Inspired #18 | Tech-Driven Science, AI Startups, Utility AI, Socratic Love, Growth Maze, Learning Routines, Director AI, Gaming Slump, The Age of Taste, Universe as Data, Strong Form AI Companies
This text was originally written in Korean on February 16, 2025, and has been translated into English and uploaded
I’ve been seeing and reading a lot of interesting things lately, so this time there’s quite a bit of content!
Keynote: The Era of Technology-Driven Science
It’s hard to even guess how fast technology will advance from here.
A recent paper supports the idea that the essence of the universe is actually data
I’ve always thought of myself as a being made up of countless inputs and outputs of data, which is why I organize my personal Notion into input and output data. After watching this video, I’m even more convinced that this is the fundamental mechanism of humans and the universe itself.
Garbage in, garbage out.
Quality in, quality out.
The Start of a New Industrial Revolution (No Jeong-seok, CEO of BFactory, an AI Startup)
“If you’re not desperately immersed in something, risking everything and being in the trenches, your sense of reality will inevitably fade. That’s why I choose entrepreneurship as a way to keep myself on the battlefield."
I completely agree with this. When I step away from engaging with great ideas, talking to customers, or building products for even a month or two, I feel my sharpness dulling. Life, business, products, and relationships all rely on momentum and compound effects—you can’t let go. Of course, rest is important, but you can’t drift too far.
ㅤ
But think about it: companies like Microsoft, Google, and Meta all deal with information in the digital world—data, as we call it. These forms of data have traditionally been handled by industries like media, which, though not centuries old, have roots in things like Gutenberg’s Bible, newspapers, magazines, and television.
These were businesses that delivered content and slapped ads on it. The spaces Big Tech has dominated in the software world are just the traditional media industries moved to software.
And that’s already massive. But if you add up all the real-world industries beyond media—construction, logistics, aviation, even everyday tasks like cooking or childcare, basically all forms of physical labor—it’s an enormous market.
These have been traditional businesses, but as we all know, when machines and intelligence combine—whether it’s robots or whatever—they essentially become human-like.
Humans are a combination of software and hardware, and if you see those units as foundational models, then organizations are just agent frameworks tying them together. That’s how our systems are evolving. So, if machines and intelligence combine to create human substitutes, doesn’t it make sense that everything humans enjoy will change? I totally agree.
The essence of the internet is connecting nodes—information to people, entertainment to people, people to people. The places that did this best became Big Tech, with their moats built on network effects. Now, with advances in AI, robotics, and biotech, entirely new UX and business models are possible, and whoever nails that could be the next Big Tech. I believe AI and robotics will free up more leisure time for people, and I’m working on an AI-native game with a new UX and business model for that future.
ㅤ
Ultimately, it’s the intersection of knowing AI and knowing the domain that matters.
The grand strategy is to solve AI problems. We need to drastically increase the number of AI engineers and set up a PM in every industry rather than diving into each one ourselves.
This feels a lot like Palantir’s organizational structure and business approach
ㅤ
I tend to observe what makes things different, and I’ll say something a bit risky: the ability to use AI is clearly limited by the user’s own capabilities.
AI can only pull out what the person is capable of asking for. So, even if AI has the magic to do more, if the user can’t tap into it, it’s useless. To put it bluntly, we need to be exceptionally capable.
I think in the age of AI, it’s not about studying less—it’s about studying more. We’re entering a time when you need to be even more educated than before.
This resonates deeply. It’s not that AI will do everything for us so we can slack off—it’s that those who know their stuff can use AI to achieve incredible things. Knowledge, skills, and experience in your domain, plus a solid philosophy and attitude, are the foundation. Then, you amplify that with AI.
For example, almost every day, I throw my thoughts and ideas at GPT and ask for counterarguments. It’s like breathing for me. This helps me uncover unknown unknowns and blind spots, which is hugely beneficial for strategic thinking, problem-solving, and high-quality decision-making.
I’m also experimenting with AI in product development, especially in gaming, where there’s a lot to leverage—BGM, sound effects, 3D models, coding, and more. I’m expanding my domain knowledge by trying out various tools.
ㅤ
Even as we talk like this, if you look at Silicon Valley’s AI community—Twitter groups, Reddit threads—they’re discussing futures way beyond what we’re imagining. Some are already talking about Mars, and the gap is only going to widen. This is a tough topic, and I’ve been circling around it, but English is incredibly important. Some say you don’t need to learn English anymore, but with the overwhelming amount of English content and opportunities for English conversations, I think it’s crucial. Math is even more important.
Personally, I really want to move to San Francisco. It’s not just about tech startups—I crave being surrounded by brilliant, inspiring people and building great products with them every day. Emotionally and socially, it feels like home. I’m determined to settle there by the end of this year.
