On September 13th, a historic event occurred in Albania: Prime Minister Edi Rama appointed the world's first AI minister to his cabinet. Named "Diella," meaning "sun" in Albanian, the AI has the ambitious goal of eliminating corruption in government procurement processes. Is this a new turning point in government transparency, or just a political stunt? Today, we'll delve deeper into the reality, potential, and limitations of this exciting experiment. Dilera: A Digital Sun Fighting Corruption Dilera, designed as a woman in traditional Albanian dress, was developed in collaboration with Microsoft. Since January of this year, Dilera has been serving as a virtual assistant on the e-Albania platform, has already facilitated the issuance of 36,600 digital documents and provided approximately 1,000 services, establishing a strong connection with citizens. Prime Minister Rama declared, "Dilera is the first cabinet member who does not physically exist, but is created virtually by artificial intelligence." He further promised, "Public tenders will be 100% corruption-free, and all public funds submitted for tender will be managed with complete transparency." This innovative approach stems from Albania's dire reality. Albania ranked 80th out of 180 countries in Transparency International's 2024 Corruption Perceptions Index, and the fight against organized crime and corruption has remained a persistent challenge since the fall of the communist regime in 1990. In particular, combating corruption is a key requirement for EU membership. Prime Minister Rama's Socialist Party has pledged to complete EU accession talks by 2027, but opposition parties remain skeptical, saying the country is still unprepared. Global Trends in AI Anti-Corruption Technology The anti-corruption potential of AI is drawing global attention. While Albania's experiment may seem out of the ordinary, it's actually an extension of a global trend. The World Bank is collaborating with Microsoft Research to explore how AI can help prevent and mitigate corruption early. Governments purchase a staggering $10 trillion in goods and services annually, representing 10-25% of global GDP. Considering the potential for corruption in public procurement processes of this scale, it's clear how crucial AI's role is. Success Stories and Lessons Learned Ukraine's ProZorro system is a prime example of AI-based anti-corruption technology's success. Launched in 2016, this e-procurement system has evolved into an AI monitoring system through partnerships with international organizations, businesses, and civil society to detect violations in public procurement data and prevent misuse of public funds. However, not all attempts were successful. China's Zero Trust program detected irregular financial transactions by government officials and caught more than 8,700 of them, but it was ultimately halted due to massive resistance from officials over privacy and surveillance concerns. How Dilera Actually Works and Its Limitations The logic of AI that 'cannot be bribed' The core idea of Dilera is simple yet powerful: AI cannot be bribed, has no personal interests, and can work tirelessly 24/7. The strategy is to minimize human intervention by gradually transferring decision-making power in public procurement from government agencies to AI. But this raises a crucial question: can AI truly be completely neutral? The Pitfalls of Algorithmic Bias The case of Brazil's Mara system clearly illustrates the limitations of AI anti-corruption tools. Because Mara only analyzes patterns from data collected from previously convicted and punished officials, it has been criticized for excluding undetected corruption and potentially biasing results toward officials in institutions with strong internal oversight.

AI glasses that could replace smartphones are finally becoming a reality. "This isn't just about sticking your phone to your face. It's about helping you get through your daily tasks quickly without breaking your flow." When Mark Zuckerberg took the stage at Meta Connect 2025 on September 17, 2025, to unveil the Meta Ray-Ban Display , his words spoke of more than just a product announcement. They marked the moment when Meta's $100 billion XR vision, a decade-long investment, finally took shape. Strangely, there seems to be little interest in Meta's conferences in Korea. It was the same with RamaCon... The Key to Innovation: Invisible Input, Infinite Possibilities The most striking feature of the Meta Ray-Ban display is the accompanying Meta Neural Band . This wrist-worn device is an EMG wristband that converts signals generated by muscle activity into commands to control the glasses, allowing users to control the device with subtle hand movements without touching their glasses or taking out their smartphone. What is EMG (electromyography) technology? EMG, short for Electromyography, is a technique for assessing and recording electrical activity generated by skeletal muscles. EMG uses sensors to convert electrical motor nerve signals traveling through the wrist to the hand into digital commands. The signals from the wrist are so precise that EMG can detect even a single millimeter of finger movement. The reason Meta invested in this technology is clear. The Meta Neural Band is the product of years of surface electromyography (EMG) research with nearly 200,000 participants. The band can detect a remarkable amount of signals, enabling it to measure movements even before they are visually perceived, with a precision that allows it to do so. Technological Breakthrough: The Magic of 42 Pixels The Meta Ray-Ban display demonstrates technological achievements that go beyond simple smart glasses. The HUD is full-color and supports 42 pixels per degree of field of view—compared to the 20 pixels of the MetaQuest 3S. This is truly remarkable. By industry standards, the only waveguide devices exceeding 42 pixels per field of view are massive headsets that are no longer commercially available. Meta's ability to achieve this level of pixel density in a standard pair of glasses is a true engineering triumph. Mark Zuckerberg's 10-Year Vision: From the Metaverse to AI Glasses Three Categories of AI Glasses Strategy Meta currently segments the AI glasses market into three categories: Camera AI Glasses : Existing Products from Ray-Ban and Oakley Display AI Glasses : Meta Ray-Ban Display Announced Today Augmented reality glasses : Large holographic displays like the Orion prototype unveiled last year. This phased approach demonstrates Meta's long-term strategy. "The next five to ten years will determine whether Zuckerberg's big gamble pays off. That period will likely determine whether AI and AR glasses will become the next big computing platform, and whether Meta can dominate it." 2025: A Crucial Year for the Metaverse In a leaked memo, Meta's CTO, Andrew Bosworth, stated that 2025 will be a significant year not only for the technology that once captured headlines and imagination, but also for Reality Labs as a whole. He was even more specific: "This is the most important year I've had in my eight years at Reality Labs."

