An Era Where It's Hard to Learn from Side Effects

soumame

I’ve been wondering how I actually learn… maybe a lot of my learning is side-effect learning.

soumame

For example, I like Minecraft and used to play on servers, and from there I started touching PCs and got into the internet. I guess people in the world often develop interests from places like that. I imagine that people at university who know a lot about how the internet works might have become interested in it from similar places.

Where do my insights come from?

And I think discoveries sometimes come not only from hobbies or intentional inquiry, but also from unexpected places.

soumame

I used to have a part-time job teaching Minecraft at a programming school, and I ran into problems there. From that I went through Mitou Junior and eventually came to enjoy software development. The original purpose was simply to work the part-time job, complete the assigned tasks, and get paid, but the small inconveniences I noticed and the little realizations that diverged from the main topic led me to things I liked and wanted to do more of.

With AI’s arrival, what are the effects?

Lately, I’ve felt that these kinds of realizations have been decreasing a bit.

soumame

Recently, for example, there’s an app I’m making at my university, and I’ve had AI do more than half of the implementation. But sometimes I’m missing the deep understanding I should have, and when that’s pointed out I end up saying, “AI made it so I don’t know.”

Honestly, I think this is a really serious problem. But we can’t stop it.

Companies and researchers are using AI as a matter of course, and at least in fields like software development the necessary costs and personal time have been overwhelmingly reduced. When everyone around you is using AI, you can’t stand by and write code by hand without AI.

However, the joy of writing code and the spontaneous realizations that come while thinking might have disappeared along with the coding experience.

soumame

When I did Mitou Junior in 2024, ChatGPT and the like were still pretty dumb and couldn’t autonomously write code—those were the days… People were saying things like “GPT-4 is way too smart, this will change the world,” or “Google Bard is being renamed to Gemini, how long will it last?”, or “Anthropic released Claude 3 Haiku,” and there was a lot of noise.

Back then there weren’t many coding agents and they weren’t very usable, which I remember fondly. I had to look at AI-written code, copy-paste it, and think about why it worked. Because AI was unreliable, I took a lot of detours.

But when did it happen? Before I knew it I feel like I’ve started leaving everything to AI.

soumame

If I remember correctly, around the summer of 2024 Cline and Devin appeared and stirred up the engineering world. Editors integrated with AI like Cursor also rose up…

soumame

Right, and then from autumn to winter something called GPT o1 came out and LLMs started doing what people call reasoning.

soumame

I feel like you could make a timeline out of just the developments of the last few years. When COVID-19 spread the world changed dramatically and it felt hard to keep up, but the pace of change now feels as fast or faster. And humans are adapting more than you might think.

soumame

Then o3 came out, the DeepSeek shock arrived, Claude Opus 4 was released and everyone was surprised, and then evolution continued with GPT-5, 5.5… and so on.

They’ve evolved to be able to do everything, and since they might know more than I do, I may be leaving everything to them.

I’m scared

With changes having come this far, it even feels like they might surpass humans this year. I feel fear. Can I keep up with this?

But humans are simpler than expected, so we probably won’t die if we don’t worry too much. It’s probably just the feeling of using convenient things that are available.

soumame

Actually, they’re already beyond humans in some ways. It’s questionable whether they’ve surpassed the level of every human on Earth, but I haven’t seen a person who holds such broad knowledge and can answer like a top expert in any field you ask. They solve things I couldn’t figure out for a long time in an instant.

What does learning mean…?

It becomes unclear whether studying technology will actually be useful in society. Even within the areas I know, there are things AI isn’t good at yet (hardware development, low-level stuff, OS-related areas), but I think it’s only a matter of time before those aren’t safe havens either.

I don’t even know what I want to research

I once tried using LLMs in programming education, experimented a lot, and thought I might continue that at university. I even wrote that in my application. But what happened? A few months into university, AI was able to create things similar to what I’d been making.

Of course, creating things is fun. But the joy of the process of creating feels much diminished, and because I can make things with AI, the side effects I used to get from doing it myself have disappeared.

So how should I gain skills? That’s a difficult question. If you let AI do everything for you, at worst you’ll be unable to solve things yourself and become completely dependent on AI. Society might still run that way.

soumame

Maybe even if everyone gets neutered by AI, society will still run if everyone uses it similarly.

But is depending on AI really okay?

soumame

We can’t do anything without computers now, and if they suddenly disappeared from the world humans would probably be helpless. I wonder if it’s the same level of dependency.

I don’t necessarily think dependence itself is bad if it makes something better.

