<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Ivan (Jonan) Georgiev on Medium]]></title>
        <description><![CDATA[Stories by Ivan (Jonan) Georgiev on Medium]]></description>
        <link>https://medium.com/@thejonan?source=rss-900f1a93f712------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*iUgFopaAWstoTuzg97AfZg.jpeg</url>
            <title>Stories by Ivan (Jonan) Georgiev on Medium</title>
            <link>https://medium.com/@thejonan?source=rss-900f1a93f712------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 14:44:16 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@thejonan/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The uncrowned king of modern AI: embeddings]]></title>
            <link>https://medium.com/@thejonan/the-uncrowned-king-of-modern-ai-embeddings-86e892001a8c?source=rss-900f1a93f712------2</link>
            <guid isPermaLink="false">https://medium.com/p/86e892001a8c</guid>
            <category><![CDATA[agi]]></category>
            <category><![CDATA[embedding]]></category>
            <category><![CDATA[word-embeddings]]></category>
            <category><![CDATA[society]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Ivan (Jonan) Georgiev]]></dc:creator>
            <pubDate>Tue, 19 Aug 2025 08:51:56 GMT</pubDate>
            <atom:updated>2025-08-19T08:51:56.944Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>SPOILER ALERT: There’s a “proof” at the end, that modern AI is not, and cannot surpass humans in intelligence.</blockquote><p>Almost everybody vaguely interested in the field of AI, has already heard about the famous: <em>Attention is all you need</em> paper¹, and that we owe the <em>Transformers</em> this AI explosion, although I’m sure some people wonder how related is that to the movie.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*yIbUet8GmdSB-kKN" /><figcaption>Photo by <a href="https://unsplash.com/@praswinprakashan?utm_source=medium&amp;utm_medium=referral">Praswin Prakashan</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>However, I claim that another underlying, architectural concept plays a more pivotal and fundamental role — <strong>the embeddings⁴</strong>.</p><p>They are wandering around the field for quite a while, and became famous with the <em>word2vec</em> model², which for the first time <em>successfully</em> mapped this scattered realm of words, into fixed-length, numerical vectors — <em>semantically</em>! The popular example is that if you take the vector for king, subtract from it the vector for men and add the one for woman - you&#39;ll get a vector which is almost identical to the one for queen! It was so exciting back then! And still, kind-of, is.</p><h3>How do they work?</h3><p>In order to appreciate the role of this technique, one needs to have a conceptual understanding of how embeddings are built. Let’s try to do this without digging too much into technical details.</p><p>First of all — what is the challenge? Most machine learning models that we’ve come up with, are based on some algorithms that work on numbers. Think of an air conditioner that only has ON/OFF switch, vs. one that has a thermostat knob — with which one it is more likely to achieve and maintain, the desired temperature? Exactly.</p><p>The current belief is that the brain works in this fashion — it has hundreds of trillions of “knobs” (in many regions, and many levels), that it turns slightly and constantly in order to adapt our behaviour to the external environment. “Muscle memory” is a good example of how many repetitions of the same activity allow the brain to “tweak the knobs” (more precisely — synaptic connections) good enough for the desired outcome.</p><p>The problem with words is that they are not “numerical”, not <em>knob-like</em>. They are more — <em>switch-like</em>, and are not easy to fit into this smooth, continuous numerical world. Many attempts were made, based on grammar knowledge, linguistics, statistics, etc. to “parametrize” the words turning them into vectors based on some fixed, explicit rule, but inherently each of these a-priori methods stigmatize irreparably the model, so it cannot efficiently embed any real semantic afterwards.</p><p>Until the <em>word2vec</em> approach came. The guys said — “Well, we need a vector for each word, and <em>we don’t really care what exactly this vector is</em>, as long as it successfully embeds some word’s semantic!”. But how do you define “good embedding”? As it often turns out in machine learning field — it is easier to set the goals, than the actual knowledge.</p><blockquote><em>So a successful semantic embedding is such that words with different meanings are mapped to very different vectors, and synonyms are mapped to similar vectors. That’s it.</em></blockquote><p>The exact <em>mechanics</em> for achieving this are different and more technical, but the short story goes like this:</p><ul><li>Each word (along with its variations) is put in a vocabulary and assigned a vector. Initially — all vectors are set at random.</li><li>The system converts every input word it is about to process into its assigned vector before doing anything with it down the road.</li><li>The actual <em>doing</em> with the word-vectors can differ, but a simple (and totally valid) example is: <em>“Given this sentence with one word removed — tell me which one do you think it is?”</em>. And the system guesses the index for the missing word in the vocabulary. This approach is called <em>skip-gram</em>.</li><li>In the process of training — large corpuses of text are used, so it is very easy to “feed” the system randomly hiding words, and with every guessed answer (terribly wrong in the beginning!), provide the correct one (i.e. — word), and living in the numerical world — fine-tune all internal “knobs” of the system, <em>including the actual vector for this word</em>.</li><li>This process is repeated thousands of times, on hundreds of thousands of words of text, and gradually the vectors for each word start to contribute better and better in the task (whatever it is), which is only possible, if they <em>represent better and better, the semantics of each word</em>.</li></ul><p>At the end of the training, we have this wonderful equation:</p><p>king - men + woman ≈queen.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*VWiMTx9cdjC7ey6R" /><figcaption>Photo by <a href="https://unsplash.com/@timgid?utm_source=medium&amp;utm_medium=referral">Timothée Gidenne</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Why is this important?</h3><p>However exciting this sounds, it has various drawbacks, but let’s see what the benefits are first. And they are fundamental:</p><ul><li>The <em>uneatable</em> realm of words is easily (and very performance efficient, by the way!) fitted into the world of numbers. Aside from using this embeddings mechanism as part of a larger model, pre-trained embeddings (so called — static), can be taken and used as well. With caution, of course — context matters, as stated a bit later.</li><li>It is now (relatively) easy to mix modalities — since the images are easier to turn into numbers (they <em>are</em> represented with numbers digitally, anyways), mixing the two worlds is not just <em>easier</em> — it became <em>possible</em>. There are different approaches which go beyond the scope of this article, but the word embeddings are the key.</li><li>Dimensionality — how big these vectors are — 4 numbers? 16 numbers? 1024 numbers? The developer of the model is free to choose, but this aspect is <em>the blessing and the curse</em> of embeddings, so we’ll spend more time on it later.