Hacker News Reader

Super-compression: document description + fingerprint = original document.

Ie.: [white caucasic woman smiles arms crossed, dark dress year 1500, background desert mountains road lake forest, match F8A91DDEB45C39] produces La Gioconda loselessly up to 512px wide.

View on HN

That doesn't sound like compression, that sounds like a way to look up a stored picture from a massive dataset that's been turned into neural net goo. I'm guessing the neural nets that can do that are WAY bigger than a 512px wide jpeg and a jpeg renderer.

It's still cool, but calling that compression is like calling a link to a website compression.

Reply on news.ycombinator.com