alper.sol

- Super cool project

[alper.sol @alperr_sol]

Smart contracts can't compute.

Even Black-Scholes, one of finance's most standard formulas, reverts on-chain as "too expensive."

The same wall blocks the big one: verifying AI outputs on-chain.

My research project at EPFL DCL, Wastral, solves it. 🧵

[alper.sol @alperr_sol]

Sources:
- zkVM 100,000x-1,000,000x overhead + zkML proving times: EZKL benchmarks, arXiv 2502.18535
- Arbitrum: zero fraud proofs since launch (Cointelegraph)
- Wastral design + code: https://t.co/RS4ilSssdA

[alper.sol @alperr_sol]

Wastral runs your algorithm off-chain, then settles any dispute by re-executing a single instruction on-chain.

The obvious question I get: why not just use ZK proofs?

I started there too. Then I hit a wall. 🧵

[alper.sol @alperr_sol]

The tradeoff is real: a challenge window before finality, and you need one honest watcher. ZK gives instant finality.

I'm betting generality + a near-free good path wins for off-chain compute.

Let me know what you think.

Building it in public: https://t.co/9c7x42wSga

[alper.sol @alperr_sol]

I'm building all of it in public from here.

Website: https://t.co/5mcEZsAxqM

Follow along if fraud proofs / ZK / verifiable compute / on-chain AI is your thing. Devlogs incoming.

[alper.sol @alperr_sol]

Smart contracts can't compute.

Even Black-Scholes, one of finance's most standard formulas, reverts on-chain as "too expensive."

The same wall blocks the big one: verifying AI outputs on-chain.

My research project at EPFL DCL, Wastral, solves it. 🧵

[alper.sol @alperr_sol]

The destination: trustless AI inference for on-chain agents.

ZK is brutally expensive here; fraud proofs barely cost more than running the model.

Not there yet, but the dispute engine works. The path:
memory ops -> a small neural net verified on-chain.

[alper.sol @alperr_sol]

@monad

[alper.sol @alperr_sol]

Can anyone explain what this is?
I have no clue
@haha_app @monad

[alper.sol @alperr_sol]

@AfelNFT 🤫🤫🤫🤫🤫