All posts, by topic.
Doing research
Research as a craft, a career, and a community.
- One Source, Many PapersPublish the source, let each reader’s AI compile the paper, and keep the claims fixed
- Fifty ProceedingsWhat 1,528 OSDI and SOSP papers say about how systems research changed
- Research Under ContentionPeer review is the validation phase of an optimistic protocol—and that protocol is thrashing
- A Lab of OneThe single-person research group, and the one thing it cannot produce
- Skin in the GameWhen everything can be generated, the last valuable thing is a claim you pay for
-
The Three Pillars of ResearchTaste, execution, communication—and why a PhD is still an apprenticeship
- Good Ideas Are Rare; Taste Is the SearchThe first pillar—how to tell good work from bad, and how to make more of the good
- Execution: The Idea Is the Easy PartThe second pillar—turning a judged idea into a real thing that works
- Communication: Standing Where the Other StandsThe third pillar—moving the work out of your head and into other heads
Correctness and verification for AI
What it means for AI output to be right, and how we could check it.
-
Same Matmul, Different BitsTwo accelerators, one operation, and the function nobody wrote down
- Fix the Tree, Free the ScheduleA spec language for arithmetic, where rounding is explicit and the hardware cannot reorder a sum
- Two Chips, Written DownThe H100 tensor core and the Trainium 2 matrix unit as specs, and the one that leans toward zero
- Check, Then CompileWhat a bit-exact spec makes possible: checking kernels against it, and replaying one chip’s arithmetic on another
- Accountable by ConstructionAn evidence tree for AI: commit to everything, check what you choose, and read only what no machine can vouch for
- Loaded DiceWhen every outcome is legitimate, the attack is in the choosing—and AI now chooses everywhere
-
Token-Based Formal MethodsFormal methods check arguments in boolean logic. Engineers usually reason in words.
- The Old BargainFormal methods deliver certainty, but in a language few engineers write or read
- Token-Based LogicKeep Hoare’s structure, change the assertion language, and check small steps
- Does It Work?One algorithm, a planted bug, and the limits of using a reader as an oracle
- Replay Comes BackA debugging framework for AI agents, and why it has to reach all the way down to the kernels
- Correctness Without a ReferenceExistential equivalence: a definition of correctness for AI
Consistency checking
Checking that what a database did could have come from a legitimate run.
- A Checker Built Like a CompilerBoomslang, and what changes when uncertainty gets a name
- Consistency Checking Is the New VerifierNotes from the first Symposium on Consistency Checking Principles
Systems in the AI era
What AI changes about how systems are built.
- C Is BackHigh-level languages traded machine time for human time; coding agents change the price
-
What Would Make a System AI-NativeA decade of machine learning for systems, not yet widely deployed, and the placement mistake that explains it
- Learning at the Wrong TimeMachine learning for systems did reach production—but only after being compiled, distilled, or rewritten into ordinary code. That pattern is the whole story
- Every Decision Has a Binding TimeFour levels of systems decision-making, and why only compiled code can live at the bottom one
- AI-Native: Autonomy as Revision AuthorityGrade systems by what the AI may revise, not by where it sits—and a definition of AI-native I’m willing to defend
Old systems ideas, seen again
Classic systems concepts used as a lens on something larger.
-
Nothing New Under the Sun
- The Second Time Is CheaperFirst in a series — Nothing New Under the Sun — on the handful of ideas computer science keeps rediscovering under new names
- Everything Is a Search ProblemSecond in the series — Nothing New Under the Sun — on the handful of ideas computer science keeps rediscovering under new names
- Keep It CloseThird in the series — Nothing New Under the Sun — on locality, the oldest performance trick, and the consistency bill it always sends
- The Dream Is to DisappearOn the ultimate problem of computer systems
- Serial Is GoodThe world is concurrent, the machine is parallel, and every layer in between pays to tell us a sequential story
About this blog
Why this blog exists.
- Generation Got Cheap. Judgment Didn't.Why I’m blogging again
- Hello, and what this is