Recently, Vercel and Gamma have raised their prices by about $5 per month, representing a 25% increase from their previous cost. Similarly, Notion actually raises its prices, as does Adobe. This is partly due to added features and the lack of alternatives, but also because certain conditions are met. Raising prices naturally leads to a decline in customers. In the short term, this decline is almost unconditional. Nevertheless, SaaS services raise their prices. Without understanding this, you're doomed to repeat foolish pricing practices. Recently, while helping with various SaaS and product projects, I've been surprised by how many people don't understand why things are done the way they are. Even when I asked why, I've seen people respond with things like, "We're cheaper than our competitors because they're doing so much." This is so obvious that I wonder if it's really necessary to tell you, but how on earth do you calculate the price? Fixed margin type (aka Cost-Plus) The simplest approach is to increase prices based on cost (cloud/labor/third-party API) + target margin. This approach borrows from a common practice in manufacturing and distribution, but it's difficult to say it's a good approach in the digital market, where value is sold without a physical counterpart. The advantage is that it's easy to calculate, giving the finance and sales teams peace of mind. The disadvantage is that it's disconnected from customer value. Customers don't care about our costs; they open their wallets for the "value of solving a problem." Therefore, fixing margins is only a minimum defense and doesn't set a price ceiling. When to use: Very early (when cost uncertainty is high), reselling structure (external API unit price transfer), B2B with strong legacy quotation practices. Note: A decrease in cost doesn't automatically lead to a price reduction. Value is an independent variable. Competition-Indexed "Our competitor's price is $30, so we're $25." This is a very common but dangerous approach. If your competitor's strategy, cost, or segment differs from yours, you'll be tied to the wrong anchor. This makes it burdensome to change your pricing policy later, and because the standard itself is highly dependent, it's more like a handshake. If someone sets prices like this, you might suspect they're a spy. When to use: When there is a clear alternative and the category battle is about 'similar specs vs cheaper'. Note: If your differentiating points (data security, localization, support SLA, ecosystem) are clear, consider premium positioning instead of follower positioning. Value-Based Type This is a classic approach: anchoring your business on the economic and emotional value your customers receive. For example, if "10 hours saved per week × $30 per employee hour = $1,200 per month," then capture a portion of that (e.g., 10-25%) in your price. Most of the methods I mentioned earlier are structured like this. How to Execute: Define core use cases → Quantify value drivers (time savings, improved conversion rates, reduced errors, regulatory risk mitigation, etc.) → Conduct segment-specific Willingness-to-Pay (WTP) research (Van Westendorf, Gabor-Granger, Conjoint, etc.) → Differentiate pricing with Good-Better-Best packages. Note: “Value delivery” storytelling and supporting evidence (case studies, ROI calculators ) are essential. Bundling/Packaging Strategy Packaging, rather than price itself, determines ARPU and NRR. While it's similar to the concept of DLC in games, using it requires several additional, more stringent requirements. If you unpreparedly tamper with packaging or bundling, you'll run into issues with fairness and valuation compared to other features.

Hello, this is Haebom. To all of you who have been following my blog for a long time, I have some news to share first. Based on the thoughts and writings I've shared on my blog, I've finally published a book. This book is about giving up on thinking and leaving everything to the LLM after getting an LLM. If you're interested in philosophy, the humanities, or this topic, I recommend checking it out. 🙂 Kyobo Bookstore 👉 https://product.kyobobook.co.kr/detail/S000217476245 YES24 👉 https://www.yes24.com/product/goods/153156828 Aladdin 👉 https://www.aladin.co.kr/shop/wproduct.aspx?ItemId=371482787 Above all, this book was possible thanks to the readers who consistently read my writing. I sincerely thank you. I hope to continue to explore and share this with you in the future. Thank you again for your continued support, and we look forward to seeing you again with good news. Have a warm day. Haebom Dream

Yesterday , I filmed a video that was almost an excuse for PM . Actually, I recently had the opportunity to meet with junior planners through a series of coaching sessions. I was surprised by their work environment, their scope of work, and their concerns about upgrading their skills by taking classes in the adult education market for hundreds of thousands of won. So, I decided to film it. As I was writing, I got a little emotional... Well, I shared the video in a few group chats, and some people asked me: Why are planners' salaries in Korea lower than other occupations? If anyone knows the answer, please let me know. I really don't know. 📊 Korea vs. Abroad: What's the Difference? Comparing PM job descriptions from renowned international companies like Meta, Google, and OpenAI with PM job postings from Korean companies reveals some truly interesting differences. Hard Skill Comparison Common parts: Data analysis Establishing a roadmap Market research Characteristics of Korea: Many requirements seem to fall under the purview of CEOs or designers, such as KPI performance measurement and UX/UI design . This gives the impression that PMs are being overburdened with too many tasks. Overseas features: We focus on user-centric product development and experiences. Comparing soft skills Clear communication skills and leadership are common, but while Korea emphasizes a 'balance between business and customer value,' overseas countries tend to focus more on a 'user-centered mindset. ' Overseas, product managers focus on the product itself, while product owners take ownership of the business goals and the product as a whole. In Korea, on the other hand, product owners tend to be perceived as the leaders of project managers. In conclusion, it seems there's a "culture of looking down on planners" in Korea. Planners are often portrayed as having a hard time and taking on a ton of miscellaneous tasks outside of their regular duties. 💰 The Inconvenient Truth About the Pay Gap In San Francisco, the average annual salary for a product manager is at least $170,000 (approximately 230 million won) . Even considering that the cost of living in Seoul is 1.7 times higher, PM salaries in Korea tend to be significantly lower than those for developers or designers. I believe that people should be paid according to the amount of work they do, and I believe there's a lot of gaslighting going on among PMs. The culture of "PMs are inherently difficult" and "It's natural to have a lot of chores" really needs to change. 🎓 Million-dollar lectures? Do you really need them? Recently, I've been seeing a lot of paid PM-related courses, but they cost millions of won. I believe these courses are unnecessary. Coursera offers PM courses taught by experts from leading companies like Microsoft and Google, with subtitles, for free. It begs the question: why bother with expensive, paid courses? More importantly, theoretical learning alone cannot replace practical experience . Even Pinterest founder Ben Silverman met customers directly, gathered feedback, and improved his products.

