jdw64

I agree. Sol followed my instructions well and wrote good code.

In the end, can aging itself not be cured, and is the only option to prevent the program that causes it? I'm curious

Finally, it seems like a good time to do some 'load-bearing' work on my project for a while

I've observed this too. I've also started using AI-style writing a lot.

I wonder how much my post would come across as AI-written. I wrote almost all of it by hand. Of course, I do use AI for technical reinforcement, but I wonder how this kind of principle would come across.

I think this is a form of path dependency. We talk about unmaintainable messes, but honestly, if you look at Kairosoft or famous game codebases, you'll find 30,000 lines in a single file or just completely chaotic code. I am actually in a position where I see bad code all the time.

The truth is, "good code" is relative. It is determined by the specific domain and the composition of the team. Is incomprehensible FP (Functional Programming) code good? No, it isn't. A programmer must assess the team's capabilities and adapt accordingly. Good code is ultimately something that morphs based on the shape of the organization. Once defined this way, good code might share certain commonalities (like readability or a shared mental model), but its actual form varies wildly.

So, what is good code? That definition is missing. To be blunt, the Hacker News posts insisting that we must write "good code" are essentially a form of self-hypnosis.

Just look at paradigms. The mechanics of OOP have changed significantly, FP approaches have evolved, and DOD or DDD are fundamentally different from their early days. Whenever paradigms are discussed, someone claims, "That problem was solved in the past, and nowadays we do X," only for someone else to reply, "I don't think that's actually solved," leading to a fragmented breakdown in consensus. Ultimately, which knowledge remains as tacit knowledge is entirely dependent on the organization's capability.

You could argue that AI is terrible at simplifying code. However, I am skeptical that AI coding needs to be identical to human coding. When you actually code with AI, it often produces structures humans would call anti-patterns, including God Objects. Yet some of those structures can be faster or simpler for machines to navigate. There is no reason to assume that the optimal modularity for AI maintainers must be identical to the optimal modularity for human maintainers.

Of course, I am not denying that the rewards of good architecture are delayed, or that there comes a point where maintenance becomes impossible. But as the AI era ushers in an age of overproduction, software could become disposable, strictly personal, highly tailored to small niches, or ultimately, heavily polarized.

Realistically, programming domains fall into two major categories: "ship it and forget it" (one-offs) and continuous services. I agree with the OP's point that AI struggles to understand boundary delineations. But honestly, you can enforce those boundaries by injecting them into the spec. How those boundaries are drawn in the first place, however, is purely a matter of personal experience.

Personally, I define "good code" as code that allows the entity responsible for the software to achieve its purpose with a sufficiently low cost and error rate, factoring in the software's expected lifespan and future changes.

If you ask an AI to generate work based on this standard of what level of code is "adequate," you might get entirely different results. The biggest problem with discussions around AI is not just that ideological identities prevent proper evaluation (as seen in that article), but that the AI itself scales proportionally to its input. It is an incredibly difficult issue to judge because you don't know an individual's workflow or exactly how they are utilizing the tool.

I do think the value of reading code is important. However, much of what we are discussing in the AI era is actually rooted in the path dependency of how to become a good human senior developer.

Instead, the core focus of AI-driven development might shift toward defining broader abstractions: data semantics, invariant external contracts, and migration strategies.

Ultimately, I believe the paradigm shift of our era should lead us to ask: "How do we write code most economically in a system where AI is the primary maintainer?" The OP might think differently, but at least, that is where I stand.

Googling itself lacks credibility. Most of the results are plastered with Google ads. And since Google ads mostly skip any kind of review process, there are a lot of bad ads.

I think one of AI’s biggest effects is simply that it makes people realize even high-income professionals can lose their positions.

This also follows from how current LLMs work. In practice, when I use them in domains I already understand, they can produce very high-quality results. But in domains I do not know well, the results can be poor, and the bigger problem is that I may not even be able to judge how poor they are.

So my conclusion is that AI will reduce the number of jobs, but it will not eliminate the need for people.

In education, the value of memorization may decline in the AI era. We may instead place more emphasis on domain modeling, problem framing, or the ability to choose and use tools effectively. But the more fundamental issue is that the IT industry may simply lose the capacity to employ as many people as it once did.

More precisely, I mean white-collar labor.

