https://sparsenotes.com/series/weekly-papers-2026-07-25/ Recent content in weekly-papers-2026-07-25 on Sparse Notes https://sparsenotes.com/images/og-default.png https://sparsenotes.com/images/og-default.png Hugo -- gohugo.io Sat, 25 Jul 2026 00:00:00 +0000 https://sparsenotes.com/posts/2026/07/papers/openforgerl-harness-native-agents/ Sat, 25 Jul 2026 00:00:00 +0000 https://sparsenotes.com/posts/2026/07/papers/openforgerl-harness-native-agents/ An open-source RL framework that trains agents inside real inference harnesses (Claude Code, Codex, OpenClaw) via a lightweight proxy plus Kubernetes-orchestrated per-rollout containers — decoupling training from inference and matching much larger models on GUI/tool benchmarks. https://sparsenotes.com/posts/2026/07/papers/byzantine-generals-problem/ Sat, 25 Jul 2026 00:00:00 +0000 https://sparsenotes.com/posts/2026/07/papers/byzantine-generals-problem/ The 1982 paper that named the problem, proved the n ≥ 3f + 1 bound for oral messages, and showed how digital signatures lift the bound — the foundation every consensus protocol since has stood on. https://sparsenotes.com/posts/2026/07/papers/ Sat, 25 Jul 2026 00:00:00 +0000 https://sparsenotes.com/posts/2026/07/papers/ One seminal paper plus two top picks from the week: harness-native agent RL, and a lossless fix for speculative decoding at million-token context. https://sparsenotes.com/posts/2026/07/papers/windowed-mtp-draft-kv-tax/ Sat, 25 Jul 2026 00:00:00 +0000 https://sparsenotes.com/posts/2026/07/papers/windowed-mtp-draft-kv-tax/ A training-free, lossless fix for speculative decoding at million-token context: apply StreamingLLM-style sliding attention to the MTP draft head only, cutting per-step draft cost by 28–44% while the full-attention target still verifies every token.