SourcesMiniCPM-SALA: Hybrid Attention for Efficient Long-Context (8600 HF upvotes)HuggingFace Daily Papers·high signalXBlueskyLinkedInCopy linkHighest-upvoted HuggingFace paper. Hybrid sparse-linear attention for efficient long-context modeling. Addresses quadratic attention cost bottleneck.SourceSource pageHuggingFace Daily Papers↳ Follow the threadStack layer / Update threadOpenBMB Releases MiniCPM5-2B Under Apache 2.0 With a 131K Context and the Top Intelligence Index Score for Any Open Model Under 4BOpenBMB / Hugging FacePolicy dependency / Stack layerHierarchical Ransomware Agents Escalate to Dynamic and Memory Analysis Only on Specialist DisagreementarXiv 2609.04820Stack layer / Threat patternLLM decompiler output that recompiles and passes every shipped test still diverges on other inputs, and can silently erase a disclosed CVEarXiv 2609.05370Stack layer / ContrastNVIDIA and MiniMax released Sol-H3, which generates five seconds of video in 1.653 secondsNVIDIA Research (corroborated by Enze Xie on X)Stack layer / Threat patternCodex users are getting 'Cyber Abuse' warnings from OpenAI for security-reviewing their own code, with appeals rejected then reversedr/OpenAI (79 upvotes, 58 comments)Stack layer / Update threadThe Agentic SDLC Throughput Paradox: Coding Gains Attenuate Sharply Between Writing Code and Shipping ItarXiv 2609.04681Stack layer / ContrastIris trains 35B and 397B search agents by reverse-constructing questions from hyperlink structure, and reports benchmarks both with and without context managementarXiv / HuggingFace Daily PapersStack layer / Threat pattern19,325 CVEs compiled into 1,033 detection rules packaged as 172 agent skills, producing runtime evidence for 644 findingsarXiv 2609.05335