</li></ul><p>It is not all bells and whistles, however. There are some inherent problems, and specifics in building embeddings, and it is expected to be so. Otherwise, we’ll have one universal, perfect embedding to <em>rule them all</em>.</p><ul><li>However elegant and clever the embeddings are <em>left to grow on their own</em>, and capture the semantic of the words they are mapped to, this all happens in a <em>certain context</em> — the actual task in hand. And the embeddings are always biased from the task, which is used to build them. When a simple skip-gram (or n-gram, or CBOW³) is used — the actual training texts, influence the embeddings. Of course! More text does not necessarily mean better embeddings. It leads to more vague ones, for sure.</li><li>There is no recipe for the “proper size” of the embeddings vectors. As always in machine learning — too big means less flexibility (and requires a lot of data), too small — much worse accuracy. But, since we’ve landed on <em>dimensionality</em> again. Let’s dig into it.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*FQDa0bwaYbeEw8Qx" /><figcaption>Photo by <a href="https://unsplash.com/@brandialxndra?utm_source=medium&amp;utm_medium=referral">Brandi Alexandra</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>The “proof”</h3><p>The actual size of embedding vectors plays a crucial role in the capabilities of the model. Commercial models don’t disclose this information, but here are some numbers for the open and/or more familiar ones:</p><ul><li>GPT-2 large: approx.1,280 dimensions for the embedding vector.</li><li>GPT-3: 12,288 dimensional embeddings.</li><li>LLaMA‑3.1 (8B model): 4096 dimensional embeddings.</li><li>LLaMA‑3.1 (405B model): 16,384 dimensional embeddings.</li></ul><p>Two important aspects pop up: (a) embeddings vectors’ size varies <em>hugely</em> from model to model, and (b) it has a huge impact on the size of the model — number of parameters.</p><blockquote><em>And all this is quite intuitive, after all — the dimensionality of the semantics that you can embed in all the words a model uses, determines how “deep” model’s understanding is. Figuratively speaking — it defines the “world of the model”. And it has a fixed, limited number of dimensions.</em></blockquote><p>This is my “proof” that modern AI is far from becoming <em>super intelligent</em> or even <em>more intelligent</em> than humans. After all — we’ve grown up from a world of cavemen, then gatherers and hunters, then villagers, extending our points of view, adding new perspectives, new <em>dimensions</em> to the world of our mind. Something LLMs can’t do. Yet.</p><p><em>P.S. As mentioned in previous articles — let’s not mess up “knowledge” and “intelligence”.</em></p><p><em>P.S.S. Could it be that we, humans, aside from DNA, differ in how are “embeddings” are initialized at birth? And that leads to totally different personalities.</em></p><p>[1] <em>Attention is All you Need</em>: <a href="https://arxiv.org/abs/1706.03762">arXiv:1706.03762</a></p><p>[2] <em>Efficient Estimation of Word Representations in Vector Space</em>: <a href="https://arxiv.org/abs/1301.3781">arXiv:1301.3781</a></p><p>[3] <a href="https://en.wikipedia.org/wiki/Word2vec">https://en.wikipedia.org/wiki/Word2vec</a></p><p>[4] The whole article will try to explain <em>embeddings,</em> but the concept of <em>vectors</em> is expected to be understood. It’s a line of numbers. Your GPS position is a two-dimensional vector (e.g. [42.56385467372551, 23.278346793990014]); the size of your dishwasher is a three-dimensional vector (e.g. [600, 555, 820]); the CMYK mixture for your kitchen color is a 4-dimensional vector (e.g. [13,9,0,12]), and so on. The size of the vector, or its <em>dimensionality</em> is the count of numbers it contains. More from here: <a href="https://en.wikipedia.org/wiki/Vector_(mathematics_and_physics)">https://en.wikipedia.org/wiki/Vector_(mathematics_and_physics)</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=86e892001a8c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The AI impact on software industry will be more profound than we anticipate]]></title>
            <link>https://medium.com/@thejonan/the-ai-impact-on-software-industry-will-be-more-profound-than-we-anticipate-5406e1cd59f2?source=rss-900f1a93f712------2</link>
            <guid isPermaLink="false">https://medium.com/p/5406e1cd59f2</guid>
            <category><![CDATA[society]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Ivan (Jonan) Georgiev]]></dc:creator>
            <pubDate>Sun, 14 Jan 2024 18:57:29 GMT</pubDate>
            <atom:updated>2024-01-14T18:57:29.999Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*EhH1vNIx20KmE0It" /><figcaption>Photo by <a href="https://unsplash.com/@fsmurrayphoto?utm_source=medium&amp;utm_medium=referral">Fiona Murray</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p><strong><em>TL;DR</em></strong></p><p><strong><em>The advances in AI will change IT industry much deeper, not just because many tasks will be automated, but because they won’t be needed in the first place.</em></strong></p><p>M y ex-wife was convinced that our son should know how to write software, and that will keep him “safe” for a job.</p><p><em>“Master typewriting and a language, and you’ll always have a job”</em>, her father had told her. Well, knowing English, when this is not your mother tongue, definitely helps, but I’m not sure she has ever made a single penny with typewriting.</p><p>Typewriters <em>were</em> around, when I was a kid, and we too, had one — permanently situated under the bed. And, although I was in my early teens at that time, I clearly remember a conversation between my dad and a relative of ours.</p><p>He was a professional translator — fluent in English and German (which, again is not our mother tongue), and was <em>typing for a living</em> to say so. Computers were already around — and not the very first ones, but an Intel-based x86 architecture ones, with MS-DOS, Word processing, etc.</p><p>However, our relative had just bough a professional typewriter! Ok, a “modern” one — with (somehow) interchangeable fonts, assisted typing, and so on, but I still remember how difficult was for my father (an electrical engineer himself) to understand <em>why</em> such a device is still in use?! I’m pretty sure that it wasn’t in use for long, indeed.</p><p>I recall this now, when arguing with colleagues of mine, from IT industry, about the profound change that is about to happen — it seems, often the people deepest in a field are most oblivious for the upcoming changes. Remember — my relative was making his bread, typing.</p><p>One of my favorite quotes is from a famous hockey player — Wayne Gretzky:</p><blockquote><em>“A good hockey player plays where the puck is. A great hockey player plays where the puck is going to be.”</em></blockquote><p>And, although the AI threw the puck in quite a distinct direction, I’ll try to break down the <em>driving forces</em> it creates, by elaborating a bit on three main aspects of the software industry:</p><ul><li>The actual <strong>process of development </strong>— code writing.</li><li><strong>Types of problems</strong>, requiring software solutions.</li><li>Interaction with the <strong>real world</strong>.