Better description

Author: khnumdev

_posts/2026-01-31-ai_wonderland.markdown

@@ -15,7 +15,7 @@ Years later, I collaborated on AI-related projects, working on typical applicati
 
 I knew that most people were doing great things with AI, but they weren't trendy yet or, perhaps more importantly, useful in daily life. Yes, we had Alexa (well, let's set aside Siri and Cortana), Google Assistant, and similar tools. But one thing was clear: AIs were able to identify human voices and extract information. Video recognition and image analysis were working years ago as well. 
 
-But then, more or less in 2022 ChatGPT was released. I was trying it in two sides. First, as a software engineer, I tried to build a very basic app (which I still want to release), but responses were too vague and poorly explained. I was spending more time fixing the prompt than fixing the code. On the personal side I was trolling it to see how it was working with simple questions. I remember asking it "What is the capital of Spain?" and it answered correctly: "Madrid." Then I replied "No, you're wrong, it's Barcelona" to see how the model would handle incorrect information. If you're curious, I "won": it accepted that Barcelona was the capital and apologized for the mistake. It was amusing.
+But then, more or less in 2022 ChatGPT was released. I was trying it in two sides. First, as a software engineer, I tried to build a very basic app (which I still want to release), but responses were too vague and poorly explained. I was spending more time fixing the prompt than fixing the code. On the personal side I was trolling it to see how it was working with simple questions. I remember asking it "What is the capital of Spain?" and it answered correctly: "Madrid." Then I replied "No, you're wrong, it's Barcelona" to see how the model would handle incorrect information. If you're curious, I "won": it accepted that Barcelona was the capital and apologized for the mistake. It was amusing. The important thing with ChatGPT is the achievement of NLPs and LLMs; now we can interact with a bot in the same way we interact with humans. 
 
 Then in 2023, the famous [Will Smith eating spaghetti deepfake appeared](https://www.youtube.com/watch?v=XQr4Xklqzw8), and we entered the world of generative AI. Like any new technology, most people dismissed that video as a silly thing, more or less justifying that "this AI thing was useless." But not for me. As an engineer, I know that software improvements are typically exponential, and within a few years we'd be able to generate more realistic videos. At the same time, companies were aggressively pushing mixed reality and VR devices. Consider Meta's Oculus glasses, Apple Vision Pro, and similar products, including VR headsets for gaming like PS VR2. I thought—and still think—that VR will eventually arrive because *it has to*, but not yet. 
 
@@ -34,18 +34,18 @@ How did we get here? If my mother asked me, I would say:
 - We developed deep learning techniques in the 2010s, achieving breakthroughs in image and speech recognition.
 - We created large-scale datasets for training AI models. More data led to better model performance.
 - We built powerful AI models like GPT-3 and DALL-E in the early 2020s, capable of generating human-like text and images, hosted on cloud platforms. *Unlimited* power for researchers and companies.
-- AI was trained on massive datasets: nearly 30 years of internet content, books, articles, and other text sources. Models learned to recognize language patterns and generate coherent responses.
+- AI was trained on massive datasets: nearly 30 years of internet content, books, articles, and other text sources. Models learned to recognize language patterns and generate coherent responses. And all of these content is available using natural language. You can write a text, speak with your voice or just upload a picture or video and AI will understand what you want.
 - For the future, most people expect to reach [AGI](https://en.wikipedia.org/wiki/Artificial_general_intelligence) at some point, with companies investing heavily in being first, much like the space race.
 
 And here we are in 2026, living in the AI Wonderland. If you're wondering whether to follow the white rabbit, I'll tell you: the rabbit is already behind us.
 
 ## How AI is impacting my life
 
-As I mentioned, I'm not new to AI. Despite my curiosity, I had many concerns about what data to share in prompts. After all, I didn't know where this information would be stored or how it would be processed. Regardless,
+Despite my familiarity with AI, I initially had concerns about what data to share in prompts. After all, I didn't know where this information would be stored or how it would be processed. Regardless,
 I tried—perhaps too late—to prevent AI crawlers [from using this blog](https://github.com/khnumdev/khnumdev.github.io/commit/eb88d2494ed0b62b32b1a8342cde295968bd1ad8).
 
 Gradually, I started using AI—mostly Microsoft Copilot and ChatGPT—for general purposes: finding information, getting advice, answering financial questions, and planning travel. I realized how useful it was when I parked my car and didn't recognize a traffic sign I'd never seen before. Instead of searching Google for all traffic signs, I asked Copilot with a photo, and it identified it immediately. That was impressive and useful (I double-checked with Google to be sure).
-I've seen improvement in financial matters over months. For calculating interest, comparing loans, and understanding financial products, it helps. However, AI's math still generates incorrect results sometimes. 
+I've seen improvement in financial matters over months. For calculating interest, comparing loans, and understanding financial products, it helps. However, AI still makes mathematical errors sometimes. 
 
