Jyunmi Hatcher

Versus kind of what I've been looking at is,

for my purposes, if I have smaller models that do specific things and I can orchestrate them, or if I have just enough hardware, like a tweaked out Mac studio, right.

Can do 170 tokens a second can run.

It's got enough memory in there and enough unified memory in there to run a, you know,

hundreds of billions or maybe even a trillion parameter model.

And then if I cluster those, right, that becomes 40K, right?

But if I'm at that level where I just buy 40K and it's like, yeah, okay, there you go.

I just run one in my closet, right?

I would never put it in a closet because, you know, it probably, but we're just saying, you know, you could just run run of those things.

You kind of have to start taking the mindset of a individual crypto miner.

I know maybe not necessarily the best reference, but

You build your own rig.

You got something like 8, 10 GPUs on there.

It's essentially clustering a bunch of hardware.

And is that...

Is the ROI going to be enough for you because you're using your own tokens?

As things get cheaper and cheaper for token use, we're going to rely more and more on the token use.

So has the foundational AI models, hardware providers like NVIDIA and data centers, have they just created a different model?

currency in terms of time we don't like to use hours anymore we like to use deliverables right but time is the ultimate measure of currency because everyone has the exact same amount of time per day so tokens are just the new medium right

It's suspicious.