Weekly Paper Notes — three entries per week: 1 seminal + 2 top picks.
🔁 Seminal Paper of the Week
- The UNIX Time-Sharing System — Dennis M. Ritchie and Ken Thompson (CACM 1974). A general-purpose operating system in about 16K words, built by removing distinctions rather than adding features: files are untyped byte streams, directories are files, devices are files, and
fork/execstay factored so redirection needs no kernel support at all.
⭐ Top Papers of the Week
- mold: A Massively Parallel Linker — Rui Ueyama, University of Tokyo (arXiv:2608.23228). Decoupling input parsing from symbol resolution turns every linker pass into a parallel-for loop; the result links TensorFlow’s 9.9 GiB debug library in 3.23 s against lld’s 52.16 s, and an ablation shows no single optimization dominates — leaving any one pass sequential costs 40–542%.
- LeVJEPA: Efficient & Scalable Video Pretraining without the Heuristics — Kuhn, Maes, Serra, Le Lidec, LeCun, Balestriero, Buettner (arXiv:2608.27395). Swapping architectural collapse-prevention for SIGReg’s provable guarantee removes the target encoder, stop-gradient, and predictor; dropping 95% of tokens then raises ImageNet accuracy from 33.9% to 47.6%, and block-causal attention comes free.
Previous digests this month
- Managing Update Conflicts in Bayou — seminal pick from the 16 – 22 August digest, alongside notes on agentic serving workloads and hyperparameter transfer for large-scale MoE.
- Communicating Sequential Processes — seminal pick from the 2 – 8 August digest.
The full weekly Slack digest covered ~15 more papers across AI/ML, NLP, distributed computing, databases, operating systems, programming languages, and networking; only the highest-impact picks are expanded into blog notes.
