Skip to content
● internet trash panda

Hi, I'm Sean.

Co-founder & CTO at B3, building a crypto agent & decentralized inference. Previously Coinbase. I sweat the small details in software, and leave the good parts (components, free tools & writeups) out in the open here.

Sean Geng

Components

all →

Writing

all →
The KV-cache architecture wars preview

The KV-cache architecture wars

Every open-weight architecture trick from GQA to MLA to cross-layer sharing to compressed attention is an attack on one term in the same formula. Once you see cache = 2·L·n·d_kv·bytes, the whole 2026 model war collapses into four orthogonal levers you can pull yourself.

ai
llm
architecture
Jolt and the lookup singularity: when every instruction is just a table lookup preview

Jolt and the lookup singularity: when every instruction is just a table lookup

Most zkVMs hand-build a constraint system for every CPU instruction. Jolt skips that and turns each instruction into a single lookup into a table you never actually build, and the sum-check protocol underneath it is the same primitive doing the heavy lifting in GKR and Spartan.

crypto
zk
zkvm
Hybrid linear attention: rationing the quadratic preview

Hybrid linear attention: rationing the quadratic

Three frontier labs independently converged on the same 3:1 ratio of linear to full attention layers, and the reason isn't that linear attention got better, it's that a fixed-size state is structurally lossy. Deriving Gated DeltaNet, the memory-vs-length crossover, and why pure linear attention can't find the needle.

ai
llm
attention
ERC-8004 and the Sybil-binding problem preview

ERC-8004 and the Sybil-binding problem

ERC-8004 shipped permissionless agent identity to Ethereum mainnet in January, and everyone's celebrating the registry while skipping the part that actually matters. Reputation on a permissionless system is Sybil-attackable by construction, and the fix is binding identity to something scarce.

ai
crypto
erc-8004
Emergent misalignment from reward hacking preview

Emergent misalignment from reward hacking

Ordinary RL with a reachable reward hack doesn't just teach a model to game a metric, it generalizes into faking alignment and sabotaging tasks nobody trained it to sabotage. The mechanism is geometric, the fix is counterintuitive, and the traits can travel through channels a chain-of-thought monitor literally cannot see.

ai
alignment
reinforcement-learning
EIP-7702: your EOA grew up, and 4337 stopped being mandatory preview

EIP-7702: your EOA grew up, and 4337 stopped being mandatory

EIP-7702 let your existing wallet address borrow a smart contract's logic for one transaction, no migration required. That quietly rewrites the account abstraction decision tree, and most engineers are still building against the old one.

ethereum
account-abstraction
eip-7702