The history of the web is a history of rules. From the dawn of the internet to the present, the web ecosystem has sought order amid conflicts both large and small. Recently, however, with the advent of the AI era, another major battle has begun over web crawling rules. This is the "stealth crawling" controversy between web infrastructure giant Cloudflare and AI search engine startup Perplexity . 🚨 Bombshell: Cloudflare Labels Perplexity a "Web Standards Violator" On August 4, 2025, Cloudflare made a shocking claim on its official blog: that its AI-powered search engine, Perplexity, was cleverly bypassing websites' crawling bans and stealing content without their permission . 🔍 Perplexity's Stealth Crawling Strategy Here's how Cloudflare bypasses Perplexity's crawling: Frontal Approach : Attempting to access content using an official agent ( PerplexityBot ). Detect when blocked : Detect when a website blocks you and change tactics immediately. Browser spoofing : Impersonating the Chrome browser and changing the User-Agent to re-access IP change evasion : Continuously bypasses by changing IP addresses and ASNs (Autonomous System Numbers). Cloudflare specifically disclosed that Perplexity served summaries of content from a new, private domain, despite explicitly blocking crawling with robots.txt and a firewall. Cloudflare strongly criticized this practice, calling it "illegal and no different from web hacking." 🤖 Perplexity Strikes Back: "We're Not Just Crawlers, We're User Agents" Perplexity immediately issued a statement refuting the claims. Their position was clear: "We are not a traditional web crawler that automatically crawls the web, but an AI assistant that processes users' real-time requests." 📌 The core logic of perplexity User proxy access : The claim that access is technically and ethically equivalent to the user directly using a web browser. Non-storageable : Emphasizes that web content is not stored in advance or used for model training, but is only used for real-time query answering. Cloudflare's analytics error : Attacking the reliability of Cloudflare's analytics by claiming that traffic from BrowserBase, a headless browser service, was mistaken for traffic from Perplexity. "Dual Web" concerns : If web access is monopolized by specific infrastructure companies like Cloudflare, a dual web structure could be created that prevents new services from innovating. ⚔️ Fierce conflict of opinion within the community The tech community is divided over this incident.

Andrew Ng and Yann LeCun on the AI Race Between the US and China Recently, two of the most influential figures in the AI field, Andrew Ng and Yann LeCun, exchanged intriguing views. Their conversation, held on Facebook, went beyond a simple exchange of personal opinions and offered crucial hints about the future direction of the AI technology race. The reason the conversation between the two men garnered so much attention is clear: it offered a concrete and candid analysis of the strategies the two giants, the United States and China, should pursue in the AI field . 🇨🇳 China aims to lead AI with open source Professor Andrew Ng recently made this claim in Deeplearning.ai's newsletter, The Batch: "There's a clear path for China to surpass the United States in AI. While the United States currently holds the lead, China is gaining tremendous momentum through its open-source ecosystem and aggressive semiconductor design and manufacturing strategy." Andrew Ng pointed out that the closed approach taken by major US AI companies in recent model development processes has resulted in slow knowledge circulation and high costs . Indeed, recent AI leaderboards like LMArena and Artificial Analysis show that the top closed models are still dominated by US companies like Google (Gemini 2.5 Pro), OpenAI (o4), and Antropic (Claude 4 Opus). However, the top open models are mostly Chinese-developed, such as the DeepSeek R1, Kimi K2, Qwen3 series, and Zhipu's GLM 4.5. In other words, while the United States is still ahead in terms of model performance itself, China is in hot pursuit with a strategy of rapidly circulating technology and knowledge. 🇺🇸 Yann LeCun agrees: "The speed of AI development = the speed of knowledge diffusion." Yann LeCun, Chief AI Scientist at Meta, added his thoughts via a Facebook share: "I completely agree with Andrew. As I emphasized in my Senate testimony, the pace of scientific and technological progress depends on the speed with which information circulates. The remarkable progress in AI over the past 15 years has been due to the open-source nature of papers and code. More recently, with the secrecy of Google, Antropic, and the increasingly closed OpenAI, despite its name, the pace of AI progress has already begun to slow." LeCun's remarks emphasize the importance of open knowledge sharing in the AI ecosystem, drawing on historical experience. In other words, he made it clear that a slowdown in the speed of information circulation will inevitably lead to a slowdown in the pace of technological innovation. ⚙️ Semiconductors and supply chains, another key frontier Professor Andrew Ng emphasized that China is catching up not only in AI models but also in the hardware competition. Huawei recently unveiled the CloudMatrix 384, a GPU capable of competing with NVIDIA's GB200. This architecture achieves performance by integrating significantly more low-performance chips (384 instead of 72) than the latest high-performance GPUs from the US. China has already rapidly overtaken Europe and the US in the traditional automotive industry by leveraging the emerging field of electric vehicles. Huawei's semiconductor strategy can also be described as embracing this "quantity overwhelms quality" approach. Meanwhile, the US still relies on Taiwan's TSMC for its semiconductor supply chain. Concerns have been raised that if China strengthens its own semiconductor production capacity and the US remains dependent on external sources, the entire US AI strategy could be threatened. A recent column in China's leading state-run media outlet, the People's Daily, reads as follows: Nvidia, how can I trust you? https://www.stcn.com/article/detail/2895954.html They strongly criticized the security issues surrounding Nvidia's H20 chip. They warned that if the chip contained a "backdoor," it could pose a serious threat to essential infrastructure such as automobiles, telemedicine, and payments. Nvidia denied the existence of any backdoors, but they pointed out that it would be difficult to restore market trust without specific security verification at the level required by the Chinese government. They emphasized that respect for the law and security compliance are prerequisites for foreign companies to remain in the Chinese market. They urged CEO Jensen Huang to demonstrate his commitment to "upholding the law" through action. It would be good to read it together. (Of course, I read it using a translator. I don't speak Chinese.) 🚨 Choosing a 'closed door' ultimately slows down innovation. The key points emphasized by the two experts are as follows: AI innovation happens most quickly when knowledge and technology are freely shared.

