SecPRBench is a new benchmark of 265 tasks drawn from real bug-introducing commits in nine major systems, such as Firefox, V8 and OpenSSL. Across eight frontier models, recall for catching vulnerabilities ranged from 32.5% to 78.1%, and in 68% of misses the agent flagged the flawed file but blamed a different mechanism.

— via arXiv Programming Languages, Yihan Yang, Chenyuan Yang, Zijie Zhao, Anh Nguyen Hoang, Ziqi Zhang, Weidong Wang, Zhiheng Lyu, Lingming Zhang

The rush R package offers a shared-state coordination layer for asynchronously parallelized iterative algorithms, using Redis as a shared key-value store so workers run their own loops. It was benchmarked with asynchronous decentralized Bayesian optimization on LightGBM tuning using 448 workers.

— via arXiv Distributed Computing, Marc Becker, Bernd Bischl

Apple acknowledged that a small number of iPhone Duo units have inner displays cracking along the hinge, after users reported the problem over the weekend on Reddit and other forums. Affected owners can contact AppleCare for a replacement without needing AppleCare+, though some reports appear to be fake AI-generated ones.

— via Daring Fireball, John Gruber

Berkeley's Measure Z on the Nov. 3 ballot would impose a six-year parcel tax raising over $9 million in its first year to capitalize a regional East Bay public bank. Supporters see progressive investment; opponents say Berkeley is footing a regional bill.

— via Berkeleyside, Alex N. Gecan

A header-only C++23 library builds perfect hash tables at compile time, using a gperf-style association table plus SIMD key comparison. Lookups cost about 5 cycles, roughly 1.3 ns, versus 16.6 ns for std::unordered_map and 7.7 ns for gperf on HTTP header names.

— via Daniel Lemire, Daniel Lemire

DigitalOcean opened a public preview of Learning Services and Agent Data Services for building and improving AI agents. Learning Services provides Insights, Signals and Simulations, while Agent Data Services indexes documents, media, logs and sessions under governance controls.

— via InfoQ, Sergio De Simone

otvdm is an altered winevdm port that lets 64-bit Windows run 16-bit Windows programs. It pairs a CPU emulator with Wine-derived Win16-to-Win32 translation, since 64-bit Windows cannot modify the LDT, and offers installation and build instructions.

— via Hacker News (top 30), doener

A Swift Evolution pitch proposes a send expression and an @oneway marker for fire-and-forget calls onto actors from synchronous code, as a lighter and order-preserving alternative to spawning a Task each time. It also extends the idea to distributed actors, where oneway calls can use fire-and-forget transport.

— via Swift Forums, ktoso

Researchers propose a cloud-edge architecture for clinical screening in bandwidth-limited rural areas, with small edge models producing compact outputs that a cloud LLM turns into summaries. On 20 multimodal cases, it reached 98-99% tool recall at 4-15x lower token cost.

— via arXiv Distributed Computing, Hei Ting Chan, Chenwei Wu, Xueshen Liu, Zesen Zhao, Boyuan Zheng, Luis Filipe Nakayama, Michael G. Morley, Liyue Shen, Jiasi Chen, Z. Morley Mao

PastaLean is a system that deterministically converts Python code into Lean, producing a provable version for verification and a runnable version for computation. Its TypeInfer engine adds static typing for untyped Python and tops the TypeEvalPy leaderboard, with over 92% coverage on LeetCode and HumanEval+.

— via arXiv Programming Languages, Anirudh Gupta, Petros Markopoulos, Siddhartha Gadgil

EnvLunch is a warm-up stage for language-model agents that learns from executable environments such as databases without any human annotation. It beats the baseline by over 9 points on average, and the baseline needs 8 to 32 times more labeled examples to match its smallest-budget accuracy.

— via arXiv Databases, Wei Liu, Min Gao, Yali Du, Yulan He

The study introduces a sequential-dispatch measurement method and finds that, for single-stream (batch size 1) LLM inference on WebGPU, per-dispatch overhead rather than kernel quality is the bottleneck. It concludes that reducing dispatch count is the effective optimization.

— via arXiv Distributed Computing, J\k{e}drzej Maczan

The paper presents Scatter-Gather Language (SGL), a simple imperative language for programming heterogeneous, multi-level parallel platforms. It builds on BSP experience and Valiant's Multi-BSP, and aims to cover many BSP algorithms with predictable, portable, scalable performance.

— via arXiv Programming Languages, Chong Li, Ga'etan Hains

A benchmark study of 13 learned indexes across more than 4000 runs finds each design pays a different price for concurrent inserts, including synchronized rebuild bursts and, in three of five concurrent designs, lost keys or stalled progress. Buffering during rebuilds and background rebuilds sharply cut high-latency inserts on 400M real keys.

