Hey folks! This week: Financial tribalization, streamlining recursive STARKs, and destroying paper trails in government agencies, among other things.
News
What we're seeing.
- Starknet Integrates Lightning. Users of the Braavos wallet can now scan Lightning Network QR codes and pay with $STRK tokens, which are automatically converted to Bitcoin. Starknet plans to settle transactions on both Bitcoin and Ethereum.
- Pluto Labs Introduces Frames. "Frames" embed web-based crypto processes like Uniswap transactions directly into mobile apps, eliminating the need to switch between devices. Building on Web Proofs, Pluto is working to streamline financial operations within unified user experiences, lowering friction for crypto adoption while improving security with zero-knowledge proofs.
- Lagrange Announces DeepProve. The zero-knowledge machine learning library verifies AI inferences up to 158x faster than competitors, according to Lagrange, using normal CPUs without hardware optimization. The blog post frames DeepProve as an AI Safety product, replacing blind trust in AI "black boxes" with cryptographic verification that models execute according to the intentions of their designers.
- USAID Document Destruction Raises Concerns. Officials in USAID—recently targeted by Musk's DOGE operation—appear to be aggressively destroying records, reports Politico. Officials in the organization have circulated instructions to burn and shred classified documents and personnel files, potentially violating federal record-keeping laws. This destruction comes amid an 80% termination of USAID programs.
Research
What we're studying.
- DEEP Commitments and Their Applications (2024). Alan Szepieniec of Neptune Cash introduces a method for committing to polynomials that allows low-degree tests like FRI to be batched and deferred. The innovation addresses a fundamental inefficiency in recursive STARK verification, where the FRI low-degree test typically accounts for roughly half the computational cost and proof size. Merging these tests across multiple proofs has generally been quite demanding on memory. Szepieniec's technique decouples the FRI step from preceding operations, reducing an entire algebraic execution trace to a single polynomial commitment. This results in a 100-1000× reduction in memory requirements compared to previous approaches. The technique also maintains zero-knowledge throughout, so witnesses leak no information about the original traces.
- The Accidental Computer: How Data Availability Schemes Can Double as Polynomial Commitments (2025). To scale blockchains, roughly two main approaches have emerged: Succinct proofs verify statements about data without requiring validators to process the underlying information, whereas data availability schemes allow light nodes to verify data is available without downloading entire blocks. Evans and Angeris find that these two systems share significant computational overlap—specifically in the encoding step where data is prepared for verification. As a result, data availability schemes can be repurposed as polynomial commitment schemes with little to no additional overhead. The technique works with multiple data availability schemes, including ZODA tensor variations, and requires no changes to existing data availability layers.
Perspectives
What we're discussing.
- The Rise of Financialized Tribes (MAGA Edition). We've been predicting the rise of financialized tribes since we started doing Twitter Spaces about one year ago. Covering Kash Patel's supplement company, Vivek Ramaswamy's anti-ESG asset management firm, and Trump's meme coins among other things, Politico reports on the markedly... entrepreneurial aspects of the MAGA movement. Our point is not to condemn or condone any of these things in particular, but just to say: We're seeing increasingly explicit tribal dynamics, which are increasingly financialized, at increasingly high levels of the government, with various tribes fighting for survival in a world with no coherent center.
- Perhaps AI, PoW, and DOGE all have something in common. Namely, the automation of philosophy—the decluttering of "fraud, waste, and abuse" hiding inside counterfeit humanisms.
- Nockchain Technical Updates. In this week's testing, we discovered an issue with the +og core in the hoon standard library. +og is a pseudorandom number generator core that uses the SHA-256 hash. We were using the +rad arm in +og to derive keys for unit tests. The +rad arm produces a pseudorandom value between 0 and N where N is a large value. We discovered it's trivial to find collisions using +rad:og with large N. We are ensuring we uniformly use Argon2, specifically Argon2d, for key derivation across our tests and wallet. Argon2 is the winner of the 2015 Password Hashing Competition. Argon2d is a good fit for non-interactive key generation because it is intentionally slow to compute, making it costly to crack. Argon2d is also memory-hard and uses data-dependent memory, which makes it resistant to GPU/FPGA/ASIC cracking.
That’s all for this week!
Hit reply if you have any comments, feedback, or want to share what you’re working on!