I think the deeper cause is a K-shaped economy in which the lower and middle classes become poorer. When ordinary consumers become poorer, one of the first things they tend to cut back on is discretionary spending, including spending on many kinds of IT services.

The core infrastructure layer is different. Large incumbents such as Microsoft and Google already dominate much of it, and they are likely to be more resilient. Search, video consumption, and a few other essential digital services will also remain strong. But many other IT services are, in practice, discretionary goods. Those companies may be hit much harder if consumers have less purchasing power.

People talk constantly about productivity these days, but we were already living in an age of overproduction before AI. AI is moving us from overproduction into an era of explosive production. The problem is that production can expand far faster than people’s ability to consume.

The cycle is supposed to be:

*products → revenue → employment*

But if the consumers who are supposed to support that revenue become poorer, the cycle weakens. Productivity alone cannot solve that.

I agree with the author that academics need to move beyond treating papers as the primary unit of achievement. Much of what the article argues is reasonable.

But there is another difficulty. Most academics built their reputations through papers. They use that reputation to obtain speaking opportunities, consulting work, grants, and other forms of income and status. Even if one person decides to move beyond the paper-centered system, it is difficult to change much unless the larger incentive structure changes as well.

My view is that IT workers have, in a sense, been working to reduce their own jobs since long before AI. The more infrastructure becomes centralized, the more peripheral and smaller companies are squeezed first. AI is simply another example of that process.

Until recently, people often said that highly skilled IT professionals were difficult to replace. AI changes that perception. Even when it does not fully replace knowledge workers, it can put significant downward pressure on the wage premium attached to specialized knowledge.

I do think AI will raise productivity. But companies will also reduce headcount accordingly. And if purchasing power becomes increasingly concentrated among a smaller group of people, product development itself may become more biased toward the preferences of those few consumers. That can create another negative feedback loop.

The claim that universities can simply choose important problems that are cheap to validate is also more difficult than it sounds.

If validation itself increasingly depends on AI, and universities cannot afford to own enough GPUs, then they remain dependent on large AI companies. That dependency will inevitably influence which research problems are practical to pursue.

Any research program is constrained by the institutions and funding sources that make the research possible. Always.

At the same time, I actually agree with the author that universities will become more important.

People often talk about “skill” as though it were some pure and independent quantity, but in my experience hiring rarely works that way. If one candidate is highly capable without a degree and another is equally capable with a degree, employers will usually prefer the credentialed candidate.

More broadly, people tend to hire those with whom they feel cultural familiarity and trust. University networks provide exactly that. Alumni often help other alumni, directly or indirectly.

So I think universities may increasingly become both social institutions and stronger elite-training clubs.

For someone like me, coming from a poorer country and without much money, there may not be many choices in that system anyway.

Still, I think the author’s argument is far too optimistic.

JEV seems similar to BERT. Where would it be useful?

I use it mainly to check whether this code fits the rules I defined, just a yes or no. But I'm not sure if that's the right way to use it.

I don't believe the "software factory pattern" can even be established as a pattern. I wonder if it's considered normal these days to call a Silicon Valley management style a "pattern."

Fundamentally, a pattern is a reusable solution to a recurring problem. But what the author is describing here is simply an operational loop and a management strategy.

I think current agent methodologies are practically indistinguishable from human developer management theories. Isn't this just a feedback process? I would consider this an agentic workflow.

Also, people casually use the metaphor of a "factory" when churning things out, but a factory fundamentally operates on "orders"—it has a specific objective and a target production volume. The entire approach of just trying to dynamically respond to everything under this label feels somewhat contrived.

Furthermore, I always have this underlying question: why are we building software factories in the first place? Where are we manufacturing the people who will actually buy all this?

I've heard stories of people finding success by running "app factories" in the early 2010s when apps were scarce. But in the agent era, I believe we are already drowning in AI slop. We have to remember that back then, producer friction was high, and simply submitting an app was a difficult hurdle.

Now, AI handles most of the basics by default. Things that were once highly praised are now just the baseline. To create something actually worth selling today, you have to break the existing grammar entirely, build much more complex architectures, and offer deeper features. Given this new standard, I seriously question whether mass production is the right answer.

I sometimes think about this: to grasp the fundamentals of knowledge and complex phenomena, perhaps we need an external system rather than human knowledge systems. By that logic, maybe we need AI, which can handle far greater complexity.

