Stories by matt_d

Measuring AI Ability to Complete Long Software Tasks

6o6 v1.1: Faster 6502-on-6502 virtualization for a C64/Apple II Apple-1 emulator

uops.info Update: Emerald Rapids, Meteor Lake, Arrow Lake, and Zen 5

MXFP8 GEMM: Up to 99% of cuBLAS Performance Using CUDA and PTX

PyTorch Autograd and Mutation

The Future of Python: Evolution or Succession – Brett Slatkin – PyCascades 2026 [video]

SlopCodeBench: Benchmarking How Coding Agents Degrade over Long-Horizon Tasks

AutoRocq: Agentic Theorem Prover for Verification

Wax: Optimizing Data Center Applications with Stale Profile

Dijkstra's Shortest-Path Algorithm: A visual exploration, following Sedgewick

Speculative Decoding: Performance or Illusion?

Goedel-Code-Prover: Hierarchical Proof Search for Open SotA Code Verification

MLSys 2026 Papers

An Efficient Heterogeneous Co-Design for Fine-Tuning on a Single GPU

Specula: A framework for finding deep bugs in system code using TLA+

Equality Saturation for Optimizing High-Level Julia IR

UniTe: A Universal Tensor Abstraction for Capturing Spatial Relationships

Co-Design of B+-Tree Index with Emerging Zone Interfaces for Small KV Pairs

CounterPoint: Using Hardware Counters to Refute and Refine µarch Assumptions

PivotRL: High Accuracy Agentic Post-Training at Low Compute Cost

11 12 13 14 15 16 17 18 19 20