Impossible colors are so cool.
‘“I would say you might get olo by staring at a saturated red spot and then looking at a green screen of a very saturated bluish-green,” he told me’
This is well established and easy to do, eg https://www.skytopia.com/project/illusion/ipage-et.html or https://en.wikipedia.org/wiki/Impossible_color#Chimerical_colors. I wish they had interviewed someone who had tried both.
On the one hand, sure, models generalize. It’s cool they established that.
On the other hand, the copyright argument seems to be that they stole so much data that the contribution of any individual image rounds down to zero, therefore it’s not infringing? Wha??
Are they really claiming that one infringement is a a tragedy, but a million infringements is a statistic?
Noslopgrenade also reads like a slop grenade
There have been more impressive studies showing adaptation to completely new limbs like an extra thumb or a tail but this seems like it was more focused on pinning down the brain structures responsible.
I wonder if you could trigger a rubber wing illusion this way?
Could be population density? Cities are bluer and if they only had to evacuate 5 people instead of 50,000 it wouldn’t have made headlines
The server still has to send its own key, so that part is bigger. They’re making up the space by replacing traditional CA signatures and intermediate certificates with merkle tree hash-based signatures.
Raw number of bytes can’t be directly compared between cryptography schemes, cryptographic hashes make very efficient use of space and don’t have the same quantum weaknesses that old public key cryptography has. The quantum-unsafe RSA signatures being replaced are about 256 bytes, the new ML-DSA signatures are about 2400 bytes, and SHA-256 Merkle tree references are 32 bytes for the same level of security and quantum-proofness.
Total addressable market is the amount of revenue they think is available to all AI providers, combined. So it makes sense that this is in the same ballpark as SpaceX’s “enterprise services” estimate.
And yes they believe 25% of the economy can be automated with AI. Which is a big downgrade from 100%