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*heDGfZxrHPKcsMko" /><figcaption>Photo by <a href="https://unsplash.com/@lennykuhne?utm_source=medium&amp;utm_medium=referral">Lenny Kuhne</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Code writing</h3><p>This quickly became quite a boring topic. Amazed by the leap OpenAI made with ChatGPT 3.5, I’ve started some experimental projects of writing code from description, automatic solving issues in GitHub, etc. a year ago. So did many other people, and already there are numerous projects that employ that approach, and even the biggest sceptics, started using CoPilot, and are enjoying it.</p><p>Yes, most of these projects are not mature yet, and still require quite a lot of human interaction. Many of them are trying to solve the one-sentence-description problems — like <em>“Make a snake game in Python”</em>, which is cute, and sometimes amazes, but far from practical and, honestly — meaningless.</p><p>In order to utilize the new power, a change in the workflow is needed, and new tools as well — fine-tuned custom models, <em>real</em> test-driven development, task descriptions becoming part of the <em>source</em>, etc. — many details in each aspect, making it a good candidate for a dedicated article. Stay tuned!</p><p>All in all, coding will be automated — not entirely, perhaps, but largely for sure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_0mWs_dWV7IOdNE-" /><figcaption>Photo by <a href="https://unsplash.com/@uxstore?utm_source=medium&amp;utm_medium=referral">UX Store</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Problems that software usually solves</h3><p>For me, this is the deepest movement that AI (will) create in the industry, and one of the least discussed and anticipated ones.</p><blockquote><em>The types of problems software usually solves, are exactly the types of problems a generic, pre-trained large language models can solve best.</em></blockquote><p><strong>The very need for many of the software solutions will disappear</strong>.</p><p>According to different sources — either by looking at software industry revenues by types of software, or by checking the software engineering market share by industry — more than half of the software engineering happens for enterprise and productivity applications, and the easiest way to get a description of types of tasks such software solves, is to check the functionality of a typical <strong>Enterprise Resource Planing (ERP)</strong> system.</p><p>A non-exhaustive list of tasks a ERP system solves, based on the <a href="https://en.wikipedia.org/wiki/Enterprise_resource_planning">Wikipedia article</a>, include: finances and reporting, human resource management, projection… well, let’s directly address them from AI perspective:</p><ul><li><strong>Integration</strong> and <strong>Automation</strong> are quite literally what the whole subject is all about — for example, even as of this writing, the OpenAI functions/plugins already provide good means of linking with external tools, big integration (and automation) players, like Zapier, embraced and utilize the new AI tools. I don’t think much persuasion is needed to convince someone that AI tools are going to make integration and automation <em>way</em> easier, in a much broader and user-accessible way.</li><li><strong>Data analysis</strong> and <strong>Reporting</strong> are another two examples of what AI tools are already being used for. <em>Data analyst</em> is often used as a synonym for someone, who understands how machine learning works. It is important to understand that even now, the majority of <em>real</em> value in these two fields lies within defining <em>what</em> to be analysed and reported, and it depends on domain knowledge, and has nothing to do with software development itself. Actually, the need to <em>communicate</em> this to a developer, in order to get the functionality done, is usually the bottleneck. Using an AI agent will definitely render all this process obsolete.</li><li><strong>Accounting</strong> and <strong>Finances</strong>, with the former usually being standardized at a government level, and not requiring a lot of <em>new</em> software, and the latter, again, much more focused on the expert-defined meaning, rather than the actual software tools. From technical and mathematical point of view — there is nothing really complex in these areas — in both cases, the work usually goes in integration with other tools, and communication with IT guys, which again can be achieved easier through an AI agent.</li><li>When it comes to <strong>Customer Relationship Management</strong>, I’m pretty sure that majority of the entrepreneurs that decided to integrate such a system, would be much happier if they could have stayed with their Excel sheets but, somehow, put them “on steroids” and make them <em>smart</em>, so the tedious tasks can be automated, but all the flexibility remains. Sales people don’t like frames (so does the market), and this is why all CRMs tend to be overcomplicated — meeting the framing needs of current software paradigm, with the flexibility of its users. Sales people would <em>love</em> to communicate with AI agent and sheets, instead.</li><li>It is not quite the same with <strong>HR</strong> people, although there are similarities, and it’s actually pretty tight with <strong>compliance</strong> related tasks. Again, the majority of the added value lies outside of the software development realm, and a proper data-aware, or RAG¹-ed agent can serve these tasks quite well.</li><li><strong>Project management</strong> is pretty vague term, and it either includes all other activities mentioned here, or has a more focused definition of <em>time and expense management</em>. Until quite big size, the companies are doing this with (again!) Excel sheets and calendar app, with the latter being the ultimate personal time management tool. Also, quite commonly interacted in automated manner.</li><li><strong>Manufacturing and Supply chains management</strong> is something that has a much higher <em>interaction with the real world</em> aspects, thus making it less prone to AI replacement, but this is the topic of the next chapter.</li></ul><p>With all this, I am not saying that the process of creating software will disappear. The development of more fundamental tools like databases, networking, infrastructure, security and encryption tools, etc. — is here to stay for long, mainly as a backbone of everything that comes next. Just like electrical engineering, and similarly — not as the cherry on the cake.</p><p>However, I can clearly imagine how a systems architect sketches and selects those basic, building blocks, and lets the a properly trained AI agent, to glue them together and deliver the final, customized solution. This is what majority of <em>writing software </em>is today <em>— </em>making these parts work together. And it’ll become obsolete.</p><p>Btw, this type of <em>development</em> is usually done in Java or C#, hence my <a href="https://medium.com/@thejonan/why-you-should-stop-teaching-java-or-c-now-e3a35bb027fd">previous article</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*-vZFNBUpiWjQXqjC" /><figcaption>Photo by <a href="https://unsplash.com/@specna_arms_4s?utm_source=medium&amp;utm_medium=referral">Specna Arms</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Real world software</h3><p>The third driving force, is how related the piece of software is to the real world — to the non-deterministic, non-formalizable natural world. There was a nice article title few months ago, which was something like this:</p><blockquote><em>Why AI can easily write a poem, but struggles to climb a stairs?