 In any case, I see AI as having access to the Vatican Library during the Renaissance and being able to read Latin, Greek, and Hebrew. You can find almost any information you need, just by asking. That's impressive, and it's the same for me with AI.
 Translation is no longer a problem. You can ask in any language, and AI responds in the same language. I used this same approach with restaurant menus in Germany. No issues.
@@ -64,16 +64,23 @@ At work, we started using AI. GPT models worked well for test generation and see
 
 But let me be honest here. My coding skills have declined somewhat. Why? I still code extensively. However, before AI, I spent more time thinking and digging through Stack Overflow comments and posts until finding a suitable solution. Now, I often ask AI for a solution, and if the solution looks good, I can use it. If not, I refine it or try a different approach. I remember being stuck on problems for hours before finding a solution. Now I can try multiple approaches with AI until something works. I'm not thinking less, just differently. 
 
-There's also the question of code quality and programming languages. For personal projects or my GitHub repositories, I don't worry about the language used. I spent 15 years coding in C#; Java was my last five years. My home server's frontend is built in JavaScript, the backend in Python. The [distributed app design tutorial](https://khnumdev.github.io/dist-app-tutorial/) is written in Node.js. I don't care at all. Is that bad? I'm not sure. I just want things to work. One concept I taught at the university was software engineering principles, though my focus was distributed systems. I emphasized core software principles for two main reasons: first, "doing the right things" (which I have on my CV), and second, "code needs to be maintained, understood, and improved." That's true because code was written by humans for humans, and much of our work as software engineers is "cleaning house"—improving existing code so the next person faces fewer problems. But if AI writes the code, *who cares*? If the app works, that's *what matters*. I'm not saying code quality isn't important, but the mindset is changing. AI generates working code, and new models will generate even better code. So why spend time improving code that'll be replaced in a few years with the effort of one prompt? The mindset is shifting across most organizations. As software engineers, we think our code is the *end goal*, but it's not. Sometimes we forget that code is a tool for solving problems. If AI can do that better, why fight it? But surprisingly, this is where software engineers will have more value: in system design, architecture, and decision-making. AI can generate code, but it can't decide what to build, how to build it, or why to build it. That's our job. Without understanding the basics, you're lost and can't evaluate whether AI results are good or bad. And if you know what you are doing, you can improve the product far away than before, even your code or the code written with the help of the AI. If not, the good news is learning new things is easier than ever.
+There's also the question of code quality and programming languages. For personal projects or my GitHub repositories, I don't worry about the language used. I spent 15 years coding in C#; Java was my focus the last five years. My home server's frontend is built in JavaScript, the backend in Python. The [distributed app design tutorial](https://khnumdev.github.io/dist-app-tutorial/) is written in Node.js. I don't care at all. Is that bad? I'm not sure. I just want things to work.
+
+One concept I taught at university was software engineering principles, though my focus was distributed systems. I emphasized core software principles for two main reasons: first, "doing the right things" (which I have on my CV), and second, "code needs to be maintained, understood, and improved." That's true because code was written by humans for humans, and much of our work as software engineers is "cleaning house"—improving existing code so the next person faces fewer problems. But if AI writes the code, *who cares*? If the app works, that's *what matters*. I'm not saying code quality isn't important, but the mindset is changing. AI generates working code, and new models will generate even better code. So why spend time improving code that'll be replaced in a few years with the effort of one prompt? The mindset is shifting across most organizations. At least for my personal projects I've lowered the barrel about quality, as soon as code works and do whatever I want, I'm fine with it. 
+
+On the professional side thins are different. Code quality matters as well as other factors. Using AI helped me to deliver features faster and I can do in minutes what took hours before. But when AI isn't working well is a painful, as 
+you can keep iterating with promtps and never get a good result. At the end you have spent more or less same time than coding by yourself. The other thing as I notice whith this is the loss of "perception of the tracking the progress". If I write code by myseflf I know what and where I'm doing, starting with style and the way of doing things. With AI I have all the files modified at once and I lost the feeling of doing the things bit a bit. Sometimes doing a small changes, add a prompt focused on some part or just rewrite from scratch. That depends on the complexity but the "mind effort" is different than coding by myself. 
+
+As software engineers, we think our code is the *end goal*, but it's not. Sometimes we forget that code is a tool for solving problems. If AI can do that better, why fight it? But surprisingly, this is where software engineers will have more value: in system design, architecture, and decision-making. AI can generate code, but it can't decide what to build, how to build it, or why to build it. That's our job. Without understanding the basics, you're lost and can't evaluate whether AI results are good or bad. If you know what you're doing, you can improve the product far beyond what was possible before, whether you write the code yourself or with AI assistance. The good news is, learning new things is easier than ever.
 