Human capability, when you think about it, is complex. Why is Newton praised as being so damn great? He established the law of universal gravitation, F=ma. Why is that such a big deal?

He distilled countless phenomena in an open system into a single mathematical formula.

What makes it great is that he found common state variables and relationships across entirely different phenomena like falling objects, planetary motion, collisions, and artillery trajectories.

But does F=ma hold true for the entire macroscopic world? No. There are various conditions and specific situations in motion, but within most scenarios and a certain range of approximation, it outputs values that are useful to humans.

Why is the Schrödinger equation so great? Because it turned the time evolution of quantum states into a calculable mathematical law.

Human thought is essentially creating a closed system by deciding what to cut out and what to keep from the infinite degrees of freedom in reality. Academia is what reinforces that closed system.

A great theory is great not because it perfectly replicates reality, but because it compresses the immense complexity of reality into a small, closed formal system while still managing to explain a multitude of phenomena.

In that process, it feels like human thought and progress are shifting into a different framework. What LLMs do well is primarily exploring within the ontology and representation space that humans have already built.

I think there are two broad categories of discovery: One is forming a new closed system, and the other is connecting fragmented knowledge within that closed system. I feel that the vast majority of research focuses on the latter.

What LLMs excel at is finding unvisited points within a given representation space. This is typically the process through which master's and PhD students connect dots, build their skills, and form their own mental models. But the logic behind criticizing LLMs seems to be that they eliminate the very work these graduate students need to do in order to grow.

However, looking at it from another angle, perhaps our current knowledge systems and classifications have reached a limit, suggesting that we might actually need a completely new classification and knowledge system.

What is the core principle of an LLM? It's predicting the probability of the next sequence.

Let's say you type the word "cat". Cat - is cute (90%), want to eat it (6%), furry (4%). Because "is cute" has the highest probability, the next sequence proceeds in that direction.

Within this framework, human knowledge and logic largely operate the same way. Once an initial logical proposition is established, we follow it up with whatever makes logical sense next. From that perspective, I think LLMs will actually do this better.

But what is it that LLMs cannot do right now? They cannot create that initial logical proposition. I believe they lack the ability to carve out a closed system from an open system.

Stacking logic step-by-step within a closed system—LLMs do this exceptionally well. But whether that constitutes true "intelligence" is a different matter.

I feel that being logical does not necessarily equate to having intelligence.

Humans preserve and create different mental models and knowledge systems within an open system. Just as your thoughts differ from mine, LLMs lack the ability to form these distinct mental models.

If so, within these limits, what humans must ultimately do is construct the logical frameworks that LLMs can then fill in. Perhaps a new kind of logic dedicated to designing these frameworks will become the next major trend.

Viewed from this perspective, I have no idea if we are in a mere technological transition or something else entirely. Or whether it is even correct to say humans are strictly necessary to build that framework. Maybe my learning is just lacking.

Now I'm starting to really feel that I should use Rust, but I have no idea where to actually apply it.

Xom titters.

I guess that's the only way. Thanks for the advice.

Thanks for the advice.

For me, reading is fine, but writing and speaking are hard. Especially speaking. What should I do about this?

Should I change my environment first, after all?

The biggest obstacle to learning a foreign language is that it's so much slower and more frustrating than using your native language, right? Compared to your native-language typing speed, even English typing speed is only half...

So it's really frustrating. how do you all relieve that frustration?

Wow, but it's surprising that even those prestigious computer science graduates have to worry about employment.

Personally, I think current LLMs have clear limits, so would engineering majors be at risk? That's a bit of a difficult point.

Because LLMs are good at writing code, but as complexity increasingly grows, if a person handles it, the number of lines one individual has to deal with will grow exponentially. To maintain that or find bugs, you'd need major-specific knowledge. After all, building a program is fundamentally about controlling complexity.

There's a slightly difficult point here. It's true that frontier AIs are good at coding, but once you start running agents at a large scale, token efficiency varies depending on whether a human intervenes or not. Ultimately, if you compare current token costs, isn't hiring a junior person cheaper? Is the US a different situation? Thinking about the API prices of GPT 6 or Fable makes me think that even more.

But if a designer improves an AI design and I feel it's fine, does that make me strange? If you ran a blind test on this, I think it would come out with high preference...

When talking with LLMs, especially recent frontier models, what's frightening is that they speak more accurately than any expert.