— via arXiv Databases, Emmilly Immaculate Namuganga, Laurent Bindschaedler

The work presents three algorithms for converting RDF 1.2 data into labeled property graphs. Structural is lossless, Direct recovers annotated edges, and Fold merges descriptions into edge properties for a more compact but lossy result. All scale linearly in the experiments.

— via arXiv Databases, Maxime Jakubowski, Dominik Tomaszuk, Daniel Fern'andez-'Alvarez, Ruben Taelman, Ruben Dedecker, Jose-Emilio Labra-Gayo, Katja Hose

OasisKV is an LLM inference system built on vLLM that moves the KV cache out of HBM into other memory tiers, prefetching only the most relevant entries predicted via speculative decoding. With 2,048 KV tokens in HBM it stays within 0.61 accuracy points of full attention and reaches 1.86x vLLM throughput on AIME24.

— via arXiv Distributed Computing, Can Xiao, Sukmin Cho, Junbong We, Zhixiong Niu, Jianyi Cheng, Yiren Zhao, Youngjin Kwon, Yongqiang Xiong, Rui Ma, Junyi Liu

Nemo is a proposed index-and-buffer design for storage-disaggregated OLTP databases that push the index down to storage. It uses a tree-structured record cache, caches negative search results, and offers phantom protection, outperforming page-based B+Trees, 2-Tree and PCL in experiments.

— via arXiv Databases, Wenjie Hu, Xinjing Zhou, Zhihan Guo, Xiangpeng Hao, Xiangyao Yu

The paper presents Probabilistic Agentic Programming, which turns responsibility boundaries for LLM agents into explicit programming abstractions, backed by a mediation kernel, proofs and an exploratory study. In 1,225 assessed observations, native Touba matched the top observed counts for each model, including ties with existing frameworks.

— via arXiv Programming Languages, Khadim Lo, Soukaina Alaoui Ismaili, Jean-S'{e}bastien Dessureault

Holo-NN is a proposed framework for private approximate nearest neighbor search that replaces data-dependent branching with one linear projection built using compressive sensing. Encrypted queries reduce to a single non-interactive matrix-vector multiplication, avoiding FHE bootstrapping, and the authors compare it with other private ANN methods.

— via arXiv Databases, Bryant Gipson

A research paper argues that single-version concurrency control suits cloud transaction processing better than MVCC, since cloud-native HTAP designs decouple transactional and analytical engines. It proposes a lighter SVCC design and a research roadmap for next-generation DBMSs.

— via arXiv Databases, Lam-Duy Nguyen, Gabriel Haas, Muhammad El-Hindi, Viktor Leis, Philip A. Bernstein

JWST/NIRSpec spectra of 16 faint submillimeter galaxies at redshifts of about 1 to 4 in the Hubble Ultra Deep Field show that most host obscured AGN and have moderately high metallicities. The galaxies follow standard scaling relations, so they appear to be typical massive star-forming galaxies.

— via arXiv — Astro-ph, Tomokazu Kiyota, Masami Ouchi, Daisuke Iono, Seiji Fujimoto, Kotaro Kohno, Yoshihiro Ueda, Kimihiko Nakajima, Moka Nishigaki, Hidenobu Yajima

The paper introduces Pat, described as the first programming language built around mailbox types, a behavioural type system for actor languages. Its typechecker catches protocol violations, unexpected messages, forgotten replies and self-deadlock, and is tested on a factory automation case study and Savina benchmark examples.

— via arXiv Programming Languages, Simon Fowler, Duncan Paul Attard, Danielle Marshall, Simon J. Gay, Phil Trinder

A systematic study using a four-parameter Maxwell-CSS approach maps where twin stars could exist, requiring a strong hadron-to-quark phase transition. Under 2-sigma observational filters, twin stars reach at most 2.05 solar masses, with a discontinuity up to 7.76 times saturation density.

— via arXiv — Astro-ph, Shamim Haque, Atharva Shinde, Asim Kumar Saha, Tuhin Malik, Ritam Mallick

RT-HiSS is presented as the first exact GPU similarity search algorithm for high-dimensional data built on ray tracing cores, with candidate refinement on CUDA cores and a two-pass method for batching under memory limits. It reports speedups up to 8.37x over state-of-the-art GPU algorithms on six real-world datasets.

— via arXiv Databases, Revanth Reddy Munugala, Michael Gowanlock

Researchers use scattering theory to study two-excitation interactions in subwavelength atomic arrays. At critical points of the band structure, scattering becomes universal and photon-field processes are suppressed, isolating a unitary nonlinear channel between collective dark spin waves.

— via arXiv — Physics, Yidan Wang, Oriol Rubies-Bigorda, Valentin Walther, Susanne F. Yelin