Penn Engineers have developed a new chip that uses light waves, rather than electricity, to perform the complex math essential to training AI. The chip has the potential to radically accelerate the processing speed of computers while also reducing their energy consumption.
So the whole chip is a complicated lens, that somehow can perform multiplication using ‘analogue computation’.
Arxiv link
Imo, analog computation is the way forward with this whole AI thing. It seems like a waste to perform calculations bit-by-bit when neural nets are generally okay with “fuzzy math” anyway.
I don’t want fuzzy math anywhere near autonomous armed machines. You want ED-209? Because that’s how you get ED-209.
“fuzzy” isn’t the right word.
I mean, I personally agree, but the military has already made it clear they don’t mind. ED-209 is basically an inevitability at this point.
Idk, maybe. But i think you may have issues with tolerances and reproducibility. With analog and neutral nets your going to have edgecases where some devices will give vastly differing outcomes. For something that’s fine but not for others.
Sounds like we’re building a true android then.
So first we tricked rocks into doing math, and now we’ve figured out how to trick glass into doing math? This is truly amazing.
We tricked the same rocks we use for doing math into bending light like the glass, and we use that for doing math, yes.
Reality is mind boggling sometimes.
Digital is also analog.