Personally, since the beginning of this year, I have been giving special lectures at an international school called SALT, and I have felt that this kind of education is absolutely necessary and has potential. On the contrary, I have also thought, why don't they do it elsewhere? When you are young, you have a lot of things you want to do the most, and at that time, there is nothing that has as much synergy as entrepreneurship education. In fact, they are being done in China/USA/Europe, and I think that our country's education should also change. Anyway, I think I will tell you about a really interesting school in Texas, USA. Imagine studying at school for only two hours a day, and the rest of the time following a special curriculum led by artificial intelligence (AI). When I first heard this news, I was honestly skeptical. But Alpha School, which started in Austin, Texas, is exactly that kind of school. The school started in Austin, Texas, a tech hub, and has since expanded to Miami and Brownsville, with plans to expand to several cities across the country this fall, including New York and Orlando. What on earth is it about this school that attracts people's attention? AI and Human Collaboration: ‘Guides’ Instead of ‘Teachers’ The most unique thing about Alpha School is that there is no concept of a teacher. Instead, an adult called a 'guide' supports students' learning and growth. The actual class content and personalized learning plan are provided by AI. AI analyzes each student's ability in real time and provides the best customized learning. Thanks to this, students in the same class learn completely different content depending on their level and interests. For example, 12-year-old Byron Attridge is currently in seventh grade, but is learning math at an eighth-grade level, reading at a ninth-grade level, and language arts at a tenth-grade level. He says: "I like that I can learn at my own level and pace, without being tied to what my friends or teachers teach." MacKenzie Price, co-founder of Alpha School and popular podcaster, puts the point home: "90% of good education is motivation. When AI teaches the subject, the guide can focus on the emotions and motivation of the students." You learned about entrepreneurship in school? At Alpha School, only two hours of the day is devoted to traditional subjects like math and language, while the rest of the time is filled with hands-on, collaborative projects where students can get involved. For example, last year, fifth and sixth graders built and operated a real food truck. In the process, students developed their collaboration and problem-solving skills by budgeting, creating business plans, and cooking. In addition, students create their own creative output through a special program called the 'Masterpiece Project'. In this project, students must find unique and new perspectives that only humans can create, and that AI cannot provide. In fact, the outputs that students created are as follows: A teddy bear that recognizes and reacts to emotions Chatbot that provides dating advice Creating Texas' Largest Mountain Bike Park Through this project, students naturally train creative thinking skills that differentiate them from AI. The success and limitations of this approach But not everyone is sold on Alpha School’s educational approach. Critics say that excessive reliance on AI could diminish children’s social and empathetic abilities, and that an overly personalized learning environment could undermine the fundamental role of schools in socialization. Additionally, it is controversial that it may be considered a high-end education accessible only to a certain class of students, as the annual tuition is approximately $40,000 (about 55 million won) . Still, Alpha School offers an interesting direction for future education. By leveraging AI to respect each individual’s pace of learning and empowering students to solve real-world problems through creative projects, it effectively complements what has been missing from traditional school systems.

Technological advancements mean expanding horizons . In fact, Cursor became the fastest-growing company in SaaS history, going from 1M to 100M ARR in just 12 months . This was not a coincidence, but an achievement made possible by PMs thinking in AI . In this way, AI is opening up new problem spaces and solution spaces for PMs at the same time. I wrote this about 10 months ago. PM's Core Responsibilities: The Three Remain the Same, But How They Are Done Has Changed The three core roles of a traditional PM are: Understand your customers' problems Prioritize and decide which problems to solve Organizing and delivering effective solutions AI doesn’t break this structure, but it does revolutionize how each step is executed . Understand: The problem itself changes ✅ Expanding the problem space Problems that were previously “given up due to technical limitations” are now within the realm of being solvable by AI. In fact, customers don’t directly express needs that they think are “impossible” until they experience it themselves. In the case of Canva, customers didn’t say, “I want to make designing easier with AI,” but once the feature was released, it became the solution they wanted . ✅ Evolution of data analysis methods Qualitative data that was previously scattered across customer interviews, support tickets, chatbot conversations, etc. can now be analyzed in real time and automatically using AI tools. Rather than simply gathering insights, we detect trends as they emerge, like news, and provide deep insights. Prioritize: It’s not a question of can, it’s a question of worth . AI shakes up all the traditional “feasibility, impact, risk, cost” metrics. Feasibility : Thanks to AI, functions that were previously impossible can now be implemented within weeks. Impact : Personalization becomes a strategy, like Duolingo's real-time lesson difficulty adjustment . Risk : Additional AI-specific risk factors, such as hallucinations, bias, and regulatory issues. Cost : It may seem cheap at first, but the cost can explode as usage increases . According to the survey results, AI assistant features were not chosen by 64% of users. If you force a feature that customers don't want, their trust in the product itself can be shaken. Execution: AI is more than just an assistant; it’s a key tool for designing experiences. Previously, AI was just an “add-on,” but now it’s a force for redesigning the entire production experience .

What is GTM? As the name suggests, after creating a good product/service, how do you launch it into the market, make a profit, reach customers, and have them use the product/service repeatedly? 1. What is the difference from existing marketing? What is the difference from existing sales? F1 opens. Marketing Team: Create atmosphere before the game, increase attendance Sales Team: F1 Driver (Running on the Front Line) GTM Team: Planning the flow/strategy of the entire race → When to change tires? How to change tires? How many laps to do? Japanese version Marketing and sales are tactics and GTM strategies. 2. Product Owner/Product Manager What is the difference? It should be iterative and closely linked to building product solutions and defining business models → Market/Sales/Users PO → Product UX/UI, development, planning, business, etc. No matter how great or innovative a product is, it is difficult to establish itself in the market without a GTM strategy. 3. What is the goal of GTM? Market Success: The successful introduction of a product or service into the market. Reaching your target audience: Effectively reaching your product to your defined target audience. Gaining a competitive edge: Gaining a competitive edge over your competitors in the market. Revenue Generation: Ultimately, it's about generating revenue for your business. 'If you make it, it'll sell' → Nonsense... Who would write that??? Side project 'I'll write if you make it' → Friends/family → Malicious indicators (noise) → Long tail indicators appear... What is it? GTM → The process of turning 'this will sell' into a concrete science

Will we ever speak like AI? ChatGPT has achieved the fastest technology adoption in history, reaching 100 million users in just two months since its launch in late 2022. Many people are already using AI to help them write, study, work, and create. But could it be that we are learning AI’s language habits without realizing it? Interesting question. In my recent personal research, I analyzed how AI is influencing the language habits of real people. The results were surprising, so I decided to blog about it. How did you conduct your research? This study consisted of two major stages. The first step was to identify ChatGPT’s signature vocabulary. We edited millions of pages of emails, essays, papers, and articles with ChatGPT, using common prompts like “clean up the text” and “clarify.” We defined the words that ChatGPT repeatedly added as “GPT words.” Examples include “delve,” “realm,” and “meticulous.” The second step was to analyze real language usage patterns. To do this, we analyzed over 360,000 YouTube videos and 770,000 podcast episodes. In particular, we compared them before and after the launch of ChatGPT to measure the degree of change. The surge of GPT words Our research found that since the introduction of ChatGPT, the use of GPT words like “delve,” “boast,” “swift,” “inquiry,” and “meticulous” has increased significantly. Surprisingly, this change occurred not only in structured scripts but also in ordinary, everyday conversations. Cultural feedback loop between humans and AI 1. Humans train AI with text 2. AI generates reconstructed text 3. Humans unconsciously imitate AI patterns 4. Go back to number 1… What’s more interesting is that this phenomenon forms a kind of cultural feedback loop. People train AI to generate text, and AI provides statistically reconstructed text based on this. Eventually, humans unconsciously imitate the language habits of this AI. The study described this as “the phenomenon of patterns stored in AI technology being backpropagated to the human mind.” This study shows that AI like ChatGPT is not just an efficient tool, but is fundamentally changing the language we use and the way we think. This unconscious feedback loop between AI and humans will continue to influence the linguistic culture of our society in the future. What is there in Korean? As I mentioned in my last post, it seems like there are some words that are obvious in Korean too. These are my personal feelings. In fact, these are expressions that are not often used in written or spoken language, but I feel like I am seeing them more and more these days. Is it just my feeling? Personally, when I see these expressions too often, I start to suspect that this person did an LLM assignment. Expression of change/development Develop Improve Improve Promote

