Alice Han

because it's cheaper and faster, seems to suggest that the use cases for this are very different, the Chinese models that is, than the American closed models.

Firstly, a lot of engineers and just everyday people are using and experimenting with the Chinese models.

Quinn, for instance, they're downloading them locally, fine-tuning them and using them for their particular use cases, any kind of projects that they may be involved in.

It's got improved privacy because they're downloading it locally.

It's cheaper, often 10 to 20 times cheaper than what's offered by OpenAI and Claude.

So really, they're using this very differently from, say, potentially you and me, Ed.

I use Claude a lot, the enterprise level, because the reasoning is exceptionally good when it comes to the geopolitical analysis, the financial analysis that I'm involved in.

So the way that I think about it is that the American models are really focusing on precision, on reasoning at the highest level as a really quest for AGI.

The Chinese models are really figuring out how to create a market in which everyday people, engineers, corporates even, are downloading locally, experimenting, fine-tuning, and just capturing the benefits of cheaper, faster products.

models.

So I don't think there's a direct comparison.

I think that the market exists for both these models, so to speak, to coexist.

Well, I love that you asked that question, Ed.

I think often that the benchmarks can be red herrings.

There's a good Harding principle in computer science that is applied to this, which suggests that even although they may meet those benchmark goals and exceed the other models, there are certain unquantifiable goals and performance outcomes that they may still be inferior in certain respects to the clods and the open AIs of the world.

So again, back to my previous point, Ed, I really think

that if you care about privacy and cheapness and speed, maybe you go for Quen, you go for Jaipur, you go for Kimmy.

But if you care about frontier, cutting-edge reasoning, precision, then you are still going to favor the clods and the open AIs.

So it's really different universes that they inhabit.

And I think benchmarks can often be red herrings, even although...