Andy Halliday

And, you know, there's,

enormous investment being made there.

But I wanted to just tell you that I came across a news item, and I haven't read the actual study that's published in the Nature Photonics Journal, but there is a breakthrough in optical computing for chips.

And one of the problems with lasers as the light for, instead of electricity and electrons running through the chips,

is that it's difficult to make it happen in parallel.

So you can send multiple electrons down different paths through a GPU, but when you're using a laser, you're putting a burst of laser light in and you either have to have an enormous array of lasers to multiplex that or whatever.

So anyway, there's these constraints that have been true

about optical computing that make optical, which is faster and better in many, many ways in terms of power consumption compared to electronics.

So anyway, there's the results of an experimental optical computing prototype with a series of comparative tests against standard GPU processing schemes.

And it kind of blows away the potential of GPUs.

And what they did was they used a conventional optical hardware component alongside a novel encoding and processing method so that they could capture and parse tensor packages in a chip

with a single laser shot.

So multiple tensor packages, a single laser shot through the entire optical package and multiple outputs through the standard electronics.

So this is a breakthrough in just concept about how you manage to get an optical computing chip as opposed to an electronic one.

It dramatically reduces the energy and the side effects of energy burning through copper.

I feel live and recording 608.

So before we move over to Anthropic and their release of Claude Opus 4.5, which kind of blew everybody else out of the water,

I will be astounded by just the rapid succession here in a matter of weeks of the displacement of the top frontier models as one releases after the other.

Wow.

You know, if we thought that there was a ceiling for