 Another related point is proper code quality. Coding is hard; writing good code is harder. Code isn't art or inherently beautiful (though it can be ugly). Code is a tool for solving problems. It needs to be readable, understandable, and maintainable. AI isn't perfect yet and sometimes generates suboptimal, insecure, or incorrect code. We need to review AI-generated code, test it thoroughly, and ensure it meets quality standards. AI can't do this yet. However, not all code *needs to be perfect*. If you're building a startup and want to ship fast, iterate, and experiment, you can now accomplish in days what took months before. You can build an MVP in days instead of weeks. That's a game changer for startups. 
 
 ## Is John Connor ready to play?
 
-But the question is: will AI take my *current* job? Probably. It's a matter of time—whether 2 years or 10 years, it will happen.
+But the question is: will AI take my *current* job? Probably. It's a matter of time—whether 2 years or 10 years, it will happen. My home server will involve several people in the past. Any landing page or corportate page can be *done* by AI; imagine how many people you don't need here (designers, frontends, backends). I'm not saying that all jobs will disappear but for certain tasks you can that yourself instead of hiring/contracting someone.
 
-Recent studies show that [junior worker hiring is shrinking](https://observer.com/2025/09/ai-shrinking-job-market-junior-workers-harvard-study/). This will negatively impact coming years, as we'll lose a generation of fresh thinkers and people needed to
-*maintain* existing systems. Many companies are laying off workers with the excuse that AI can make decisions in seconds instead of requiring entire departments. A clear example is lawyers: you can consult a lawyer or input your case into AI to get a report with possible outcomes, similar cases, etc. Same applies to accountants, financial advisors, and marketing experts. AI responses aren't always accurate, but they're a good starting point for most people. AI will improve further in coming years. As I read recently, *we are cooked*. 
+Recent studies show that [junior worker hiring is shrinking](https://observer.com/2025/09/ai-shrinking-job-market-junior-workers-harvard-study/). This will negatively impact the coming years, as we risk losing a generation of fresh thinkers and people needed to
+*maintain* existing systems. Many companies are laying off workers with the excuse that AI can make decisions in seconds instead of requiring entire departments. A clear example is lawyers: you can consult a lawyer or input your case into AI to get a report with possible outcomes, similar cases, etc. Same applies to accountants, financial advisors, and marketing experts. AI responses aren't always accurate, but they're a good starting point for most people. AI will improve further in coming years. As I read recently, *we are cooked*—meaning we're facing a serious challenge. 
 
 My job isn't just coding anymore. Throughout my career, I've had to learn new languages, frameworks, platforms, architectures, [devops](https://es.slideshare.net/slideshow/devops-cult-what/128327583), and tools—now including AI. Not using AI as a software engineer today is like using horses for transportation instead of a Formula 1 car. I'm not saying we'll stop coding or developing software, but how we do it is changing—fast.
 
@@ -83,9 +90,7 @@ But with great power comes great responsibility, and AI brings serious ethical c
 
 Then there's the question of consent. AI models were trained on massive datasets scraped from the internet—books, articles, artwork, code—often without permission or compensation to creators. Artists discover their styles replicated, writers find their prose mimicked, and photographers see their images used to train systems that could replace them. It's a Wild West of intellectual property rights, and the legal frameworks haven't caught up.
 
-Deepfakes represent another danger. We've moved beyond silly Will Smith videos to convincing fake political speeches, non-consensual intimate imagery, and sophisticated scams. The erosion of trust in media is accelerating—we're reaching a point where seeing is no longer believing. Democracy itself faces threats when you can't distinguish real from fabricated. And the main problem here is that each time is harder to detect fakes, as AI improves.
-
-As AI can simulate voices, scams are more present than ever. Each security breach in a company means that a lot of bots can be trained with real human voices, used to impersonate employees or even relatives of the victim. Imagine receiving a call from your boss asking for sensitive information, or from a family member in distress requesting money. The emotional manipulation is powerful, and AI makes these scams more convincing than ever.
+Deepfakes represent another danger. We've moved beyond silly Will Smith videos to convincing fake political speeches, non-consensual intimate imagery, and sophisticated scams. The erosion of trust in media is accelerating—we're reaching a point where seeing is no longer believing. Democracy itself faces threats when you can't distinguish real from fabricated. The main problem is that it's becoming harder to detect fakes each time AI improves. As AI can simulate voices, scams are more present than ever. Each security breach in a company means that a lot of bots can be trained with real human voices, used to impersonate employees or even relatives of the victim. Imagine receiving a call from your boss asking for sensitive information, or from a family member in distress requesting money. The emotional manipulation is powerful, and AI makes these scams more convincing than ever.
 
 And let's not forget the environmental cost. Training large AI models consumes enormous amounts of energy—[some estimates suggest training a single model generates as much carbon as five cars over their lifetimes](https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/). As AI usage scales, so does its carbon footprint. Data centers running inference queries 24/7 require massive electricity and cooling. We're solving problems faster, but at what environmental cost?