So, I’m speaking English daily, reading and writing in English, and dabbling in math when I can. Recently, I read a fascinating book called The Discovery of Zero.
Marc Andreessen: Trump, Power, Tech, AI, Immigration & Future of America | Lex Fridman Podcast #458
Corporate strategies for AI can be split into weak form and strong form. The weak form treats AI as a sixth-priority option within a company—something supplemental—since workflows and cultures are already established. It’s used for basic tasks like summarizing emails or co-writing documents.
The strong form, on the other hand, is about completely rethinking how work gets done with an AI-native, AI-centric workflow.
AI’s Similarity and Difference with Humans
AI is human-like in some ways but starkly different in others, especially in scalability and accessibility. Multiple AI agents can collaborate to finish in a day what might take ten humans a week—at a fraction of the cost. They don’t need breaks, which helps. Unlike humans, who exist separately from the internet, software, and hardware, AI is seamlessly connected to them. This means it can access and process vastly more information, faster and more accurately, in parallel. Historically, human progress came from more people accessing more data and information with increasing clarity—think papyrus, books, libraries, Gutenberg’s press, democracy, education reform, computers, and the internet. Now, there’s so much data that a single person couldn’t absorb even 1% of it in a lifetime. AI’s overwhelming advantage in accessibility—finding, learning, and organizing data we don’t even know exists—is what sets it apart most from humans.
Why We Need a Strong Form Stance
So, whether you’re starting a company or building a career, you should actively shift your stance on AI to the strong form. You’ve got to embed this into yourself somehow.
Features of an AI-Native Company
Humans handle soft skill jobs (communication, creative & strategic thinking, storytelling, etc.)
AI takes on hard skill jobs (coding, design, writing, video production, etc.)
Remove middle layers (managers)
Automate repetitive tasks as much as possible with AI
The economics of open source vs closed source in AI
When I look at model diversification, miniaturization, or LLM distillation, it feels like watching humans specialize and divide labor to grow industries. Instead of running massive models endlessly, training smaller ones keeps core intelligence intact while cutting compute costs and boosting productivity. It’s as if AI is undergoing its own industrial revolution.
The AI and DAOs Edition
In 2014, Vitalik defined a DAO as follows: "an entity that lives on the internet and exists autonomously, but also heavily relies on hiring individuals to perform certain tasks that the automaton itself cannot do…" "Automation at the center, humans at the edges"
What if the main participants in a DAO were autonomous AI agents? Could that really change things? It’s a topic worth following.
AI 101: Introducing Utility AI
The solution to this is a process known as bucketing. Possible actions such as eating at the table, watching tv, or strutting your stuff on the dance floor have different priorities based on the current context of the game. So in the Sims, motives are bucketed based on utility value. Then only the highest priority motives are considered. The action then selected is devised by looking at all the possible interactions in proximity that satisfy that motive, and are also scored accordingly. This ensures that a thirsty or hungry Sim always seeks to address that first, and won't wind up dying of hunger because they were wasting away on the couch.
If you’re building a real-world-like simulation game, you’d need to design it with:
How realistically priorities mimic a player’s
How closely the environment resembles reality
How human-like the execution of actions feels
AI 101: Managing the Experience with Director AI
Director AI is not a precise technological process, but rather a philosophy for how a gameplay system can make intelligent decisions that influence a players experience. As we've seen throughout this video, the range of titles they're used in is fairly broad and the purposes for which they're adopted can vary even between similar games. But critically, it's all about keeping the player at the heart of gameplay and ensuring that you are experiencing the game in the way that designers intended and hopefully having some fun as well while you're at it!
Item Type Generation: If a user crafts a gun, AI tags it as a weapon/projectile type automatically.
Function Generation: Based on the item’s description and type, functions like shoot/reload/aim are added.
Sound Generation: Sounds are added based on the item’s description, type, and functions.
AI NPCs: Generated based on the player’s situation, location, and playstyle, with memory of past interactions.
An Interview with Matthew Ball About the Gaming Slump
Key takeaways:
From 2011 to 2021, the gaming industry grew at an impressive 9.7% annually—three times the global real GDP growth rate—thanks to mobile gaming (billions of new players), cross-platform play (network effects), new genres (battle royale, AOS), and free-to-play models.
From 2021 to 2024, it failed to hit 30% annual growth. Mobile gaming saturated, growth drivers faded, new tech (cloud gaming, AR/VR) underperformed, AAA titles had gameplay issues, and dominant games monopolized attention.
The industry’s in a deflationary phase: rising production costs, shrinking purchasing power, and fiercer competition. Top successes rake in huge profits, creating polarization—Call of Duty has earned ~$45B, outpacing the Marvel Universe, rivaling Pokémon and Harry Potter IPs combined.
