Post by @vitalik • Hey

Since a lot of people are interested in fully homomorphic encryption (FHE) recently, re-upping my article from 2020 that goes into how the math works in de

Stats

Comments

Also a toy implementation of one scheme here: https://github.com/vbuterin/research/blob/master/tensor_fhe/homomorphic_encryption.py
Haven’t fully disgusted what you posted. But can an algo get fully trained on an encrypted data? I get you said it can do basic arithmetics. But what about operations like checking Boolean expressions, like pairwise comparison?
How should we be thinking about the timeline for sufficient compute on L1s or L2s to make this economically feasible?
This would unlock incredible amount of new use cases. In a trustless manner
Thanks for your insight!
Thank you so much 👩💻👩💻