Everyday Shikinen Sengu
from/villagepump/2025/04/10
(Context: vibe coding's story about /villagepump/blu3mo.icon's success in prototyping a new system with vibe coding)
If a system can be implemented in this short time, the system willmodular にしつつ頻繁にスクラップアンドビルドを繰り返すようなやり方の方がうまくいくことが多そう/villagepump/blu3mo.icon
Everyday Shikinen Sengu
I've got a great power word here./villagepump/nishio.icon
Yeah, I guess that means that once you get the abstract "knowledge", you can discard the concrete "source code".
Exactly blu3mo.icon*2.
Paradigm Shift./villagepump/基素.icon
Free yourself from the weight of writing code.
The dream of a world where all you have to do is knead abstractions is about to arrive./villagepump/sta.icon
kneading in abstract form only じゃなくて「AIの支援で速やかに具体を生成して、そこから新しい知見を得て、具体は捨てる」だと思う/villagepump/nishio.icon
+1/villagepump/blu3mo.icon
"Putting the generated concrete into the real world to gain new insights."experiment の作業が本質になってきそう/villagepump/blu3mo.icon
The value (or fun) is going to be in being able to do experiments that other people can't do.
I was looking down from the abstract (the heavens) to the concrete (the earth), and I imagined dropping it on the earth, letting it do all the work, observing the results, kneading the abstract, and dropping it again./villagepump/sta.icon
But realistically, I guess I have to come and go from the ground myself to some extent.
Even with cloud and IaC, you have to set the detailed parameters yourself.
I'm starting to think I've written something similar as an unexplored junior review memo./villagepump/nishio.icon
Proposals that are kneaded in the abstract alone are usually rejected.
Type-specific advice for contest applicants
It comes first as Pattern 1.
The type that just says abstract things and doesn't lead to anything concrete at all.
I also remember Mr. Ochiai saying something like, "You should have already done that."/villagepump/sta.icon
People who have never wanted to make it and have never made it.
The requirements for "inventions" handled by patents also require a specific way of making them./villagepump/sta.icon
In kneading abstractions, it seems that "the knowledge/experience/attachment/value system/ring world that only the person has" can only dwell from the concrete./villagepump/blu3mo.icon
I believe that the "knowledge/experience/attachment/values/world of the ring" that only the person has is "interesting./villagepump/blu3mo.icon
消すことが大事nishio.icon
If you have old code in the repository because you feel like it's a waste, the AI will read it and get confused.
I complained to AI when it was accidentally deleted before, and they went back to the Git log, found the deleted commit, and restored it.
So if it's in Git, you can delete it without mercy, and if you're worried, just make a note of the commits you've deleted.
I think there's a lot of confusion right now about both AIPM and the approach of having Devin do the research, and that's the problem.
Delete or create a new repository and move only what you need.
Related: Human Work: AI Ruthlessly Throws Away What It Writes
天上↔泥臭い、 smell of earth nishio.icon
In an age when linguistic thinking has been made more efficient by AI
Processes with language as input and language as output become thicker.
If that was the bottleneck, the bottleneck is likely to move elsewhere
Related: When one bottleneck is eliminated, another bottleneck arises.
One of the places to move to is "the process of making what is not yet a language into a language."
In situations where observing non-verbal things is a source of value, there is a big yield difference between observing up close and observing at a distance.
+100blu3mo.icon
Quantity is also important
N=multiple observations and then abstraction.
Abstracting from N=1 observations (compromised)
N=0 abstraction (just a delusion that is not grounded in concrete reality)
Related: Decisions supported by multiple perspectives.
It may no longer be important that the "sample is accurate and unbiased"blu3mo.icon
There seems to be a difference between "aiming for" and "waiting to be secured."nishio.icon
I'll have to continue to work on my footwork to observe up close./villagepump/sta.icon
I thought I didn't need it anymore, but today's exchange reminded me that that's not going to work.
getting used to the soil.
足を使って情報を集めろ、といった表現を思い出す。古い時代の刑事役が言うセリフのイメージだけど、結局は自力で情報集めないとダメというやつは普遍的な気がする/villagepump/Hiro Aki.icon
Related:.one must see with one's own eyes nishio.icon
Shikinen Shikinen Refactoring
construction of a new shrine
Metanote.
Replanting Notebook
You can't just skip the superficial knowledge and get a deep understanding.
You can't learn directly from something that is "hard to change."
Prototypes were originally intended to be thrown away, but people often continue to use them, saying, "I made this prototype and it's a waste of money.
Operation is the bottleneck
AI ideas are not grounded. (2024-09-01)
When AI starts doing muddy trial-and-error 100 times faster than humans, we're in trouble.
It's getting closer! (2025-04-15)
---
This page is auto-translated from /nishio/毎日式年遷宮 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.