https://sparsenotes.com/posts/2026/10/papers/ Recent content in Weekly CS Paper Digest — 26 September – 2 October 2026 on Sparse Notes https://sparsenotes.com/images/og-default.png https://sparsenotes.com/images/og-default.png Hugo -- gohugo.io Sat, 03 Oct 2026 00:00:00 +0000 https://sparsenotes.com/posts/2026/10/papers/congestion-avoidance-and-control-jacobson-1988/ Sat, 03 Oct 2026 00:00:00 +0000 https://sparsenotes.com/posts/2026/10/papers/congestion-avoidance-and-control-jacobson-1988/ Seminal paper of the week: Van Jacobson's response to the 1986 Internet congestion collapse. Packet conservation, ACK clocking, slow start, a variance-aware retransmit timer, and AIMD, all added to TCP senders without changing the protocol or a single router. https://sparsenotes.com/posts/2026/10/papers/context-language-models/ Sat, 03 Oct 2026 00:00:00 +0000 https://sparsenotes.com/posts/2026/10/papers/context-language-models/ UW and Meta give the model write access to its own context, exposed as a file it edits with Bash. Zero-shot it beats summarisation and tool-based context managers on deep research, terminal coding, and 24-hour agent swarms, and it can be steered by prompt, evolved, or trained with RL. https://sparsenotes.com/posts/2026/10/papers/mixture-of-kittens-moe-megakernel-nvl72/ Sat, 03 Oct 2026 00:00:00 +0000 https://sparsenotes.com/posts/2026/10/papers/mixture-of-kittens-moe-megakernel-nvl72/ Cursor and Stanford rebuild MoE training for a 72-GPU NVLink domain: pull dispatch, push combine, a tunable overlap granularity, and a device-side ring buffer, all inside one deterministic megakernel. Up to 2.37x over the best public baseline, 1.41x end-to-end on 512 GB300s.