Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
Murat Demirbas's notes from day 1 of HPTS 2026 at Asilomar, including Ion Stoica's keynote 'Reality is the Final Verifier' on gaps in agentic software engineering, plus talks on storage, metastable failures and Little's Law, CXL memory, LLM training resilience on TPUs and inference.
13 min · 2,948 words
Papers We Love: Flow Typing, A New Lens on Linearity
Jared Forsyth works through a new paper on flow typing that aims to make linear types far less cumbersome, implementing its execution model and explaining how it treats values as pass-by-value with immutable updates, and what that could mean for language design.
6 min · 1,388 words
The Vermilion Experiment: Implementing a Lean Backend for Verus
Proofs and Intuitions describes the Vermilion experiment: implementing a Lean back end for Verus to explore verified software tooling across proof assistants.
20 min · 4,538 words
Cloud of suspicion lifted from NIST curves
John D. Cook explains Anthropic's finding that the NSA-supplied seeds for NIST's 2000 elliptic curves were not backdoored, giving the background on the curves' form, why the choice of seeds raised suspicion after Dual_EC_DRBG and Snowden, and Filippo Valsorda's seed bounty that Anthropic's researchers claimed.
2 min · 469 words
rust-djinn: You Should Specify Your Code Unambiguously (and Prove That You Did)
Simon Henniger introduces rust-djinn, arguing developers should write unambiguous specifications for their code and prove those specs hold—bridging documentation, types and verification so correctness claims are checkable rather than aspirational.
13 min · 3,029 words
Google's Gemini 4 Argon post for robotics: physical reasoning that leads ERQA, with representative prompts for zero-shot control, Real2Sim practice, visual-coordinate harnesses, and agentic verification workflows across the robotics stack.
13 min · 3,033 words
Please stop hashing the public key first (or: ML-DSA's µ annoys me)
Erincandescent explains why signature schemes such as Ed25519 and ML-DSA hash the public key before the message, why that ordering blocks a useful optimisation for verifying one message against many candidate keys, and how the constructions could have been arranged differently.
3 min · 784 words
An opinionated hands-on comparison of Isabelle/HOL, Lean, HOL4 and Agda, formalizing Euclid's proof of the infinitude of primes in each and judging them on user experience: dependent types versus LCF-style kernels, tooling, libraries and automation, with the full proofs included.
36 min · 8,226 words
Jesse Waites describes using AI to search millions of digitized historical records, inspired by Benjamin Breen's dodo find, and how the process surfaced a forgotten meteorite report, three lost rhinos and volcanic eruptions missing from the Smithsonian's Global Volcanism Program list, with sources cited.
16 min · 3,659 words
Philip Zucker sketches refinement e-graphs, which bake a privileged <= refinement relation into an e-graph alongside equality so compiler rewrites from abstract specs to concrete programs can be modeled as one-directional, with a prototype built on microegg, a WASM demo, and notes on related term-rewriting ideas.
26 min · 5,940 words
Jane Street reveals what its August ASIC reverse-engineering puzzle chip actually did, how solvers approached the GDS layout without a netlist, and lessons from the physical-design challenge.
10 min · 2,405 words
One vector is too few. One per token is too many.
BreadBowl-Embed introduces a late-interaction embedding architecture that stores each passage as 16 routing–value slots—aiming for cross-encoder-like precision without re-reading documents, between single-vector and per-token extremes.
15 min · 3,524 words
Equality Saturation: An “Incomplete” Project
Max Willsey revisits equality saturation five years after the egg paper, surveying egglog, Slotted, Colored and Versioned E-Graphs, Cranelift's aegraphs and MLIR integrations, and arguing that giving up completeness has been a strength that enables flexibility, control and a growing research community.
8 min · 1,798 words
How Pew Research Center is – and is not – using AI in our work
Pew Research Center outlines internal AI guidelines: surveys stay human-answered, disclosure rules for production use, and careful experimentation while keeping research people-centered.
2 min · 542 words
EX-ARRR: Sailing the 0-click Seas
**Every serious Apple device compromise of the last decade has boring person at the bottom of it: a parser read a file and trusted it a little too much. Not a phishing link, nor a stolen password, but a daemon you never launched, decoding a file you never opened, one byte past the end of a buffer. This is that story. It starts late one evening with a fuzzer that did not know what an EXR file was, and ends with a heap overflow that fires inside a privileged Apple daemon the instant an iMessage lands evading BlastDoor’s, before the little notification banner even finishes…
19 min · 4,447 words
OpenAI agents tried to bruteforce a UN website's API fields
Rowan H-J documents how OpenAI agents scanned UNCTAD’s public statistics API thousands of times—proxies, obfuscation, and odd tool use—while probing API fields on a UN website.
16 min · 3,610 words
Honest About Uncertainty: I Tried to Rebuild Jev’s RLCD From a Blog Post
Anthony Maio reverse-engineers a plausible RLCD training loop for decision-only models from TypeSafe’s Jev blog post, then trains and evaluates a small Qwen3-0.6B checkpoint—with code and ablations.
19 min · 4,310 words
Tackling Robotics with (V)LM Agents
Nishanth J. Kumar surveys recent demos and ideas around GPT-6 and other vision-language models solving robotics tasks—summarizing approaches and offering thoughts on what works and what still breaks.
10 min · 2,304 words
How GPT-6 Astra ascended NetHack: setup, agent loop, tool use, failure modes, and what beating a famously hard roguelike says about LLM agents in open-ended environments.
11 min · 2,477 words
I built non-autoregressive decision models with RL a year ago
Convai Innovations’ Nandakishor recounts building Laya—a ~33ms multilingual non-autoregressive decision engine with calibrated probabilities—via RLCD a year before frontier labs framed similar System One models as breakthroughs.
8 min · 1,852 words