The Frontier

The main job of a Cashu mint is preventing double spending by tracking spent e-cash tokens, which Calle calls the 'state of the mint'.

Calle explains that using a sparse Merkle tree instead of a database allows a mint to commit to a single hash representing all spent tokens, enabling efficient exclusion proofs.

An exclusion proof proves a token hasn't been spent. Calle theorizes using it with on-chain scripting could enable users to unilaterally withdraw Bitcoin from an offline mint.

Running a Cashu mint inside a Trusted Execution Environment (TE) prevents operators from accessing private keys, eliminating fast rug pull risk and offering plausible deniability.

Calle says CDK mint works in a TE with on-chain Bitcoin backing, but running a Lightning node inside is harder. He mentions Lexi has achieved this.

To solve the offline mint risk, Calle describes a timelock script: mint controls funds unless it's offline for two weeks, then a multisig third party can help users withdraw.

Cashu is switching cryptographic schemes from David Wagner's 1996 SecP256k1-based blind MAC to BLS signatures, which allow public verification and simplify third-party operations.

Calle's team has built the BLS upgrade and will deploy it via keysets in about two weeks, aiming for a seamless user transition.

Calle advocates using separate AI coding agents per project to preserve context, noting GPT-5 series is drier and more task-focused than sycophantic GPT-4.

Calle argues LLMs are token predictors disconnected from reality, while embodied world models trained on action-state data can learn physical laws and plan.

He describes JEPA models as a path to lifelong learning AI that could run on a single GPU and adapt to real-world changes, unlike static LLMs.

Calle believes AI empowers high-agency individuals, enabling non-technical people to build complex projects, which he finds hopeful.