The Chinese models were also developed later and allegedly didn't have to focus so hard on the initial acquisition of training data, so were able to build their models completely differently.

Think of it like developing dictionary technology. For the first dictionary, the hardest part is finding all the words that need to be defined. After that it's just a matter of adding new words occasionally as they're found.

That's the "theft" the Western AI companies accused the Chinese companies of doing. By starting with the initial training data from Western devs, the Chinese AIs (allegedly) skipped over the hard part, which means they are able to focus more of their processing on the end product.

And that's not necessarily a bad thing, depending on your perspective (ignoring all the ethical questions around AI data acquisition). Duplicating work someone else has already done is wasteful.

But it's also disingenuous to say that the guy who takes an existing dictionary and changes all the definitions did a "better job" than the guy who wrote the first one, even if the end product is, in fact, better.