5 ms·
"four-order-of-magnitude improvement in the accuracy of matrix multiplication"
by Kalanos 4y ago
"four-order-of-magnitude improvement in the accuracy of matrix multiplication"
- derbOac 4y agoWouldn't this depend on what's in the matrices? I've done work with big numbers and this format screams problems. I can see how it would be great depending on the domain, and wish there was more diversity out there in the wild in this area but this seems a little hyped? Maybe I'm missing something. I'm not saying this doesn't seem useful, just that the article seems to be presenting it as a replacement for floats, when it might be better thought of as another option?