It's embarrassing to admit, but I tend to trust the research materials LLMs bring more than my colleagues or the programmer friends I once respected.

So does that mean the Korean Peninsula is no longer a geologically safe region from earthquakes going forward? That's scary.

I feel the same way. Want to become a dumb and dumber duo? I believe you could be my friend.

Generally, merely possessing nuclear weapons incurs diplomatic disadvantages. Moreover, due to the sheer maintenance costs of the weapons themselves, they have effectively become symbols of state power for major powers rather than acting as a true 'deterrent,' and authoritarian regimes tend to fixate on such outward displays. In reality, unless a country is one of the existing major powers (US, Russia, UK, France, China), simply possessing nuclear weapons brings massive drawbacks. Looking at the current state of nuclear weapons, they practically serve no function other than as symbols for the survival and legitimacy of dictatorial regimes. Therefore, their practicality remains highly questionable.

AI, on the other hand, immediately drives people into unemployment. While AI will create jobs, I believe the jobs it creates will generally be of lower quality and lower pay than before—even if they are better than outright unemployment.

Looking at AI's current potential, it essentially forces a dependency on those who own massive data center infrastructures. Furthermore, while the principle of LLM chatbots is that output varies based on input, it is actually a dopamine-driven structure that yields instant gratification, much like cheap crack. Even though I constantly use LLM chatbots myself, they are addictive.

From this perspective, I agree that AI is more dangerous than nuclear weapons. Setting aside the actual capabilities of LLMs, I agree that the risks LLMs pose are far greater than those of nuclear weapons.

First, the greatest risk is that the knowledge industry itself is in danger of being subordinated to AI companies. Once this infrastructural dependency takes hold and 'friction' disappears, traditional knowledge providers who relied on that friction will go bankrupt, leading to a monopolistic convergence. This means that while workers at AI companies will earn astronomical sums, the knowledge workers on the periphery will become impoverished. We will see a severe wealth polarization where the 'average' wage might increase, but the overall population becomes poorer.

Second, as the areas where AI cannot perform continue to shrink, the cost of learning the skills required to handle those remaining untouched areas will increase significantly. In other words, the hurdle for knowledge labor will rise. This rising hurdle means that the educational and learning costs required to cross that barrier will become an overwhelming burden, making it even harder for the poor to succeed.

I believe that AI, in many ways, is more dangerous than nuclear weapons.

I think it depends on the workflow. I believe I extract a fairly high information density.

That said, this is ultimately a personalized experience, so I think there will be differences between people. Separately from that, I think it's also partly because I basically like writing densely myself. When I start a project, I always use error handling and templates, so I have a fixed form that comes to about 3,000 lines by default.

LLMs produce a lot of wrong answers, but in fact people make just as many errors. I dislike LLM's grammar, but LLM writing has a higher information density than short pieces, so when it's not narrative writing, I trust LLM information.

I dislike that an LLM can produce tens of thousands of lines in 30 minutes, but if I had done it, it would have taken a month.

I know that using LLMs degrades my skills and also degrades my ability to verify LLMs. But in the freelance market, these days contracts are made on the premise that you use LLMs.

I hate AI slop, but here I am actually trying to make a small indie game using AI.

I hate LLMs, and I also like them. I have complicated feelings about it.

Your figures show that Singapore has high output and a tight labour market. They don't by themselves tell us how the gains are distributed or whether median households face affordability pressure.

Here's my view. There are many points where the current economy doesn't feel all that good to me.

Because of you, I got curious and looked into it, and Singapore's income has risen, but consumer prices and especially housing asset prices have also risen sharply.

And this is only Singapore being strong in macroeconomics and export employment; I don't think this means the global economy is good.

Strictly speaking, I think this is because of our particular situation and the fact that we are knowledge workers. Of course, I'm not saying you're wrong. In your situation, you're right, but I think my view is more general. In other words, I don't think it's a case of the global economic situation being good, nor is it easy to say that China's economic situation is booming. I have Chinese friends, and while I don't know China as well as you do, since you're ethnically Chinese, I do know that the number of people who can leave China for abroad is very small in class terms, and I can also argue that China's statistics do not indicate such a good economic situation.

