CODE IN TOKYO

Watching the DeepSeek Distillation Controversy: Is It Theft to Steal from a Thief?

Watching the recent DeepSeek controversy, I can’t shake a strange feeling.

DeepSeek is said to use a technique called “distillation,” learning from the outputs of high-performance AI to build smaller, more efficient models.

In response, some Western AI companies and researchers have strongly criticized this as:

Of course, I understand the logic.

Building a massive AI requires enormous GPUs, electricity, researchers, and capital. If a latecomer can shortcut all of that, the pioneers naturally don’t find it amusing.

But watching this debate, one phrase keeps coming to mind.

“Is it theft to steal from a thief?”

I admit it’s a fairly provocative way to put it. Still, looking at the AI industry as a whole, the unease won’t go away.

What did the giants learn from in the first place?

Today’s giant AI models are trained on an enormous amount of data from across the internet.

Blogs. News. Forums. Illustrations. Photos. Source code. Among it is a vast amount of material whose creators never imagined it would be used to train AI.

Of course, each company argues for legality on grounds such as:

But at least on a gut level, part of it looks like this:

“They grew powerful by vacuuming up the world’s creative works with a giant vacuum cleaner.”

Then along comes a latecomer like DeepSeek, saying:

“We learned from that AI’s output.”

And suddenly the response becomes:

“That’s our intellectual property.”

No, I get the logic. But at the same time, there’s a certain mood in the air:

“Weren’t you also doing some pretty bold vacuuming at the start?”

Who owns knowledge?

Of course, it’s dangerous to oversimplify this.

The original data providers and the AI companies. The AI companies among themselves. Legally, these are entirely separate issues.

So no one is really a “thief.”

Still, across this industry there is unmistakably a shift in values driven by which side you’re on:

“When I’m the one doing the absorbing, I call it ‘learning’; when I’m the one being absorbed, I call it ‘infringement.’”

And the essential point of contention is probably deeper than the law.

It comes down to this: Who owns knowledge?

We’ve seen this movie before

Personally, I think this setup closely resembles the early days of the internet.

Back then, too, debates dragged on endlessly:

In the end, we landed neither on pure idealism nor on pure regulation, but in a strangely ambiguous world where giant corporations and open culture mixed together.

AI will probably end up the same way.

Except this time, the fight is over knowledge itself.

So the scale and the intensity are on another level entirely.

Less revolution, more resource war

Either way, the AI industry is fascinating to watch.

For all the talk of the future, of democratization, of human progress, what’s actually flying around is:

In other words, extremely down-and-dirty stuff.

Less a technological revolution, more a massive war over resources.

And the DeepSeek controversy, I think, made that contradiction almost embarrassingly visible.