Andy Halliday

outpace, it clips its revenues, right?

So for the 5 billion of insurance costs, you're not getting 5 billion of revenue from the people who are asking for that because a lot of it is free and a lot of it is at, you know, $20 a month and that's not accumulating very quickly despite the huge numbers of people who are, you know, using ChatGPT, you know, and they're not using it intensely through the day, you know,

Among the 800 million users are many, many people who have an account and haven't looked at it for seven days, right?

So we don't know what the distribution of utilization is, but we do know that that 5 billion cost that's money going from OpenAI over to Microsoft to pay for the inference generation is far, far greater than what it's getting in revenue.

So eventually they hope that turns around.

And that either means that inference isn't going to go

down to zero as many, you know, the cost is still there, but the price to us will continue to be very low or close to zero.

That's got to change ultimately in order for this to rebalance and actually be an ongoing sustainable kind of economy.

Well.

When you framed it as consult with myself, I had to laugh.

I was chuckling inside because I don't want to consult with myself.

I already know what I'm thinking, right?

But I do see it in the context of, you know, the progressive forgetting that happens as a natural part of our brain's processing.

And so if I could capture that,

perspective, point of view, et cetera, that was historical for me and based on experiences and the knowledge base about me is richer than my own special recollection skill, which is a special needs recollection skill, then I can see that being applicable.

I wanted to just quickly point out that the term digital twin has been used

for things other than preservation and accessibility of human knowledge.

It has been used even in the context for NVIDIA, for example, of a digital twin of your business process or a digital twin of your factory floor so that we can then

understand the interactions, both information flows and physical flows and decision points.

And that's a digital twin of a real system that's out there today in order to apply AI to that.