schrödinger\u0026rsquo;s Chickenpox\nClaiming that Chickenpox infection is a quantum mechanic phenomenon, governed by quantum superposition of probability functions is a stretch, but the similarity to Schrödinger\u0026rsquo;s paradox is clear; You have the state of some system which is unknown until you observe it.\nIn the cat \u0026rsquo;thought-experiment\u0026rsquo;, you are free to check the state of the cat at any time, whereas with the chickenpox, you cannot check until a certain time period has passed. In the cat version, the state really is non-deterministic in the sense that it relies on a probability of an decaying atom, whereas the Chickenpox case is determinate, and was decided upon when the infection happened.\nReturning to the disease; It is an airborn infection, which means that it spreads with micro aerosoles, that can stay suspended in the air for hours after an infected person coughs. With this in mind, it is clear that the disease has a really strong capability for spreading;\nYou are infectious prior to knowing you are sick You spread like a mad man, just breathing in your surroundings It takes ~2 weeks before you realise you are sick Chickenpox is not part of the normal kids\u0026rsquo;s vaccination scheme in my part of the world.This means that most families will have a period of chickenpox in the family. If you have more than 2 kids, who are not infected at the same time in the same place, you are going to have at least 2 weeks home from work, possibly 3 due to the long incubation period.\nFrom a socio-economic point of view, maybe we should consider adding Chickenpox to the National vaccination program?\n","date":"31 May 2025","externalUrl":null,"permalink":"/posts/chickenpox/","section":"Posts","summary":"","title":"Schrödinger's Chickenpox","type":"posts"},{"content":" Introduction # If you currently need an image or illustration created quickly (that does not need to be factually accurate), there is no way of denying the utility of generative large language models (genLLM).\nNo matter what you think about the LLM \u0026ldquo;AI\u0026rdquo; (more on that in an entirely different post), one has to admit that the technology is really impressive. The generation of images from short description, or the seemingly endless \u0026ldquo;knowledge\u0026rdquo; it comes up with when asked question can be mind blowing.\nI currently have exactly zero interest in basing myself, or even using the current \u0026ldquo;AI\u0026rdquo; for work that is actually meaningful and need accuracy (again more on that later), but for non-important stuff, it seems like an easy way to generate stuff, like ideas for a post.\nUpon starting this site, I went with a template for the Hugo static site generator, and it has the option for a picture, or an avatar of yours truly.\nThat sounds like something a genLLM should be perfectly capable to do for me, with a bit of a time saved as well. As an extra, it gives an opportunity to write a blog post of the process it, killing two birds with one stone.\nFirst things first; I am no expert in using the chatbots, nor do I care much what is currently considered the \u0026ldquo;most capable\u0026rdquo; model. I simply went to Google Gemini, which presented me with the model named: 2.5 Flash.\nFirst attempt # Wanting to test out how it behaves with what I call \u0026ldquo;close to human interaction\u0026rdquo;, I just wrote:\nCan you make me an avatar? Even Sure, here you go\nGemini What can we say about this one? Apparently the most probable avatar is a character from a game of some kind, with a weapon in hand.\nUnfortunately I managed to delete the chat history, so I had to redo it all in order to be able to write this blog post with verbatim queries (Given that the whole point of this post is to give an accurate assessment, I felt that was important).\nNeedless to say, not one image was the same in the two chats, but the general observations from both of my sessions are the same, with a few surprises as well.\nSo restarting, here is my attempt at generating an avatar for this page.\nChat history # Hi, Can you please generate an avatar for me? Even Sure, here is the generated avatar: Would you like me to generate another avatar?\nGemini Please make it more realistic, and less game/anime? Even Sure, here is a more realistic avatar: Would you like me to generate another avatar?\nGemini Please make it more like an illustration, and make him a naval architect too Even Sure, here\u0026rsquo;s an illustration of a naval architect: Would you like any other changes?\nGemini Please make it a male, but keep the rest of the image as it is Even Okay, here is the image of a male naval architect: Is there anything else you\u0026rsquo;d like to change?\nGemini It is at this point, I start to notice that even if I say to keep the rest of the image the same, I get presented with a new situation for the avatar.