matt_d's submissions | Hacker News

1.		Test-Time Scaling Makes Overtraining Compute-Optimal (arxiv.org)
		1 point by matt_d 3 hours ago \| past \| discuss
2.		Analyzing Reverse Address Translation Overheads in Multi-GPU Scale-Up Pods (arxiv.org)
		1 point by matt_d 4 hours ago \| past \| discuss
3.		Tracing a Full MoE Training Step Through the XLA Compiler (patricktoulme.substack.com)
		3 points by matt_d 6 hours ago \| past \| discuss
4.		Breaking Down the Cerebras Wafer Scale Engine (wafer.substack.com)
		2 points by matt_d 3 days ago \| past \| discuss
5.		The need for better compiler frontend benchmarks: Carbon's benchmarking approach (llvm.org)
		2 points by matt_d 3 days ago \| past \| discuss
6.		DAXFS: A Lock-Free Shared Filesystem for CXL Disaggregated Memory (arxiv.org)
		2 points by matt_d 3 days ago \| past \| discuss
7.		AST Edits: The Code Editing Format Nobody Uses (geometricagi.github.io)
		4 points by matt_d 3 days ago \| past \| discuss
8.		KernelEvolve: Meta's Ranking Engineer Agent Optimizes AI Infrastructure (fb.com)
		3 points by matt_d 4 days ago \| past \| discuss
9.		Software Engineering Is Becoming Civil Engineering (christophermeiklejohn.com)
		2 points by matt_d 4 days ago \| past \| 1 comment
10.		Adaptive Block-Scaled Data Types (arxiv.org)
		2 points by matt_d 5 days ago \| past \| discuss
11.		AC4A: Access Control for Agents (arxiv.org)
		2 points by matt_d 5 days ago \| past \| discuss
12.		Rethinking Language Model Scaling Under Transferable Hypersphere Optimization (arxiv.org)
		2 points by matt_d 5 days ago \| past \| discuss
13.		Distributed builds of LLVM with CMake, recc, and NativeLink (reidkleckner.dev)
		1 point by matt_d 6 days ago \| past \| discuss
14.		A Pattern Generation Language for MLIR Compiler Matching and Rewriting (radbox.org)
		1 point by matt_d 6 days ago \| past \| discuss
15.		Gram Newton-Schulz: A Fast, Hardware-Aware Newton-Schulz Algorithm for Muon (dao-lab.ai)
		2 points by matt_d 7 days ago \| past \| discuss
16.		Compiler as a Service: C++ Goes Live – Interactive C++, interop, and beyond [video] (youtube.com)
		2 points by matt_d 7 days ago \| past \| 1 comment
17.		Measuring AI Ability to Complete Long Software Tasks (muratbuffalo.blogspot.com)
		4 points by matt_d 7 days ago \| past \| discuss
18.		6o6 v1.1: Faster 6502-on-6502 virtualization for a C64/Apple II Apple-1 emulator (oldvcr.blogspot.com)
		3 points by matt_d 7 days ago \| past \| discuss
19.		uops.info Update: Emerald Rapids, Meteor Lake, Arrow Lake, and Zen 5 (uops.info)
		3 points by matt_d 7 days ago \| past \| discuss
20.		MXFP8 GEMM: Up to 99% of cuBLAS Performance Using CUDA and PTX (danielvegamyhre.github.io)
		2 points by matt_d 7 days ago \| past \| discuss
21.		PyTorch Autograd and Mutation (ezyang.com)
		4 points by matt_d 7 days ago \| past \| discuss
22.		The Future of Python: Evolution or Succession – Brett Slatkin – PyCascades 2026 [video] (youtube.com)
		3 points by matt_d 9 days ago \| past \| discuss
23.		SlopCodeBench: Benchmarking How Coding Agents Degrade over Long-Horizon Tasks (scbench.ai)
		2 points by matt_d 9 days ago \| past \| discuss
24.		AutoRocq: Agentic Theorem Prover for Verification (github.com/nus-program-verification)
		2 points by matt_d 10 days ago \| past \| discuss
25.		Wax: Optimizing Data Center Applications with Stale Profile (github.com/ice-rlab)
		1 point by matt_d 10 days ago \| past \| discuss
26.		Dijkstra's Shortest-Path Algorithm: A visual exploration, following Sedgewick (joshmpollock.com)
		2 points by matt_d 10 days ago \| past \| 1 comment
27.		Speculative Decoding: Performance or Illusion? (specdecode-bench.github.io)
		3 points by matt_d 10 days ago \| past \| discuss
28.		Goedel-Code-Prover: Hierarchical Proof Search for Open SotA Code Verification (goedelcodeprover.github.io)
		3 points by matt_d 10 days ago \| past \| discuss
29.		MLSys 2026 Papers (mlsys.org)
		1 point by matt_d 11 days ago \| past \| discuss
30.		An Efficient Heterogeneous Co-Design for Fine-Tuning on a Single GPU (arxiv.org)
		3 points by matt_d 11 days ago \| past \| discuss
		More