</em></blockquote><p>Because — from certain perspective — it <em>is</em> harder to keep the balance, while climbing a stairs. Big parts of our brain, btw, are dedicated to motor skills.</p><p>Boston Dynamics, the best robot producing company in the world, funded by the USA military, being under the Massachusetts Institute of Technology for quite some time, managed to develop their robots to be more “robust” to the environment after decades of development. And still the results are <em>far</em> from what an average human being can do. Very far.</p><p>In the light of software development, that makes me think that the more one is involved in real-world related type of development, the less probable is AI to take over — embedded developers, in this sense are much <em>safer</em>. Their expertise just stretches way beyond the virtual world of the software, and they (must) have good understanding of technicalities of completely different matter.</p><p>Another part of development process that is closer to the real world, and is much, <em>much</em> bigger is actually UX development — making something aesthetic and ergonomic is not a trivial task. It involves a lot of creative decisions. However, majority of these decisions are made by the UX designers, not the developers. Combine with the fact that frameworks and software languages involved in front-end development, are so popular — they are among the best to be consumed by the LLMs.</p><h3>Conclusion</h3><p>I don’t see the perspective of AI replacing software developers as a reason for fear, or any kind of threat. Automation has always replaced human labor, and I don’t see anything dramatic this time either.</p><blockquote><em>The biggest concern is not that AI will take a lot of jobs, the problem is that so many jobs are replaceable with automation — clerical and depleted from creativity.</em></blockquote><p>And, as the essence of software development has always been to develop tools, that solve real world, human problems, using this new machine, called computer — similarly will be in the next chapter: solving real world problems, with this new “machine” called AI agent. And, just like with the raise of software, there is an explosion of new concepts, tools to be mastered and new horizons to be reached now.</p><p>The sooner we start to look where the puck is going to be, not where it is now, the less we’ll be surprised by the ever changing reality. And yes, I’m not pushing my son to be a programmer, like his father — there are better things to be learnt — even in this area.</p><p>But, that’s a topic for the <em>next article</em>.</p><p><strong><em>[1]</em></strong><em> Retrieval Augmented Generation — when task-specific data is provided to the AI agent, and specifics in the answer are heavily based on this data, while the understanding comes from the model itself</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5406e1cd59f2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why you should stop teaching Java or C#. Now.]]></title>
            <link>https://medium.com/@thejonan/why-you-should-stop-teaching-java-or-c-now-e3a35bb027fd?source=rss-900f1a93f712------2</link>
            <guid isPermaLink="false">https://medium.com/p/e3a35bb027fd</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[software-languages]]></category>
            <category><![CDATA[future]]></category>
            <dc:creator><![CDATA[Ivan (Jonan) Georgiev]]></dc:creator>
            <pubDate>Tue, 20 Jun 2023 08:54:24 GMT</pubDate>
            <atom:updated>2023-06-20T08:54:24.486Z</atom:updated>
            <content:encoded><![CDATA[<p>Unless you believe your students will be lucky enough to become like one of those 128 COBOL/Fortran¹ developers that are extremely well paid, albeit being deep in their pension days, you better present them with something more future proof.</p><blockquote>Why Java and C#? Why excluding C++ and Python, for example?</blockquote><p>Well, as mentioned in a previous article — I see the software development industry having a sobering shower with the advancement of LLM systems, but language like C++ and Python are <em>on the edge</em>, so to speak. The former is used exclusively to deal with the world of hardware, and the latter — with the world of data. Any data, not just software-related. Both are used to deal with problems/tasks from a realm outside of coding itself.</p><p>Java and C# and related platforms, are “inland” animals — they serve the software industry itself. And as being such, (almost) everything they deal with is an easy target for AI automation. Not that existing systems will be replaced quickly, but… remember — COBOL and Fortran are still around, and not the top choice in the textbooks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*7rSUbNYENulTaleU" /><figcaption>Photo by <a href="https://unsplash.com/@momentance?utm_source=medium&amp;utm_medium=referral">SwapnIl Dwivedi</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><blockquote>What about JavaScript?</blockquote><p>It’s quite a mixed animal. On one hand it <em>does</em> serve the UI world, which is much closer to humans, than the back-end parts, but on the other hand — it’s not a high bar for AI. And also — the web is abundant of existing JS code that AI can learn from. <em>However</em>, the ecosystem around that language is already very easy to use, very quick to add value, and try common tasks. In other words —</p><blockquote>like PASCAL — easy to teach, easy to try, and not entirely useless from production point of view.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*lgZuYDTm_DN24nVN" /><figcaption>Photo by <a href="https://unsplash.com/@pistos?utm_source=medium&amp;utm_medium=referral">Jeffrey Hamilton</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p><strong>One may ask — Why teaching any software language at school — as AI will write all the code?!</strong></p><p>Just within the last few months, ChatGPT provoked an explosion of new areas of learning and development — prompt engineering, (efficient) model fine tuning, context selection, reinforced learning with human feedback, chain-of-though techniques, new infrastructure challenges for scalability, data protection and management, and so on, and so on. And just like we’ve been teaching students in programming classes to code linked lists in Java — over and over, <em>and over</em> again, regardless of the fact that it has better built-in implementation (of course!), introducing them to Python will give a bit deeper understanding of what the core of all these AI systems, actually is based on.</p><p>If you ask why I keep mentioning C++… I believe this will stay the language of industry “backbones” for quite some time — embedded devices, operating systems, networks, actual AI engines, robotics, etc.</p><blockquote>But what should be taught to students regarding the software industry ?</blockquote><p>In order to answer that question, we first need to elaborate on the consequences of emergence of such smart tools capable of solving clerical tasks with ease — like ChatGPT — on the whole software industry. But that’s a topic for another few minutes article. Coming soon.</p><p><em>[1]: Of course, this number “128” is made up, but you’ve got the idea.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e3a35bb027fd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Few more cents on ChatGPT]]></title>
            <link>https://medium.com/@thejonan/few-more-cents-on-chatgpt-5e72c66ffbf9?source=rss-900f1a93f712------2</link>
            <guid isPermaLink="false">https://medium.com/p/5e72c66ffbf9</guid>