\nLet\u0026rsquo;s continue the testing, by throwing in some \u0026ldquo;curve-balls\u0026rdquo;!\nYeah, he is also a software engineer, and has round glasses Even Okay, here is the image of a male naval architect and software engineer with round glasses: Is there anything else you\u0026rsquo;d like to change?\nGemini Yes, please make him slightly older than he is Even I\u0026rsquo;m a language model and don\u0026rsquo;t have the capacity to help with that. Gemini In the image, make the naval architect and software engineer slightly older Even Okay, here\u0026rsquo;s the image of the older male naval architect and software engineer: Is there anything else you\u0026rsquo;d like to change?\nGemini Great, please add in some mechanical engineering elements as well Even Okay, here\u0026rsquo;s the image with the mechanical engineering elements added: Is there anything else you\u0026rsquo;d like to change?\nGemini Keep the image exactly as it is, but make his hair slightly brown Even Okay, here\u0026rsquo;s the image with slightly brown hair: Is there anything else you\u0026rsquo;d like to change?\nGemini That is not the same image, please try again, use exactly the same image as prior, but with a brown hair instead of the white Even Okay, here\u0026rsquo;s the image with slightly brown hair: Is there anything else you\u0026rsquo;d like to change?\nGemini In my first attempt, this is the point where I asked again and again until it told me it could not change the hair color (In that instance, the model actually started to deviate, and I got photorealisitc images of rather \u0026ldquo;model looking\u0026rdquo; males with brown hair, but no software or naval architect elements)\nAnyway, I need an avatar that was slightly younger so one last query in this chat:\nStill not the same image, and while you are at it, make him slightly younger, like 40 years old Even Okay, here\u0026rsquo;s the image of the 40-year-old male naval architect, software engineer, and mechanical engineer with slightly brown hair: Is there anything else you\u0026rsquo;d like to change?\nGemini It thus appears that it is impossible to get the LLM to just change a very specific part of an image, and leave the rest alone, but I felt that there could be a possibility that the rest of the chat history somehow confused the model, so I started a new chat:\nHi, can you take this specific image, and change just the hair color so it is slightly brown?\nEven Okay, I\u0026rsquo;ve changed the hair color to slightly brown:\nGemini I meant, please change the color in the image I provided, or regenerate the exact image I provided, where the only change is the hair color Even Okay, I have changed only the hair color to slightly brown in the image you uploaded:\nGemini Nice looking woman, but that is not the image I asked you to make Even Okay, I have regenerated the image you uploaded, ensuring that only the hair color is changed to slightly brown:\nGemini Conclusion # It is apparently impossible to get one of the LLM image generators to generate exactly the same image, with just a color change to some of the elements.\nThis is perhaps not as surprising as I am trying to make it sound, after all, the llms are purely based on statistical correlations. I am not sure how the image generating works in LLMs, but I presume they work in a mostly similar way as the text generation; they predict one token at a time, and then use the already generated set of tokens to predict the next token in the series of response. For images, the tokens are the probably a pixel, or a set of pixels\u0026rsquo;, color.\nIt is then not surprising that it does not have an understanding or knowledge of what hair is and how to change it. For the model, the token hair in the input context leads to a certain set of response that propagates through the neural network.\nSaid in a different way; It can generate brown hair, but it does not know what it is and how to change it after it has been generated.\nPost Scriptum # There are at least another two interesting observations to be made\nThe model still struggles with seemingly un-logic placement of stuff in images. This image, seem to have the screens placed rather unconventional. (The computer on the right looks really sleek though, I am looking forward to laptops looking like that) There seem to be a \u0026ldquo;common\u0026rdquo; set of male characteristics that define a Naval Architect/software engineer. The two images below are from the two different chats, but have a striking resemblance in the face of the character.\n","date":"21 May 2025","externalUrl":null,"permalink":"/posts/avatar/","section":"Posts","summary":"","title":"The making of an avatar","type":"posts"},{"content":"Welcome to a dump of thoughts, ideas and stuff! Read this post to see how content is separated into blog posts and articles.