Like people are familiar with DSP programming, they might be aware of this, but maybe you can talk kind of high level first about, you know, what is LNS and how do we apply it to AI?

Yeah.

So LNS, first of all, just has that basic idea of, you know, instead of representing numbers like integer would, where, well, you have an integer and a fractional part, or how floating point represents numbers, which is you have an exponent, it's an integer exponent, and then a mantissa, which is basically one plus the mantissa times two to the power of the exponent.

Let's just say you only have an exponent, but the exponent, can also have numbers after the decimal point.

So it's not just 2 to the power of 4 and 5 and 6, but it's 4.2 or even 4.27345. And the advantage from that is that you can just add two of these exponents and you have effectively multiplied two numbers.

So what normally costs the most, which is multiplication, suddenly becomes what normally is the But then, well, then the real interesting stuff starts to happen.

So you then have to convert that number, that basically multiplication result, back to a linear representation.

So I'm curious about why you picked this 72 number.