tech
-
Essay
A compute operator who claims to have run a particular model can be lying, and the logs that would settle it are written by the party under suspicion. VerInf produces zero-knowledge proofs of LLM inference, bounding the information in an output stream that a committed model does not account for. This living document details what the system certifies, what it does not, and what I contributed to it during the MARS V fellowship.
-
Essay
Software engineering does so little epistemic work that it hardly earns the name “engineering.” The field has never learned to distinguish provenance — where an artifact came from — from warrant — why anyone is entitled to rely on it. Authorship, a passing test suite, and a completed review are routinely mistaken for the second when they are only ever the first; libraries are the rare exception, where warrant is actually constructed and amortized across users who never read the source. LLM-generated code inherits neither comforting story, and the discomfort that provokes is not a new problem but the oldest one in the profession, finally felt without the anaesthetic that provenance usually supplies.
-
Essay
A first-person, code-heavy companion to the preprint Near-Critical First-Moment Lower Bounds for Growing-Radius Domination in Random Regular Graphs. Rather than reproducing its theorem-and-proof form, this page traces where the problem came from — a cops-and-robbers hypergraph question that collapsed into a domination bound — why the answer carries an unnecessary coupon-collector logarithm, and how a chain of computational detours (a failed concavity conjecture, a catastrophic cancellation, an independent audit that caught a stale constant) repeatedly redirected the proof before it reached its final shape.
-
Essay
As we approach AGI, the increase in the ability of Artificial Intelligence models to infer a robust specification from a sparse prompt will lead to a devastating trend of homogeneity. We argue that this is the primary concern regarding the interaction of AI and human intelligence, rather than blanket claims that “AI reduces human cognitive ability.”
-
Essay
TCP/IP, RIP, UDP, and DNS implementations in Go, supporting file transmission of up to 1 GB across networks of up to 8 virtual machines. Extended with a fully RFC-compliant SSH implementation (2,000+ additional lines) supporting sustained sessions of arbitrary length.
-
Essay
Full Unix-like kernel in ~7,000 lines of C, written for Brown CS 169 (Operating Systems with Lab): virtual memory, VFS, system calls, threading, device drivers and interrupt handlers, and file systems. Custom linker support for running userspace x86-64 ELF binaries.
-
Essay
AI labs are likely deliberately reluctant to scale because they are aware that any imminient shift to locally run models as the norm would render their compute redundant. We take Anthropic as a principal case study to validate this hypothesis.
-
Essay
We systematically decompose the sources of SIMD speedup for ML-KEM (Kyber) on Intel x86-64 AVX2. By benchmarking four compilation variants, we demonstrate that GCC’s auto-vectorizer provides negligible benefit, and that hand-written AVX2 assembly delivers a – performance increase for core arithmetic operations. This drives an end-to-end KEM speedup of –.