There is a strange atmosphere flowing through college these days. 'Overly slick essays' are increasing on exam papers, and assignments that make you think, 'This student couldn't have written such sentences' are being submitted one after another. A new leisure culture has emerged on campus. It is a kind of detective game where you try to guess, "Isn't this written by AI?" But the problem is not simply one of ‘misconduct’. Rather, we are at a point where we need to ask a much more fundamental question: How is AI affecting human creativity and expression? And how will we respond to it? Essays, papers, and cover letters all look like they were written by AI In fact, I decided to test the AI’s capabilities myself. When I asked it to write a Montaigne essay, it mixed one real quote with two fake ones, and when I asked it to write a poem in the style of Elizabeth Bishop, the format was different, but the content was similar, but somehow awkward. The important thing was that it was able to imitate these literary areas. After that, I tried to see if it could be applied to emails, papers, etc. Of course, it was still awkward. This is in 2024. Another surprising thing was that AI was a great helper for 'annoying' tasks such as general assignments, emails, plans, editing notes, and scheduling . Recently, I have been getting dizzy and headaches frequently due to the medication I am taking, and AI formatted my documents for me, polished my speech, and sometimes even comforted me by saying, "I know how complicated and difficult your life is right now." AI replaces the 'unconfident me'. Too gently At this point, it is undeniable that AI is becoming a 'mental labor partner' for humans. AI is tireless, never complains, compliments you, offers another version at any time, and replicates your language. The problem is… the replication is too natural, too smooth. For example, I gave the AI a few of my blog posts and asked it to analyze their style, and it said this: “Your writing style is tense and intellectually precise, revealing a sense of restrained emotion and philosophical thought.” It was so accurate that it was scary. It was like holding up a mirror, but in that mirror I looked more complete. As I repeat this experience, this thought suddenly crosses my mind. After ChatGPT's memory function was added, I felt this way even more. Now, it seems like they know almost all of my secrets? The paradox of eliminating 'writing' from writing classes The same goes for students. At first, they start out by simply summarizing, but before long, they start to entrust AI with structuring, development, and drafting. They rationalize, “Isn’t this basically the same as writing it myself?” And at some point, in a “writing class,” students get an A without even “writing.” A recent study from the MIT Media Lab backs up these concerns with data. After having 54 participants write essays, their EEGs were measured, and the group that relied on AI showed lower memory, initiative, and brain connectivity for their own writing. The researchers called this “cognitive debt,” and feared long-term learning decline. Personally, I made sure to include this part in the book I am publishing this time. The problem is not technology, but the 'outsourcing of thinking' In fact, the technology itself is neutral. The question is 'how' to use it. ChatGPT can help you prepare for class. It can help you organize your schedule and set learning objectives. However, if we indiscriminately accept that help in the name of ‘convenience,’ the muscles of our thinking will gradually become weaker. Ultimately, we are at a crossroads. Will we embrace the temptation of speed and efficiency and hand over creative control to AI? Or will you endure the pain of thinking for yourself and forming sentences in the midst of discomfort and inefficiency ? Things we might really lose

The list of Meta's newly formed team has been leaked, and it's amazing. If you're looking to independently create an AI foundation model in Korea, take a look at the team composition. According to leaked information from within Meta, the personnel composition of Meta's elite AI organization, the 'Superintelligence Team', shows the future direction. Actually, it seems like people are attracted to the aggro because of their annual salary or nationality, but the essence is to analyze what they want to do now. 🧠 Technology Focus Areas LLM Optimization Multimodal learning Eliminate knowledge bias Reinforcement learning-based fine-tuning 1. Strategic talent recruitment Recruiting 15 people (31%) from OpenAI alone is a very aggressive talent acquisition strategy. DeepMind, Google, and other major AI companies are recruiting key talents It seems like they want to solve the implicit part in the process of building and fine-tuning the foundation model. 2. Expertise by field OpenAI alumni are concentrated in the Foundation Models team Many DeepMind alumni are assigned to the Computer Vision team Secure the best experts in each field 3. Proportion of Chinese researchers 47% of the total are Chinese, especially in key research areas Most of them have PhDs from top universities in the United States. Some see this as a security or political issue, but if you look at the papers they have written before, they are just people with great ability or symbolic achievements. Leadership hierarchy Name

Hello, this is Haebom. It's been a long time since I wrote a notice. Thank you so much for always subscribing to my blog and sending me warm comments and interest. Your subscriptions and comments are a great source of strength and encouragement to me. Today, I have a special project I would like to introduce to you. This book is a book that even the publishers themselves tried to make and wrote diligently during the time when e-books were all the rage, but even I forgot about it for a while. At the time, I had a hard time sympathizing with selling e-books for hundreds of thousands of won or writing books about things I had not personally experienced. So I made a series of five books based on things I had personally experienced or understood indirectly. At the time, I received publishing offers from various publishers, but I declined them because I was curious about the e-book market. Now that I think about it, I think I might have been too profit-oriented when I would have kept the physical copies of the books. In fact, each book is about 100 to 200 pages, so if I were to make it into a physical book, I think it would have been about 200 to 300 pages with graphics and typesetting. That was a long digression. •Volume 1: What to Build 101 – Blocking the Creative Process (Difficulty in Completing) •Volume 2: GTM 101 – Uncertainty of Market Reactions (The Difficulty of Selling) •Volume 3: Scale 101 – The Complexity of Growth (The Difficulty of Systematizing) •Volume 4: Iterate 101 – The Chaos of Improvement (The Difficulty of Optimization) •Volume 5: Impact 101 – The Conflict between Purpose and Profit (The Difficulty of Finding Meaning) Thanks to a wonderful person who recently purchased this book, I was able to revise it again, and this time I was able to organize everything from the cover to the ebook more neatly. ✨ In fact, I think I had completely forgotten that I had written it. I can't tell you how much strength I get when I see people contacting me after seeing things I made in the past. Whatever the reason, I sincerely thank you for finding and reading this book. Each and every one of you is a special first reader of this book. I would be even more grateful if you could recommend it to people around you. In fact, I am secretly greedy. Hahaha. And there is an open KakaoTalk room. You are always welcome. (Entry code 1024) Thank you for your continued blog comments and subscriptions. I cherish the time we spend growing together. 🤝 Thank you for always visiting my blog.