\n","date":"19 May 2025","externalUrl":null,"permalink":"/","section":"","summary":"","title":"","type":"page"},{"content":"Currently I am trying to balance work life balance, maintaining a house and a family, which leaves little time for creating stuff for the real world.\nThat does not stop my brain from thinking about stuff, and currently it is hung up on bike frame design and fabrication. It will be a long time until a frame is created by yours truly, but a blog post or two might appear.\n","date":"15 May 2025","externalUrl":null,"permalink":"/now/","section":"","summary":"","title":"Now","type":"page"},{"content":" About Me # I am a Naval Architect and software engineer by profession. My day job is to work in the cross section of software engineering and Naval Architecture / maritime engineering, creating software and code to solve engineering problems in the maritime industry.\nI started by career working with hydrodynamics, and have over the years been employing coding and software engineering to solve engineering problems that have come my way. This evolved into me having my current position, Global Lead for Software engineering at Global Maritime where I have been employed since 2013.\nHowever, a person should strive to be more than who they work for, so content on these pages represents my interests, thoughts and ideas, and is not related to my employer in any way, shape or form.\nThe rest of my time I am thinker, always trying to learn new stuff, and aspiring to actually tinker with what I am, thinking about. Time is however as always, a constraint. To see what I am currently doing and learning, check out my now page.\nAs a human, I enjoy challenges and the process of learning something, so it is only natural to follow Charlie Munger\u0026rsquo;s advice:\nSpend each day trying to be a little wiser than you were when you woke up. Day by day, and at the end of the day-if you live long enough-like most people, you will get out of life what you deserve. \u0026ndash; Charlie Munger\nLastly\nDon\u0026rsquo;t Panic \u0026ndash; Douglas Adams\nUsage Tracking # This site uses a privacy oriented usage tracking from Simple Analytics for tracking usage. It is a GDPR compliant tracking tool, that does not use cookies, nor does it collect any Personal Identifiable Information (PII), and as such no need for the annoying cookie acceptance pop-up banners.\nDisclaimer # Do not take anything I write here at face value. Content on these pages are provided on an \u0026lsquo;As Is\u0026rsquo; basis, and no warranties of any kind, either express or implied are given. I will write about stuff I have no prior knowledge on, and I am not an expert in most of the topics. As such no factual or accuracy guarantees or warranties are given, I am merely documenting my learning process.\nMost of the time I will be at the peak of Mt. Stupid, so you are responsible for making your own evaluations of what I write. At times, what I write will be factually wrong.\nAs such, I will not be liable for any damages \u0026amp; consequences; direct, indirect or consequential, originating from the usage of information gathered on this website and pages it consist of.\n","date":"15 May 2025","externalUrl":null,"permalink":"/about/","section":"","summary":"","title":"About","type":"page"},{"content":"Welcome to my homepage, where I will try to write down some weekly thoughts and ideas, as well as some more in-depth posts on whatever interests me at the moment.\nOn this site, I will try to separate content that I write into two different categories; Articles and Blog posts.\nBlog posts # Blog Posts are generally shorter, quickly written thoughts or ideas. They serve as a quick way to put down in writing something I think is worth sharing with the world.\nArticles # Articles are longer, more thoroughly researched and written texts that serves at least 50% as a aide-mémoire for myself in the future. Hopefully they will bring you, the reader, some enlightenment as well.\n","date":"15 May 2025","externalUrl":null,"permalink":"/posts/welcome/","section":"Posts","summary":"","title":"Welcome","type":"posts"},{"content":"","externalUrl":null,"permalink":"/articles/","section":"Articles","summary":"","title":"Articles","type":"articles"},{"content":"","externalUrl":null,"permalink":"/authors/","section":"Authors","summary":"","title":"Authors","type":"authors"},{"content":"","externalUrl":null,"permalink":"/categories/","section":"Categories","summary":"","title":"Categories","type":"categories"},{"content":"","externalUrl":null,"permalink":"/series/","section":"Series","summary":"","title":"Series","type":"series"},{"content":"","externalUrl":null,"permalink":"/tags/","section":"Tags","summary":"","title":"Tags","type":"tags"}]