TikTok’s eating mobile gaming time; 2024 US mobile game downloads hit a post-2014 low. Consoles stagnated, except for the Nintendo Switch. PlayStation’s 7% profit margin is low for its 100M MAU and $30B revenue, outpaced by top mobile/PC games.
PC gaming thrives: Steam offers 110K+ games, 200M MAU, 6B hours spent yearly, $22B revenue—with just 350 employees. Discord’s gamer-centric platform has 80M DAU, 90% of whom game every other day.
Vercel's Guillermo Rauch on AI and the Future of Coding - Ep. 47
With coding becoming increasingly abstracted by AI, I think it’s crucial to think at a higher level. Coders need to see the bigger picture more than ever—considering product direction, decision-making criteria, and data while overseeing code principles, architecture, and tech stacks. That way, they can properly direct AI and evaluate its outputs. In the short term, it feels like everyone’s becoming a conductor—it’s all about where, how, and what to orchestrate.
An operator’s guide to product strategy | Chandra Janakiraman (CPO at VRChat, ex-Meta, Headspace) An operator’s guide to product strategy | Chandra Janakiraman (CPO at VRChat, ex-Meta, Headspace)
The mention of leaders having “pet ideas” is a fun, intuitive way to put it. Sometimes leaders avoid sharing ideas because they worry it’ll feel like implicit direction or micromanaging. But whether you act on pet ideas or not, it’s important to share them and clear up any curiosity—otherwise, they’ll just linger in your head. I’ve been there, so this resonates.
The Growth Maze vs The Idea Maze
If we want to build a truly great product or business model, deep reflection and thought experiments on why Y over X are essential.
Unlocking simplicity: Customer empathy as a (not-so-secret) weapon
These are basic things we often fail to do. It’s crucial to make them a regular practice. For early-stage teams, I think it’s fine to spend 70% of your workweek on this alone.
Selfish Software
I believe balancing profitable ideas with curiosity-driven ones is key. Without money, pursuing curiosity and fun becomes a runway issue; without curiosity and fun, burnout hits before the money does. The takeaway is you need both, but sometimes you have to do unenjoyable work—finding or creating fun in it is crucial for sustainability.
Taste is Eating Silicon Valley.
In a world of scarcity, we treasure tools. In a world of abundance, we treasure taste. The barriers to entry are low, competition is fierce, and so much of the focus has shifted — from tech to distribution, and now, to something else too: taste.
In scarcity, we value tools; in abundance, taste. This sentence captures exactly what I felt last year. In an era overflowing with products and services, including software, things keep fragmenting into vertical niches of taste. Going forward, many software products will increasingly evolve in the realm of taste (except for legacy or heavily regulated industries). A founder’s taste seeps into the product and team-building process, directly shaping the user experience. Early UX and user interaction strategies hinge almost entirely on the founder’s taste—including their understanding of what people like.
What Socrates Teaches Us About Love, Politics, and Death
The greatest gift others can give us is pointing out our mistakes, and the only way to respect others is to answer their questions or question their answers.
Love Grows
From yearning to bonding, love is a collection of intentional acts.
People rely on apps, algorithms, and chance to find partners, leading to a casual attitude. Media often portrays love as something stumbled upon, making it seem like it just happens rather than something you actively participate in. But love isn’t found by accident—it grows through consistent effort and care, like tending a garden. Love is a collection of intentional acts, from longing to connection.
Three AI Apps That Deserve the Hype
I like poking fun at the internet because I really love it and I know it can be better,” Agarwal says. “I think the web is one of the best creative mediums ever made, up there with paintings and movies, and somewhere behind the enshittification lies so much potential and I want to explore that.
This is one of the things I completely agree with! Remix and spread—that’s how the internet works.
Optimize Your Learning & Creativity With Science-Based Tools
The nervous system can change itself through conscious thought or external feedback. Neuroplasticity comes in short, medium, and long-term forms, each optimized in daily life based on arousal:
Short-term: Temporary changes (e.g., tomorrow’s dinner plans)
Medium-term: Changes within a limited period (e.g., hotel location during a business trip)
Long-term: Lasting changes (e.g., learning a new language or skill)
Routines for effective arousal:
Morning sunlight → Activates the circadian-adrenal circuit → Cortisol release (wakefulness)
Delay caffeine for 2 hours after waking → Strengthens the circadian-adrenal circuit → Clears adenosine receptors for maximum caffeine effect later
Early morning exercise → Boosts neurotransmitters like epinephrine → Increases alertness in late morning to afternoon
First meal at noon, low-carb, regardless of exercise
If drowsy at 2-3 PM, do low-cognitive-load tasks. At 4 PM, practice non-sleep deep rest protocols. Resting well here can give you a second wind for new learning.
When over-aroused, a quiet environment helps; when sleepy, a bit of noise aids learning.
I’m planning to share some thoughts on a new product I’ve started working on soon. Hope everyone has a great week!