Andy Halliday

It's just it's illegible.

Okay, I want to talk about I want to talk about, you know, something that's, you know, really present in the conundrum of

of ai development which is the open source approach that the chinese you know labs are taking and the closed approach that the main you know frontier model developers in the united states are taking uh with the exception of the you know the

historical work that Lama represents, you know, from Meta as an open source path for AI development.

So just in the last couple of weeks, there's been a lot of news about Kimi 2 from Moonshot Labs.

And that's a Chinese startup.

And they developed this model called Kimi K2 Thinking.

And it's an open source reasoning model.

It can think

long and hard, 300 steps straight through without losing a step.

And it's basically, it matches or beats the best closed models.

So here's some examples.

It outscores GPT-5 on expert level reasoning, 45% to chat GPT-5's 42%.

It demolished humans on web research tasks, 60% versus 30%, right?

So it's way beyond human capabilities when it comes to deep research.

And it was developed in a training run that cost a total of $4.6 million, whereas GPT-5 would have cost many billions of dollars to train.

So it's better in many respects.

It's open source, so you can use it.

And this is part of the underlying pressures that caused Jensen Wang of NVIDIA recently to say China is going to win the AI race.

We had this big sort of crisis.