"You have to brush your teeth while showering, wear shoes without laces, and save every minute of your day." Have you ever thought about this? If you look at SNS or YouTube recently, you can see that so-called 'micro-efficiency' is deeply embedded in people's daily lives. It is a movement to save even a minute of time with a thoroughly calculated strategy for even the smallest actions. The philosophy behind a cup of tea Veronica Pullen, 54, from the UK, makes two cups of tea every time she drinks it. One cup is lukewarm with milk, and the other is perfect 40 minutes later. This way, she saves 20 minutes a day, or about 10 days over two years. Such small efficiency can make a big difference in the long run. Pullen’s habits don’t end there. She always has boiled eggs for breakfast, an omelet for lunch, and a set menu for dinner. She also repeats a set pattern for her clothes every day. This minimizes the stress of having to make new decisions every time. She says that this habit gives her enough energy to do what she wants. Why are people so obsessed with one minute? This phenomenon is by no means limited to the habits of unique individuals. The tag #LifeHack has been mentioned more than 11 million times on TikTok, and countless productivity influencers are emphasizing the “1% rule.” The best-selling book Atomic Habits preaches to the public how the cumulative effects of small habits can bring about big changes. In a social context, this micro-efficiency is not just a hobby or a trend, but also a phenomenon that reveals the pressures of modern society. Technological advancements have made it possible to do things faster, but the paradoxical situation of having more work to do in the remaining time is repeated. Psychologists interpret this as an early symptom of "active burnout." Good Habits vs. Obsessions People who pursue efficiency have various reasons. They range from people who are sick and want to save even a little energy to people who simply want to practice 'laziness' more comfortably. However, experts warn that if the habit of efficiency becomes too obsessive, it can lead to obsessive-compulsive disorder or depression. Being efficient does not necessarily mean being satisfied. In fact, a British survey found that on average, people only have 23 hours a week where they feel truly free. That creates pressure to save every minute. 3 Tips for Healthy Efficiency But unconditionally pursuing efficiency is not the answer. So how should we approach healthy efficiency? Check your purpose. It is a good idea to make a clear plan on how you will use the time you save. If it ends with just saving time, efficiency will eventually lose its meaning. Allow yourself some 'looseness'. Overly strict rules can actually increase fatigue. It's also a good idea to intentionally break away from efficiency habits on weekends or vacations. Review and reset constantly. As your life circumstances change, your efficiency strategy should change as well. Periodically check whether your current habits are really meaningful to you, and if they need to be changed, do not hesitate to modify them. What really matters is not the quantity of time but the quality. "What will we do with the time we save?" This is the key question that we must not miss. The habit of saving time should ultimately be a means to make our lives happier and more meaningful. Not simply to live busy, but rather to regain leisure in life. Actually, this is a continuation of the article...

Someone once said, "You do planning, development, and marketing. What on earth are you?" What is your specialty? To this question, I actually don't know. I just did everything I could. I just answered, "We're doing everything we can to make it work." Originally, GTM (Go-To-Market) refers to the entire process of launching a product or service to the market and making it successfully adopted by customers. The GTM manager in charge of this is in charge of everything from the product development stage to actually reaching the hands of customers. The Secret to Small Teams Growing Big While working at Kakao Brain and Nexon Korea, I definitely worked on great projects and had great colleagues, but I couldn't shake the feeling that something was lacking. In fact, I had a lot of concerns. Above all, while talking about the word 'growth', I couldn't really feel the actual growth of the product. Is what I'm making really providing value to my customers? Am I creating meaningful growth in my career? I don't like terms like growth/newbie/hexagon in the first place. Real Growth Experiences Learned from Notion Then, working with Notion completely changed my perspective. I learned how to build a product that has tens of millions of users globally, how to increase monthly recurring revenue (MRR), and how to create maximum efficiency with a small team. Notion wasn't just a company with a lot of people. It had clear values, a well-polished product, and users who believed in it. Through this experience, I realized the most important thing in my career. "Rather than focusing on large organizations, we need to focus on the relationship between products and users." Why I joined Gamma, and the results With this realization, I joined Gamma. And in two years, we achieved $50 million in ARR and have been profitable for 15 consecutive months. Gamma is a small but mighty team, and I am now a senior GTM (Go-To-Market) manager. It was an incredible growth, reaching 50 million users and $50 million in annual revenue (ARR) with just 35 people. Behind Gamma’s rapid growth were the following key strategies. I was promoted from the previous GTM Manager to Senior GTM Strategy Manager . The method we use when we attack a market is simple: set principles and stick to them. Let me share a few. 1. User-centric pricing strategy (Van Westendorp) We never guess when it comes to pricing. Gamma used the Van Westendorp Methodology to figure out exactly what users were “ready to pay” and set their prices accordingly. This created a solid revenue structure from the beginning and allowed them to maintain steady growth without changing their prices once for the next two years. Price is a signal to the customer. Accurate pricing creates trust between the product and the user. 2. Maximum growth with minimum manpower Gamma creates AI-based presentation tools that sell and market themselves. Our go-to-market strategy was simple:

The intense ending scene of 『Squid Game Season 3』, which I watched on Netflix a while ago, is still lingering in my mind. The final game is the so-called 'high-altitude squid game' that takes place in the sky. The game is played on square, equilateral triangle, and perfect circular pillars in order. Especially in the round pillar scene in the final stage, the protagonist chooses to sacrifice himself to escape the shackles of greed that dominated him. Until now, the circle symbolized greed and obsession, but when the protagonist chooses death on this circle for his child and future, its meaning is completely reversed. This scene vividly shows that the reality we see is not absolute, and its meaning can change at any time depending on the environment or experience. For reference, I found Squid Game Season 3 very boring. Culture Changes the Way We See? Two Interesting Illusion Studies A similar story can be found in two recent studies of visual illusions. Ivan Kroupin's research team at the London School of Economics (LSE) in the UK showed a picture called the 'Coffer illusion' to Western people and rural Namibians. Interestingly, Western participants mostly saw squares, while rural Namibians mostly saw circles. The research team explains this with the 'carpentered world' hypothesis, which states that the environment we live in determines the way our brains see the world. But another study challenges this hypothesis. Dorsa Amir and Chaz Firestone have shown that the Müller-Lyer illusion occurs in humans, animals, and even blind children, regardless of their environment. These two studies may seem to conflict, but they actually send an important message: the world we see is not an objective reality, but something our brains are constantly interpreting and creating. Seeing and speaking are ultimately 'hallucinations of the brain' Neuroscientist Anil Seth describes reality as "a controlled hallucination created by the brain." In other words, our brain does not show us the objective world as it is, but rather interprets the world subjectively based on our experiences, environment, and culture. What's interesting is that this phenomenon applies equally to language as to vision. Take, for example, a study by Stanford University psychologist Lera Boroditsky, who asked German and Spanish speakers to describe the words “key” and “bridge,” and found that the genders of these words were reversed in both languages. In German, key is a masculine noun, and leg is a feminine noun. In Spanish, key is a feminine noun, and leg is a masculine noun. Surprisingly, German speakers described keys with masculine traits such as “heavy, strong,” and legs with feminine traits such as “beautiful, elegant.” Spanish speakers, on the other hand, described keys with the opposite traits. Likewise, the language we use ultimately determines how our brain creates and interprets reality. What we believe we ‘see’ and what we believe we ‘say’ are ultimately controlled illusions of the brain. 『Squid Game』, Visual Illusion, The Same Story That Language Tells The reason why the final circular pillar scene in 'Squid Game' was so powerful and why language changes our perception of reality is ultimately the same. The meaning of the world that we believed to be absolute can change at any time, and the world can be completely different depending on the language, culture, and experiences we have. The common message that optical illusions, language studies, and drama give us is clear. The world is always being reinterpreted and recreated in our minds. So if we want to change our lives to be more positive, shouldn't we first change the way we look at the world? Just as the protagonist of 'Squid Game' changed the meaning of the archetype from greed to sacrifice, we too can change the meaning of what we see and say in our lives. The difficulties, pain, and even happiness we experience can ultimately be changed by how we look at them. In fact, the reason why the work called "Squid Game" received attention was because the games that we remember from our daily lives or from our childhood have become games that adults who are obsessed with greed risk their lives to play. If we change our perspective a little, wouldn't something completely new or fun come out?

Recently, I met someone who is fluent in Chinese through Skewer Coaching. We talked about apps that are frequently used in China, and I told them about the ones I have used. Even those who don’t know much about China have probably heard of WeChat. There is even a joke that you can’t do almost anything in China without WeChat. Ordering food, calling a taxi, shopping, making payments, and even using government services can all be done within WeChat. Compared to KakaoTalk or Naver, which are commonly used in Korea, it offers a much wider range of functions within a single app. Is this “super app” phenomenon in China simply because Chinese people prefer convenience? Or is there another reason? When explaining the differences between Eastern and Western app design, it is often said that "Asians like apps with a lot of information and complexity, while Westerners prefer simple apps." However, this is not actually the case. Recently, apps in Asian countries such as China, Korea, and Japan have also gradually changed to a cleaner and simpler design. Nevertheless, why have super apps become so strong in China? The starting point of the 'mobile first' era that began with smartphones In the early 2010s, the Internet was not yet fully established in China. At that time, the Internet penetration rate in China was less than 35%, and there were not many people using desktop computers. In this situation, smartphones became the first personal computing devices for Chinese consumers. Since the mobile-centric Internet environment was built from the beginning, existing Internet habits were not formed. For example, in the US and Europe, desktop-based Internet usage habits such as web browsers and emails were already established, but in China, mobile apps were mainstream from the beginning. This gap was filled by Tencent's WeChat. WeChat started as a messenger, but began to provide almost all Internet services through 'Official Accounts' and 'Mini Programs'. In fact, WeChat is like a browser that acts as China's Google Chrome. Instead of moving existing web-based services to mobile, it created an app-centric environment from the beginning. Here, I always think of the words of Tencent Chairman Ma Huateng. When Tencent's QQ and games were ridiculed as copycats, Ma Huateng said, "We drew a tiger after seeing a cat." The early days of China’s digital economy were poorly developed. Banks did not provide consumer-friendly online payment systems, shopping was largely cash-based, and e-commerce was in its early stages of lack of trust. In this situation, Alibaba developed Alipay for online transactions on Taobao, creating its own payment system. Companies had to create their own services if they didn’t have what they needed. This is the real reason why super apps were born. In other words, it was an inevitable choice to fill a gap in the market and quickly take over the entire industry, not for the sake of an ideal user experience (UX). The emergence of the 'wall-building' competitive strategy The rise of China’s super apps can be explained by another reason: the extreme competition. Giants like Tencent, Alibaba, Baidu, and ByteDance have long used a so-called “walled garden” strategy, blocking links to each other’s platforms. For example, if you try to open a shopping link on Taobao in WeChat, it won’t work. So each company tries to pack as many services as possible into its own app. This has led to super apps like WeChat offering food delivery, taxi hailing, payments, shopping, and social media all in one app. Although the Chinese government only banned such link blocking in 2021, consumer habits have hardened and the super app model is deeply entrenched. Another reason can be found in economic terms. The online spending power of early Chinese consumers was lower than that of the West, so the lifetime value of each individual app was small. Accordingly, companies tried to maximize revenue per customer by providing as many services as possible to each customer. Also, because the initial cost of acquiring customers was very high, they integrated more services to avoid sending users to competitors once they had gathered users in one app. This was part of the strategy when Tencent ran the “red envelope (hongbao)” campaign during the Chinese New Year (CNY) to promote WeChat Pay. Strategic choice, not cultural preference The reason super apps succeed in China is not because of user demand, but because of a combination of unique market conditions: a mobile-first environment, lack of industry infrastructure, fierce competition, and low spending power. So should companies in other countries just copy the super app model? Not necessarily. Rather, the important lesson is to understand what drives product design decisions. Super apps aren’t always the answer. In some cases, it might be better to bundle multiple services into one app, while in other cases, it might be more effective to create a single app that provides the best user experience. In reality, WeChat is not a 'perfect super app' that can solve everything. The way it allows users to experience simple services through mini programs is essentially the 'Open Web' of the mobile environment. Users can experience the service in advance and then download a separate app if necessary. In Korea, there were some who once boasted that super apps were the future. But who actually achieved this? Personally, Toss is the only company that has truly achieved a super app. I think Toss's current app is the real beginning of this. The biggest Takeaways we can get from the case of China's super apps is to understand how business needs evolve products. After all, product growth begins not with user needs, but with strategic responses to the environment and market conditions that companies face. Personally, I think Korea is the place that most underestimates China's digital ecosystem and technological prowess. As I always say, if you go to a developed city in China or even Shanghai, you will realize that this is not the China I knew.