            <category><![CDATA[education]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[advertising]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Ivan (Jonan) Georgiev]]></dc:creator>
            <pubDate>Mon, 23 Jan 2023 13:01:49 GMT</pubDate>
            <atom:updated>2023-01-23T13:01:49.827Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>TL;DR:</strong> <em>It is cool and controversial, and definitely not just about software development, and surely more than few cents.</em></p><p>When I was asked 6 months ago:</p><blockquote>“Ok, when do you think the software development will be automated?“</blockquote><blockquote>“Five, to ten years”, <em>was my answer.</em></blockquote><p>Well, OpenGPT caught me a bit off-guard, and now I realize that more precise answer is <em>“three to five”</em>, and probably more on the sooner end.</p><p>And, of course, not all software development will be automated. I keep repeating for several years now, that majority of the software development is deprived of any creativity, and therefore — it is doomed to be automated. Just like compilers now convert the high-level language code into machine instructions better than humans — and it was not like this 20 years ago.</p><p>The AI-powered tools will convert the idea of a product into code in very similar manner. So far any attempts to do that were failing because stating <em>the idea</em> into formal, ambiguous way is at least as much of a philosophical question, as is an engineering one. But now, with the conversational, aware-of-context, incrementally detailing tool like ChatGPT this barrier will fall. And will fall soon.</p><p>But, let’s leave the software development industry on its own merits. After all, it is already too bloated with far more personnel than it <em>actually</em> needs, and the creative ones will always be needed.</p><p>A more significant aspects are coming, however.</p><h3>Search engine revolution</h3><p>I remember the time of <em>AltaVista</em> search engine, and how they’ve fought with <em>Yahoo!</em>. I also remember how quickly they were blown away by <em>Google</em>, just by delivering much, <strong>much</strong> more relevant information to the question asked, and delivering it in much, <strong>much</strong> more focused and usable way — <em>AltaVista</em> result page was bloated with additional, non-relevant, distracting ads, links, suggestions, etc.</p><p>Nowadays few people click on “Page 2” of the searches, and for a good reason — most of the time you get what you are looking for on the first one, and if you don’t — you refine your question and still look on the first result page.</p><p>However, gathering an overview on a certain topic <em>is</em> quite an effort — collecting information from different sources, each potentially involving multiple searches and a dozed page visits, peeling off the information and synthesizing the essence. And this is a <em>zero creativity</em> process, more or less — one is just trying to obtain more <em>knowledge</em> on an existing subject.</p><p>Now, with ChatGPT all that can be reduced in time dramatically — a short conversation with the bot, and “Voillah!” — you get the whole human knowledge on the topic synthesized and elaborated in details — <em>as deep as you want</em>. Sweet!</p><h3>Education as one of the victims</h3><p>Actually, the education <em>is</em> already a victim of the industrial mindset, and its own conservatism — stuck into the realm of <em>knowledge</em>, which is by far superseded of tools like Google, <em>already</em>. But this time, it will get exposed violently and nobody will be able to circumvent the fact that it is not about <em>testing</em> students for knowing facts, but about helping them develop the <em>capabilities of doing something new</em>. Some schools and universities already started the “stick your head in the sand” game, by limiting the Internet access to OpenAI servers. That, of course, will fail.</p><p>At some point, hopefully, most will understand that only the unique parts of the process of education are important, and the only agent that can bring uniqueness is the teacher -</p><blockquote><em>… no testing procedures, no website restrictions, no detailed curriculums can replace the creativity of a teacher, engaging her students into a unique process, that ultimately develops </em>them<em> into something they value </em>themselves<em>.</em></blockquote><p>All of the above — just like textbooks, exams, etc. — are just tools <em>for the teacher</em> to make her life easier and allow her to focus on the important — build and deliver a unique environment for students to grow.</p><p>Hopefully, ChatGPT will speed up the process of educational institutions <em>and the society itself</em> of realizing how depleted of meaning the current system <em>already is</em>.</p><h3>The shadow of advertisement</h3><p>The majority of revenue for Google comes from ads. They are clearly marked in the search results, although the process of <em>ranking</em> the results is not clear at all, and is heavily influenced by objective facts like general popularity of terms, the source, its connectivity, etc. But also with a lot of unknowns — like internal preferences on own products, for example.</p><blockquote><em>Now imagine the new, almighty ChatGPT search engine, which is “rigged” for recommendations on certain brands or products, just based on the companies’ budgets?</em></blockquote><p>Quite scary, huh! Because current results from Google are still referring to the original, genuine content which can always be tracked. This is not the case with AI systems! Neural networks are notoriously bad at explainability — a sore topic in scientific fields. This also makes it harder “to rig” the results, but not impossible, given that they are properly trained.</p><p>So, it will be quite a skill to distinguish the <em>paid ad</em> in the answer of the bot. This might be impossible even for the creators themselves — depending on how the advertisement is facilitated.</p><h3>Stay creative</h3><p>As a software architect, with Master degree in Computer Science, with a special interest in AI, and particularly in neural networks for decades, I would not say anybody should be scared. Automation is ever present topic in human history — from labor separation, to stock-raising, to modern agriculture, to mass production, etc.</p><blockquote><em>What I see as a gauge is how much creativity is involved in a certain task or job, and if there is none left… well, it deserves to be automated! And I doubt that people performing it are very happy, anyways.</em></blockquote><p>What stands out on <em>this</em> type of automation, however, is its wide range of coverage — it touches most aspects of modern human life, and the more general AI becomes, the easier it is to get transferred between fields. And that is a phenomenon we are yet to witness.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*-KB6bltHjSj8X3R2" /><figcaption>Photo by <a href="https://unsplash.com/@kellysikkema?utm_source=medium&amp;utm_medium=referral">Kelly Sikkema</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5e72c66ffbf9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The bad, the ugly, the weak and the good.]]></title>
            <link>https://medium.com/@thejonan/the-bad-the-ugly-the-weak-and-the-good-21509e90c9ac?source=rss-900f1a93f712------2</link>
            <guid isPermaLink="false">https://medium.com/p/21509e90c9ac</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[team-management]]></category>
            <category><![CDATA[agile]]></category>