The AI debate is, in fact, somewhat ideological. The problem is that in the process, whether advocating for it or pushing for progress, there is a failure to face its limitations directly. In other words, it is emotional. Because AI's mere existence is a threat to knowledge workers, much like machines were a threat to blue-collar workers. That is, if the value of intellectual labor drops, it damages the value of the labor force through which workers earn capital in a capitalist system.

I have tried solving a few math problems with AI (they were Erdős problems), but because I know absolutely nothing about that math, I couldn't just take the AI's word for it, so I am actually a bit skeptical. A problem arises where you arrive at the answer without actually understanding it. It's not that AI is bad. The problem is that AI destroys the equilibrium between the knowledge I have and the knowledge I lack. That boundary collapses, making it feel as if I can know everything.

Actually, academia is fundamentally about mental models. It's a kind of internal worldview, and that worldview is shared. When you actually listen to the thoughts of scholars and professors, there are subtly different aspects. That forms the person's worldview... and I get the feeling that sharing it is what constitutes intellectual activity.

However, as you mentioned, unlike academics like yourself, academia and knowledge communities seem disconnected to someone like me (meaning they lack accessibility). Even if I were to make a discovery, it would probably be hard for me to become recognized, and I do think AI could actually play a role in opening up those closed communities. But apart from that, I find it hard to say that this only has positive aspects.

I followed Karpathy's research from start to finish to build a small LLM like nanoGPT on my own, and people say that because of the positive transfer that comes from feeding diverse data through modern LLM multimodal encoders, there will be new discoveries. But in reality, the types of problems AI excels at are generally those that humans have found but overlooked. In my opinion, rather than being a knowledge machine, LLMs (or AI) make me feel that what we call "intellectual activity" is closer to a kind of serialization work. A method of stacking things up one by one in sequence, so to speak? After all, the actual operating principle of an LLM proceeds according to the probability of the token coming in the next sequence.

In other words, I think the serialization method of our knowledge activities is similar to how LLMs operate, but I also think a different kind of thinking might be necessary. I am not that smart, and I have never interacted with scholars... (As you know, I am a subcontract worker. Of course, I have been hired by startups run by professors in my country, but it's not like I modeled that intellectual design myself.)

On the contrary, I feel that the evolutionary approach will slow down after GPT 6 Astra. They can continue to increase the size, but the issue lies in the cost-effectiveness of token costs.

Anyway, I agree with most of what you said in your discussion, but rather than anti-intellectualism, I consider this a direct threat to survival.

Watching the discussion unfold, here is what I feel:

Before arguing whether mathematics must strictly be done by humans, there are different motivations at play. Some people love the sense of solidarity within the community that forms during the process. Those excluded from that community might resent it, while others just purely want to solve problems.

Many things are being discussed, but looking at the overarching narrative, it seems that AI's true function isn't necessarily opening new horizons of specific knowledge, but rather excelling at 'serializing' topics that have been heavily fragmented until now.

In that sense, the concern is that because AI is solving the very problems needed to cultivate mathematicians internally, the stepping stones required for human growth are disappearing.

However, on the other hand, as the world and industries become increasingly complex and hyper-specialized, you could also argue that AI is the exact tool needed to unify this fragmentation across academia and industry. It is a highly complex dilemma.

From the perspective of researchers and the mathematical community, those 'problems for growth' must remain. But conversely, AI has the distinct ability to serialize siloed disciplines. Usually, when you go to graduate school, you often hear professors say that even within the exact same major, they cannot understand each other if their sub-specialties differ.

thanks!

That's fascinating. Does it mean it verifies mathematical proofs directly within the Rust code? Does anyone know the underlying principles of how this is possible?

Programming education in the LLM era will be different from what it is now. Many of the learning methods emerging now are the practices of a "good senior programmer."

But realistically speaking, choosing LLM programming ultimately means pouring out an enormous amount of code, and it's difficult to verify all of it. Common sense says that if you produce 10,000 lines in an hour, you can't read all of it, and even if you do read it, you'd have to rewrite it. The problem is that LLM code differs from human abstraction. Or more precisely, it lacks a programmer's habits, so it's hard for me to maintain.

Clearly, programming in the LLM era will be different. The problem is that I can't get a sense of what that way of doing things actually is.

I think that low-priority frontend work will probably be handled by LLMs, while only complex animation work will be handled by humans, and humans will end up working only on things like payment modules, which are hard to fix if something actually goes wrong.

LLMs are now better at optimization than most people.