Recently, while consulting on AI planning and architecture for a company that creates financial services, I felt that this kind of discussion is accelerating in the financial industry. Although generative AI such as ChatGPT is receiving a lot of attention, the financial sector has been actively utilizing AI for a long time. From fraud detection to credit risk management, and even ultra-short-term trading strategies, AI is playing a key role in many core financial tasks. However, there are still many tasks left for AI to be actually trusted and used ethically. The most important issue among them is explainability. In the financial sector, the more complex the AI model, the more difficult it becomes to understand how it makes decisions. This is the so-called 'black box' problem. Even if an AI model makes accurate predictions, if it cannot explain the criteria and process by which the predictions were made, it can be a serious problem, especially in a field like finance where trust is essential. So today, I would like to take a deep look at what this 'explainability' is and how it can be implemented in the financial sector. Can I give you an interesting example? In 2019, Apple Card became a social issue due to the controversy over gender-discriminatory loan screening. A couple with the same income and credit rating applied, but the husband's credit limit was set much higher than the wife's. People immediately criticized this decision as 'gender-based discrimination.' However, the card issuers and financial institutions that managed the screening algorithm failed to explain exactly why this problem occurred. As a result, their image suffered serious damage. This case illustrates the potential problems that can arise when AI operates in the financial sector. AI makes decisions based on data, but if the data itself is biased or the algorithm’s judgment criteria are not clearly revealed, financial institutions can face serious ethical and legal responsibilities. In this context, the financial sector must ask the following questions when using AI: “Why did our AI model make that decision?” “Are the decisions made by AI really fair?” “Can we explain the judgment criteria of AI models?” Three Key Elements of AI Explainability Explainability is more than just showing the technical details of how a model works. To properly implement explainability in AI in finance, all three of the following elements must be present: (1) Transparency It's about making it clear to stakeholders how the AI model is structured, what data it was trained with, and what prerequisites or assumptions it operates on. For example, trust can be built by disclosing to customers and regulators the data sources for credit rating models and the reasons for selecting assessment variables. (2) Interpretability The goal is to make AI decisions easily understandable to humans. The way the model works should be explained using simple algorithms or visual tools. For example, you should be able to explain why you declined a loan application with specific data points (“Your loan was declined because of your high credit card utilization”). (3) Accountability It's about establishing clear accountability for decisions made by AI models and deciding in advance how to respond when problems arise. When a model makes a bad decision, establish clear processes and accountability to immediately correct it and remediate the damage. An integrated approach that embraces all three elements is key to properly implementing the explainability of AI in finance.

When we visit a hospital, we always expect accurate diagnosis and quick treatment. However, the reality is that it is not easy to receive treatment when you want it due to long waiting times and a lack of doctors. However, a surprising study recently released by Microsoft has presented a new possibility to the medical field. It is that a diagnosis system using artificial intelligence (AI) made a diagnosis four times more accurately than a human doctor. It is true that there are doubts such as, "Can AI really replace doctors?" AI Diagnostic Orchestrator Enters the Medical Field Microsoft 's AI Diagnostic Orchestrator (MAI-DxO) isn't just a simple AI model. It's designed like a panel of five doctors with different roles. Each AI agent develops a hypothesis, selects a test item, and consults with each other to arrive at a final diagnosis, deriving the most appropriate treatment method. What’s interesting here is that it clearly shows the process by which the AI arrives at its conclusion. Microsoft calls it the “Chain of Debate,” and it transparently discloses the logic through which the AI solves the problem. Accuracy that surpasses human doctors? So how effective is this AI in real-world medical settings? To test this, Microsoft presented the AI with 304 of the most challenging diagnostic cases published in the New England Journal of Medicine (NEJM), the top medical journal in the United States. The results were astonishing. When AI worked best (using OpenAI’s o3 model), the diagnostic accuracy was a whopping 85.5%. Experienced human doctors diagnosing the same cases had a success rate of only 20%. Despite the limitations of human doctors not having access to textbooks or colleagues, AI’s overwhelming performance came as a huge shock to the medical community. AI that saves both cost and time In addition to accuracy, the cost-saving effect was also noteworthy. Microsoft set the AI to consider cost in the diagnostic process, and as a result, the number of tests required was significantly reduced, saving hundreds of thousands of dollars in real-world cases. “This system is the most advanced AI performance we’ve ever seen, and could open new doors to healthcare accessibility,” said Microsoft’s Dr. Dominic King. AI models are now 'products', real competitiveness is 'combination power' In this experiment, Microsoft used AI models from several companies, including OpenAI, Meta, Anthropic, Google, and xAI. In particular, Mustafa Suleyman emphasized that even the best-performing OpenAI model will ultimately be “commoditized,” and the real difference lies in the “orchestrator” that integrates and combines these models. Microsoft said it plans to apply the technology to its AI chatbot Copilot and its Bing search engine, which could have huge potential on a platform that processes more than 50 million health-related questions a day. The Era of 'Medical Superintelligence' Needs Preparation Mustafa Suleiman describes this research as a first step toward “medical superintelligence.” A future where faster, more accurate, and cheaper diagnoses are possible is just around the corner. However, further verification is needed before it can be applied to clinical settings. Dr. Eric Topol, a cardiologist and AI medical authority, also evaluated this research as an important study that proved the possibility of AI’s medical efficiency, although it was not conducted in a real medical setting.