            <category><![CDATA[org-dev]]></category>
            <dc:creator><![CDATA[Ivan (Jonan) Georgiev]]></dc:creator>
            <pubDate>Mon, 15 Jul 2019 07:55:38 GMT</pubDate>
            <atom:updated>2019-07-15T07:55:38.718Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/624/1*HxMRQ4b1JbkGl0QEKc1ZYQ.jpeg" /><figcaption>I’ll put the name on this stone…</figcaption></figure><p>There are couple, major problems in development teams, and none of them is technology related. Problem number one is <strong><em>Communication</em></strong><em>.</em></p><p>There is a wrong illusion that a software developer should be good in communicating with the machine. <em>No!</em> The compiler does that. A software developer communicates with people, and with most of them — indirectly. The creators of the libraries he uses; the people from other teams in the company; the end-customers, by trying to fulfill their requirements; the management body, communicating the tasks, and so on. Unless you work for <em>Intel</em>, <em>AMD</em>, or <em>NVidia — </em>you are not likely to really communicate with the hardware directly.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ot1UeCPZKI5HyfxnLfYyfA.jpeg" /><figcaption>You’ll make a wrong move…</figcaption></figure><p>The second, and it comes partially as a consequence of the first one, is <strong><em>Trust</em></strong><em>. </em>Actually — the <em>lack</em> of it. It is not difficult to see why. Discussing a certain software product, is like standing in Louvre, talking about Mona Lisa’s posture, the spark in the eyes, her hands, etc. Then, going out and telling the development team about your insights, giving them few <a href="https://mymodernmet.com/tibetan-buddhist-monks-sand-art">tons of sand</a>, and asking them to paint the Vatican square with Mona Lisa’s portrait… That big is the jump between high-level, feature related talks, and the granularity of the details software developers need to deal with.</p><p>Of course, industry is getting mature — there are a lot of libraries, frameworks, templates, development tools, etc. There is an open source movement, which significantly raised code reuse. But, the point is that making such big jumps <em>always </em>bring errors. Big and small. Design-related and implementational. <em>A lot</em>. When this happens the rest of the team, naturally, as part of our human nature, becomes less trustful. As a result, many teams start relying <em>heavily </em>on code reviews, thinking this is some miraculous panacea. They’re essentially saying <em>“We don’t believe you, anyways, so we’ll stick several other people, to make sure you’re making less mistakes”</em>. Well, the role and importance of code reviews is a totally different topic — I’m not anti-CR advocate. <em>No</em>. Still, they add up to the lack of trust. And, there is no such thing in the world, as — <em>teamwork without trust</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/759/1*cFobNO4jtDB0Fkgyf61Zag.png" /><figcaption>Gotcha!</figcaption></figure><p>And the third major problem comes from the lack of <strong><em>Completeness</em></strong><em>. </em>Many companies talk Agile. Few understand it. Even less do it. The deeper into the development team people love the idea, the more useful it is. And we, engineers, naturally, don’t. We live in our brains — construct buildings, tools, machinery, software — this is very important skill for any engineer to have. If not — better do something else.</p><p>In the same time, this attitude makes us reluctant to verify, to confront our imaginary creatures with the reality. And this is what Agile stands for. An attitude that marketing people have. Sometimes — sales people too. But, we don’t like them very much either… Kind-of… Anyways, where was I?! Ah, <em>completeness</em>! What does engineers’ love with marketing have to do with it?! Well, not much, really. But, overcoming the inherent fear and discomfort of matching imagination with reality, by settling on a solution, even when it does not feel perfect — this has a lot to do with both <em>completeness</em>, and <em>Agile</em>.</p><blockquote>The process usually consists of many decisions, and like always in life — decisions are about what should be <em>left behind</em>, not about what will be taken away.</blockquote><p>In software industry, it is full of opportunities, possible paths, tools, branches of possibilities, and on each step there is plenty to be left behind. It is what it takes to complete something. And this is why we so often fail to do it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UW_XqUVTrwg2n3uPsfc8xQ.jpeg" /><figcaption>Easy, easy…</figcaption></figure><p>The beauty is that same pill works for all three of these problems. Does not solve them completely! But, helps a lot. It is called <strong><em>Responsibility</em></strong><em>.</em> Or, more accurately — the push to bring it back in <em>the business</em>. Splitting the work into responsibility zones, i.e. modules, sub-products, libraries, etc. and assigning a single person, because there’s no such thing as shared responsibility, to each of them, facilitates a type of behaviour which resists all of the three major problems.</p><p><em>Completeness</em> is immediately enforced — from architects to settle on certain structure, to developers to deliver something usable by the other team members. The <em>trust</em> improvement comes as a by product of the fact that being solely responsible for a piece of code, makes you care about its quality much more. As a result, the code gets better and better over time. So does the trust from the others.</p><p>And, finally, the <em>communication</em> aspect. Splitting the work into standalone pieces, means that they need to be assembled at the end, in order to have the final product in all its glory, and the glue for this process, is <em>communication — </em>in the form of documentation, diagrams, presentations, or whatever.</p><p>Only, after addressing these aspects of the team setup, one can turn to discuss the technologies to be used — for both the development itself, and the tasks of managing it. But, that’s another story.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*VbpP2jBYCjcegQs8eGMK8Q.jpeg" /><figcaption>“No name” is the name.</figcaption></figure><p><strong><em>Note:</em></strong><em> All images are from </em><a href="https://www.imdb.com/name/nm0001466/?ref_=tt_ov_dr">Sergio Leone<em> </em></a><em>’s 1966 movie “</em><a href="https://www.imdb.com/title/tt0060196/?ref_=nv_sr_1?ref_=nv_sr_1"><em>The good, the bad and the ugly</em></a><em>”.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=21509e90c9ac" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How much a software architecture design should cost?]]></title>
            <link>https://medium.com/@thejonan/how-much-a-software-architecture-design-should-cost-1af390b9501f?source=rss-900f1a93f712------2</link>
            <guid isPermaLink="false">https://medium.com/p/1af390b9501f</guid>
            <category><![CDATA[architects]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[architectural-design]]></category>
            <dc:creator><![CDATA[Ivan (Jonan) Georgiev]]></dc:creator>
            <pubDate>Tue, 05 Mar 2019 21:45:20 GMT</pubDate>
