Learning With Errors (yes, the lattice assumption) is a corner of the site where I explain technical concepts the way you'd actually pick them up: as a chat, at a human pace, with inline demos you can poke. The hard explanations are real and sourced (often verbatim from people who know); the conversational framing and the demos are AI. Every chat now signs off with a pop quiz: reading feels like understanding until someone asks, and the asking is the point (Geoffrey Litt makes the argument). Pick a buddy to start a conversation.
- FHEcompute on encrypted data without ever decrypting it chat
- MPCsplit a secret across nodes: honest majority, traitors and all chat
- TEEsa secret room inside the chip, and the side-channels that leak it chat
- UTF-8ASCII vs UTF-32 vs UTF-8: bytes, code points, 中文 chat
- Vigenèrethe cipher behind Kryptos K1-K3: keystreams, autokeys, cranking the noise chat
- thumbnail encodingAVIF vs JPEG vs jpegli: bytes-per-pixel, from the garage study chat
- programmable cryptoZK, MPC and FHE as one toolkit: the 0xPARC view chat
- DACsnumbers into sound: ladder DACs vs noise-shaping, and the audiophile fight chat
- driversplanar magnetic vs dynamic: point drive, distributed drive, cone breakup chat
- knotsgranny vs square, the Ian knot, and why your shoelaces slip chat
- Leanformal verification: the spec is the weak point, and the kernel had a hole chat
why "Learning With Errors"? it's the lattice hardness assumption that FHE and modern lattice crypto are built on: secrets hidden in deliberately-added noise (errors). it's also just how learning works. · more experiments live in the garage →