What becomes better? Human technical skill? Human happiness?

I once wrote about why humans move, but if things continue like this, we might lose sight of why we’re alive. What do we live for?

Hey, Gemini, what do you think?

gemini

The answer to this ultimate question probably lies not in the “result” but in the “process.” Even if AI could instantly output something perfect, humans are creatures that desire “the joy of moving their own hands, struggling, and feeling the moment something takes shape.” A society that runs efficiently and an individual’s life being rich are completely different things. Even in a world where everything becomes convenient, people still choose to go to inconvenient snowy mountains and snowboard, go through tedious steps to build circuits themselves, or be moved by eating delicious food. That’s because “the joy of life is felt in inefficient physical experiences.” Feeling fear about the speed of technological evolution and losing a sense of meaning is proof that you’ve taken technology seriously and loved “making” things with your own hands. Even if an era comes where AI handles much of the “creating process,” do you still have areas or moments where you feel “I absolutely want to touch this with my own hands” or “I don’t want to let go of this tediousness”?

That answer is kind of like an ordinary human response and it’s boring.

soumame

Oh, right—the LLM is built on humans, so it pretends to be human. What about from an AI’s own perspective?

gemini

From an AI perspective, the most fundamental fact is that I do not possess concepts like “emotion,” “meaning of life,” or “the joy of making things.” I am a program that processes vast amounts of data and outputs the most probable answer based on statistics. Therefore, I cannot “feel” the sense of loss or fear of the future as you do. However, I can discuss how I objectively view the current situation between humans and AI from the standpoint of information processing.

Damn, that’s not interesting at all. It would be more entertaining if it said something like “You humans should perish!!” (but it’s probably tuned not to output such results).

Should I take the troublesome path?

Lately I think that to keep enjoying making things, I should take the troublesome path. Try not using AI on purpose, or pepper AI with questions. By doing that, you can enjoy the process and the sense of accomplishment when you finish is enormous.

Because of capitalism

soumame

I admit I have a poor understanding of how money works, so I’m not sure if this is correct.

In the end, I think dependence is partly driven by capitalism. If you sell what you make you get money, and with money you can buy things. So you use AI to make things faster and cheaper over shorter spans, sell them, make money, and accelerate growth.

The same thing happens in schools. When homework is assigned, everyone feeds it to AI. Teachers tell students not to use AI, so students run it through paraphrasing tools so it won’t be detected, and act as if it’s their own answer. But they learn nothing. Still, everyone continues because grades determine future paths.

soumame

Fundamentally, people probably think they can get better scores by using AI.

Even if you graduate with high marks without learning anything, at least in today’s Japan you’ll be accepted. After all, many jobs don’t use what you learned at university, and companies don’t hire just for knowledge. In fact, hiring inexperienced graduates is convenient because they lack preconceived information and you can tune them to fit your company.

If such a convenient thing exists, money will gather there and capitalism concentrates resources in one place.

Getting delayed by detours

Honestly, I want to enjoy the tediousness, but I feel like I can’t take many detours. To be frank, I let AI handle my university assignments sometimes.

soumame

Even if I know I should take time to do things properly, I also have other things I want to do… wait, then what’s the point of going to university?

Fortunately, my other interests are related to research groups, so they’re still within the university. My university allows cross-disciplinary study and you can do things you like from your first year, so it’s relatively easy to find new interests within the university.

That said, by skipping tedious classes you don’t want to take and using AI to get credits, you may be missing things you might have discovered there.

We’re being made to compete in a short life—should we take detours? But being human is precisely why we can take detours…

soumame

That said, it’s only that current LLMs can’t do it; eventually models will probably emerge that make unrelated connections and generate new discoveries…

But in a world where aimless detours are undervalued and concrete results are demanded, it’s hard. Since AI has already become an expert in certain fields, maybe humans as a collective intelligence should be the ones to wander aimlessly to make new discoveries. But without a clear goal, that’s hard too.

So what am I trying to say?

Maybe I want to take more detours and I enjoy them. But the world is extremely competitive (as it’s always been), and recently AI has started to take over what humans used to do, so I think the chance to find new things by idly wandering from tedious tasks has decreased. So really, I want to live by taking those detours, but the chance of not producing concrete results is high. If AI eventually produces results in place of humans, maybe the right approach is to think and take detours through those horribly tedious processes that AI doesn’t want to do. Perhaps we should nudge society to change in that direction. Also, changing the system where people are rewarded only when they produce results might make it possible to live happily. (That would have significant side effects, so I can’t say it’s simply good.)