            <atom:updated>2019-03-05T21:45:20.402Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>A lot!</strong> The importance of setting the right direction of a journey before your very first step, can never be stressed enough. While it should be obvious. But that’s a title for another story. This one is about <em>what part of the budget for a software system development should go to the architecture design process</em>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*uCxzSfBtfb0msvxt" /><figcaption>Photo by <a href="https://unsplash.com/@sunburned_surveyor?utm_source=medium&amp;utm_medium=referral">Scott Blake</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>I’ve recently been asked this question. Since I’ve always loved relating the software development to civil construction, I’ve asked a close friend of mine, an architect, what is considered a normal share for the design of a building. His industry is couple of thousands of years older — should be more mature, I guess.</p><blockquote>A tough question… depends on the size… On one end of the spectrum are large, multifunctional public buildings, and the design takes around 1.5–2%. On the other end are small, single family houses, where the percentage can reach 10–14%…</blockquote><blockquote>But this is usually related to the total value of the finished building. The structural work also varies, but can go anywhere between 30% and 50% of the total cost.</blockquote><p>That was a good starting point. And a bit surprising one. I felt like in software development, the share of architecture design efforts does not vary that much. I gave it a second thought… And a third one. Then realized how for so many years, I’ve missed an important difference point in the analogy between making a building, and making a software. The potential for <em>effort amortization.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JlYoB14_flhdBPPy" /><figcaption>Photo by <a href="https://unsplash.com/@rawpixel?utm_source=medium&amp;utm_medium=referral">rawpixel</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>In both endeavours the design work can be easily reused — you work hard to think about a floor plan, that suits the whole ten story building. And it works for all of them. With some tweaks, of course. Similarly — you design the modules, libraries and protocols in a software product once for the whole system (<em>ok, it could be twice, or more, sometimes, but again — this is a subject of another article</em>). But, when it comes to actual development/construction — you simply <em>cannot copy &amp; paste a floor</em>… You need to annoyingly and meticulously lay brick, after brick, on a same plan for each of the ten stories, until you have it built. While in software development you <em>can</em> copy and paste (although, you shouldn’t!), and most importantly — you can <em>reuse</em> “constructed” parts many times.</p><p>That explains the big variance in design share for civil construction — processes have different scalability. It also means that a software design architecture part should be closer to that of a single family house scenario. Additionally, since we often see the <em>development</em> part as the “structural work” only, and the content (which is part of the “finished building”) is not part of <em>that</em> calculation — the percentage of the software system architect design should go as high as 20%. Even more, I’d say.</p><p>That is not all. Having each <em>copy</em> (floor in our example) be individually “developed”, also means that it doesn’t pay off overthinking the structure and making it too reusable, because <em>it can’t be</em>. In software design, on the other hand, abstractions and structural reusability potential (if done properly) <em>does</em> pay off. Not to mention, that you don’t suddenly decide to add 3 more stories to a finished building. Civil architecture design finishes <em>at last</em> when the construction ends. In software design it doesn’t, which again — raises the bar for a <em>good, thoughtful </em>design, capable of handling unforeseen expansions smoothly.</p><p>As I was saying… <strong>a lot!</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*folM6dap8j84MUOD" /><figcaption>Photo by <a href="https://unsplash.com/@jbrinkhorst?utm_source=medium&amp;utm_medium=referral">Jonathan Brinkhorst</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1af390b9501f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cars?! No, thankyouverymuch!]]></title>
            <link>https://medium.com/@thejonan/cars-no-thankyouverymuch-df9d24a50786?source=rss-900f1a93f712------2</link>
            <guid isPermaLink="false">https://medium.com/p/df9d24a50786</guid>
            <category><![CDATA[carless-cities]]></category>
            <category><![CDATA[livable-cities]]></category>
            <category><![CDATA[self-driving-cars]]></category>
            <category><![CDATA[lovable-cities]]></category>
            <dc:creator><![CDATA[Ivan (Jonan) Georgiev]]></dc:creator>
            <pubDate>Wed, 21 Feb 2018 22:16:28 GMT</pubDate>
            <atom:updated>2018-02-21T22:16:28.365Z</atom:updated>
            <content:encoded><![CDATA[<iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FwbSwFU6tY1c%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DwbSwFU6tY1c&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FwbSwFU6tY1c%2Fhqdefault.jpg&amp;key=d04bfffea46d4aeda930ec88cc64b87c&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/337928f9ef3e119e2f139570c9cadc6f/href">https://medium.com/media/337928f9ef3e119e2f139570c9cadc6f/href</a></iframe><p>Elon Musk is making history. He’ll be our companion and opponent. Not, being aware of that, of course… On the very same day, when his team launched <a href="https://www.youtube.com/watch?v=wbSwFU6tY1c&amp;t=1803s">Falcon Heavy caring Tesla into the open Space</a>, a <a href="https://www.nytimes.com/2018/02/06/world/middleeast/syria-bombing-damascus-united-nations.html">Syrian government strike, with 80 casualties</a> happened, and some media jumped to compare the two events, implying meaninglessness of the former compared to the harsh reality of the latter. That’s frustrating. Indeed, 80 innocent people dying is a tragedy. What about <strong>3 thousand? Every day! On average!</strong> Throughout the year! And the previous one! <em>And the next one, most probably</em>!!! These are the statistics about road accidents’ fatalities. And, even though 3 thousand it just a number, everyone was someone’s beloved person! No less loved, than those 80 people who died on the day of the Falcon Heavy launch.</p><p>So, I’ve came up with an idea — let’s make a website, and match each of these media-friendly tragedies about plane crashes, terrorism, war, etc., with real cases of people that died on the road somewhere on the Globe, in the very same day!<br>— <em>It’s disgusting!</em>, said a friend of mine — <em>showing so much tragedy in one place!</em> <br>— <em>It’s disgusting!</em>, I totally agreed — having so many people die from same disease, <em>every single day</em>, and nobody talking about that! <strong>Number of people dying on the road per year, is closely shaving the civilians being killed per year, during the World War I¹!</strong> Think about that!</p><p>There’s already a push for solution, they say. Elon Musk, aside from sending cars into space, is also working on making them self-driving. Other big companies too. <em>A humanoid robot, cutting the grass with a scythe</em> — that’s the scene I’ve recalled the very first time, I’ve heard about making cars, as they look now, self-driving. <em>A humanoid robot, cutting the grass with a scythe</em>. Think how <em>ridiculous</em> this is!? If, the technology is so advanced, so you have humanoid robots, you’ll never, ever use scythe! Right? Same with self-driving cars. The problems lie much, much deeper. It’s pure physics. And, I’m not talking about combustion engine efficiency. Not yet.