The Era of No One Knowing Again

Chapter 9

A new word came up during the meeting.

A newly joined developer was explaining the materials and said,

"I think it would be better to go with RAG."

I didn't misunderstand.

I knew what RAG was.

However, I wasn't sure why this function was needed.

The youngest developer asked right away,

"Why are we using RAG here?"

I raised my head.

The question came out too easily.

The developer who was presenting said,

"Oh, that's because external documents change frequently, so it would be better to retrieve them when needed rather than putting everything into the model."

The youngest asked again,

"Then wouldn't it be enough to just attach a search function?"

The meeting room fell silent for a moment.

I was secretly surprised.

Was it okay to ask like that?

The developer who was presenting didn't seem too upset.

"That's right. In terms of structure, it's similar to search. However, the model doesn't just show the search results; it uses them to create an answer."

"Ah."

The youngest nodded.

Then tapped on his laptop.

A few seconds later, he said again,

"But if the search quality is bad, wouldn't the answer be strange too?"

"That's right."

"Then we need to look at the embedding model as well, right?"

"Yes. That's important."

I glanced at the youngest developer's screen.

The AI window was open.

He was typing in the words that had just come up in the meeting.

Explain the difference between RAG and a general search simply.

What is the impact of embedding model selection on search quality?

I looked at the screen again.

That friend didn't seem to have any intention of hiding what he didn't know.

He asked questions during the meeting, immediately asked AI again, and if the answer was strange, he asked again.

I couldn't have done that before.

No, to be precise, I didn't.

When I didn't understand something, I would write it down in a notebook.

I stayed still until the meeting was over.

If I could pretend to know, I pretended to know.

At an appropriate time, I nodded my head, and when someone asked "Do you know?", I mostly said "Yes".

After the meeting, I would go back to my seat and search.

If I couldn't find it even after searching, I would look for a book.

If I still didn't understand, I wouldn't ask anyone until then.

That order rarely changed.

I hated being caught not knowing.

Especially when someone with a higher academic background was in front of me. I graduated from a vocational college, and they were from well-known universities.

No one ever brought it up,

but I was always conscious of it.

If I asked a question wrong, I thought I would see an expression like "As expected". I don't remember seeing that expression for real.

Maybe not.

But even if there is no expression, imagining it for a long time makes it feel real.

When the meeting was over, the youngest developer came to my seat.

"Senior."

"What?"

"About RAG earlier."

"Yeah."

"Isn't it enough to just do a good search?"

I laughed.

"Didn't you ask that earlier?"

"I heard the explanation, but I still don't understand."

"Then ask more."

"Who should I ask?"

I pointed to the youngest developer's screen.

"Him."

The youngest laughed.

"I already asked him."

"What did he say?"

"He explains so well that I'm more confused."

I laughed too.

That made a little sense to me.

AI can explain what you don't know, but explaining well and making someone understand are different things.

The youngest pulled up a chair and sat down.

“Senior, how did you learn this kind of stuff at first?”

“This kind of stuff?”

“When new technology came out.”

I thought for a moment.

“I bought books.”

“Books again.”

“Why?”

“They say old developers used to buy books all the time when they heard about a senior.”

“Because there was no information.”

“What if there wasn't an answer in the book?”

“They asked people.”

“What if they didn’t know either?”

“They just tried it.”

The youngest nodded.

“It's similar now, isn't it?”

“What is?”

“We ask AI, and if that doesn’t work, we ask people. And even then, if we don’t know, we just try it.”

I didn't answer.

I had never thought of it that way.

I thought I had worked hard to learn back then.

I thought these kids were getting answers too easily.

But when you looked at the order, it wasn't that different.

The only difference was the speed.

It used to take us a day to find one answer, but now it takes a few seconds. The problem is that if the answer comes too quickly, you don't have time to think about why that answer came up.

The youngest spoke again.

“But I think it’s better to just ask when I don’t know.”

“That's right.”

“Pretending to know can lead to bigger problems later on.”

His words stuck with me.

I had pretended to know many times before and made mistakes.

I hid what I didn't know and didn't ask questions, and sometimes I misunderstood and developed something wrong.

At the time, I just thought of it as a mistake.

Looking back now, it was often a matter of pride.

The youngest stood up from his seat.

“Anyway, I'll take a closer look.”

“Okay.”

“I’ll ask again if I don’t know.”

“Who?”

The youngest smiled.

“Both of you.”

He alternated between me and the AI.

I waved my hand.

“Go work.”

The youngest returned to his seat.

I watched him for a while.

I wasn't like that when I was young.

If I didn’t know something, I hid it. If I was wrong, I hoped I wouldn’t be found out. Next to someone who knew more than me, I became even quieter.

I thought it was humility at the time.

Actually, it was fear.

The moment I admitted that I was lacking, it seemed like someone would figure out whether it was because of my education, my ability, or both. These young developers didn't seem to care much about that.

At least the few I saw were like that.

If they didn’t know something, they said so. If they needed something, they looked for it right away. And once they got the answer, they moved on.

Sometimes I was annoyed that they were moving on too quickly.

I often saw them just copying and pasting code without understanding why.

“Why did you do it this way?”

The answer was always similar.

“Because the search results said to do it this way.”

These days, there was one more thing added.

“AI wrote it like this.”

When I heard that, I sighed inwardly.

It seemed like they were taking development too lightly.

Some of them didn't seem interested in the technology itself and just wanted it to work.

I once asked them over lunch.

“Is development fun?”

He answered without hesitation.

“It’s just a job.”

“But don't you like doing it?”

“I'm doing it to make money.”

He had such an obvious expression.

I was speechless for a moment.

I almost said that we weren’t like that back then.

But when I thought about it, we were doing it to make money too.

I also changed companies because of my salary, and I was swayed if there was a place that paid more. There were times when development was fun, but it wasn't always fun every day. At the time, though, I couldn't work without knowing technology, so I had to understand whether I liked it or not.

Now, even if I don't understand, there are more things I can make.

That difference was huge.

In the afternoon, the young developer posted a code review.

I read the code and stopped at one point.

It seemed like it would work.

But I was curious why he did it that way.

I commented.

Is there a reason why you chose this method?

A few minutes later, the answer came.

AI recommended this pattern and I applied it.

I lifted my hands from the keyboard.

I thought for a moment.

Please also check why this pattern is correct.

I was about to write it, but I deleted it.

Instead, I wrote this:

Please ask the AI what situations this code could cause problems in, see the answer, and then judge again.

I sent it.

A little while later, the answer came.

Ah, in this case, there could be a problem. I will fix it.

I looked at the screen.

It was a little strange.

While I thought I couldn't trust AI, I was teaching it to ask better questions.

It was getting close to quitting time.

The youngest member got up from his seat and said,

"Senior, I'll go first."

"Okay."

"I fixed that earlier."

"I saw it."

"Thank you."

The friend picked up his bag and left.

I looked back at the monitor.

The questions I asked that morning remained in the AI window. RAG. Embedding. Search quality.

I typed one sentence into the input box.

What's the difference between quickly asking what you don't know and studying to understand it yourself?

The answer started appearing after a while.

I didn't read it.

I deleted the question.

It didn't seem like a question to ask AI.