</p><p>I’m talking about mass. The average weight of an automobile found on the road nowadays is nearly 2 tons. <em>Two thousand kilos</em>. Not including trucks and similar in that statistic. On average, <a href="https://en.wikipedia.org/wiki/Energy_efficiency_in_transport">we have approx. 1.6 passengers per car</a>, and average weight is about 75 kilos. That makes around 120 kilos. Let’s add some baggage, make people 2 and round up that to 200 kilos. This means, that even being quite optimistic you have <strong>about 10% useful cargo</strong> in an average car. Now apply the 45% engine efficiency (at most!), and you’ll end up with <strong>less than 5% total efficiency</strong>. <em>Five f***** percent (!)</em> on our most beloved way of transportation. And we haven’t even touched the issues that self-driving capabilities could, theoretically, solve. We, also haven’t discussed the fact this efficiency is achievable, only if certain infrastructure is present — all the highways, streets and boulevards. And they are all expensive, their production is dirty and environmentally damaging, and — most importantly —they occupy vast territories of our land, especially in the most expensive and livable areas — city centers.</p><p>Don’t get me wrong —I admit and admire the role, that the auto-industry has had on the technological growth of our society. I, personally, no matter how embarrassing is to say that, love driving. I appreciate the freedom of movement the cars gave to mankind. But it’s not relevant any more in this form. It’s obsolete. It must change. There are alternatives, and all the technology is here, but that’s a whole another story.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*N9UgEN3pEjjffv5C0_i5yw.jpeg" /><figcaption>Kids’ free riding! (Photo by <a href="https://www.3rd-eye.com/">https://www.3rd-eye.com</a>)</figcaption></figure><p>The real, <em>urgent</em> reason for us to retire automobiles, is the damage they are doing to our society. Mainly — damaging the way our children grow. Everybody, or at least — every parent — knows how important, and fulfilling for our youngsters is to communicate with other kids. To do it on their own, with their own pace. This is almost totally gone. We rarely let our small kids freely go outside, in the fresh air, to play with other kids around. No, don’t tell me, that you drive them to the park and let them play there. Also, don’t tell me you have bicycle lane from you home to the park and they use it. Firstly, vast majority of people, throughout the Globe, <em>don’t</em> have such lanes, secondly I’m talking about much earlier age, before kids can ride a bicycle on their own, and last, but not least — I’m not talking about this full-time, parent-managed, supervised communication with others. I’m talking about natural, hour-per-hour playing and socializing of youngsters. That is gone, and mostly because of the danger that cars impose. Nothing else — not thieves, not robbers, not even sociopaths, are exposing danger, even close to the one from a car accident.</p><p>And raising kids like this, already shows the “nice” results — growing number of socially damaged personas, people not used to being on their own and taking any kind of responsibility for their actions. Because they never grew up unsupervised, with other kids giving them honest and immediate feedback. Because we’ve decided to inhabit our cities with cars.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*av696hFfD4gTvc_UMfjR4g.jpeg" /><figcaption>Naturalistic Allianz Arena (Wikimedia Commons)</figcaption></figure><p>This brings us to the next big topic. How much area do you think is dedicated to cars in our cities? 5%, 10%? The answer is <strong>more than 30%</strong>! In US usually above 40%! Even the regulatory norm of having a garage, which roughly equals two not-so-small sized bedrooms, should ring a bell of alarm. Just think — <strong>if you remove all the boulevards and streets from Munich, </strong>for example<strong>, you’ll free enough space to build more than 500 Allianz arenas²!! </strong>Not, that you’ll be able to get to them, of course.</p><p>So, what’s <strong>the alternative? </strong>Not that radical, actually. After realizing it is ridiculous to have same vehicles for <em>both</em> interstate transportation, which we do rarely and, 98% of the times, using highways; and in-the-city everyday use, <strong>we‘ll be able to see the solution is already here</strong>. We can halve the size of our roads in the city and have only small, electric, golf cart-like vehicles for our daily commute. They could be <em>much, much</em> lighter, which imposes <em>huge</em> benefits — less energy to move them, narrower lanes, <strong>much wider margin of error </strong>— a<strong> </strong>2 tons vehicle has <em>10 times</em> more kinetic energy, compared to 200 kilo golf cart. Crashes will not be deadly, will be much easier to avoid (steering is more or less vectoring the kinetic energy). This <em>opens</em> the door for more self driving innovation, lowering the currently over-regulated entry barrier. What we need for our cities is more like rain protected electrified bikes, rather than highway capable speed bulids, that we currently use.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_-mnqk04tF_jUeV39HGN-Q.jpeg" /><figcaption>Even loading ferries could be fast. (Photo by <a href="http://www.picture-newsletter.com/">http://www.picture-newsletter.com</a>)</figcaption></figure><p>And, how are we going to travel outside the city? Well, for 98% of people <em>currently</em> travelling means driving on highway for couple of hours, until they get to another urban area. This can be done with trains. And, no, it doesn’t need to be slow. Just think about hundreds of ferries which operate around the world. <em>Greeks to it fast, too!</em> And, don’t forget that lightweight, golf cart-like vehicles are, additionally, much easier to put on a train within minutes. Then the annoying multi-hour driving with 150kmh will turn into less-hours, mutli-coffee, or whatever you order into the train’s restaurant, travelling with 200+ kmh.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/655/1*WhHGUHtFY2x_H6THet03UQ.png" /><figcaption>Sending cars to where they belong — far away in the Galaxy… (Space X @ YouTube)</figcaption></figure><p>It’s all here. All the technology, all the possibilities, and plenty of reasons. We just need to add some <em>actions</em>.</p><p><strong><em>[1]</em></strong><em> </em><a href="https://en.wikipedia.org/wiki/World_War_I_casualties"><em>Civilians killed during World War I are around 6mil</em></a><em>, and it lasted for 4 ½ years, which makes around 1.3 deaths per year. </em><a href="https://en.wikipedia.org/wiki/List_of_countries_by_traffic-related_death_rate"><em>According to WHO</em></a><em>, road fatalities are ≈1.23 per year (Cited data is for 2015)! Of course, the population then was nearly 4 times less than it is now.</em></p><p><strong><em>[2]</em></strong><em> Munich </em><a href="https://en.wikipedia.org/wiki/Munich"><em>is said to occupy 310.43m²</em></a><em>, assuming 30% dedicated to road infrastructure, we’ll get 93.129km². The whole </em><a href="https://allianz-arena.com/en/arena/facts/general-information"><em>Allianz arena’s site</em></a><em> is 0.171km².</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=df9d24a50786" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>