Monday, July 13, 2026

6 stories worth your scroll — picked from 44 curated today.

1 · Copy → 2 · Share → 3 · Paste & post

View the post text

Critical kernel vulnerabilities, efficient AI inference, and scalable data architectures define today's engineering priorities.

▸ GhostLock exposes a 15-year-old stack UAF flaw in Linux, enabling privilege escalation and requiring immediate kernel patches.
▸ PoWA Archivist 5.2.0 adds PostgreSQL 19 support, enabling real-time performance metrics and lock statistics on the latest release.
▸ GATS replaces stochastic LLM inference with layered world models, reducing costs and improving reliability for agent planning tasks.
▸ CogniConsole externalizes LLM control logic, proving that inference-time framing matters more than model capability for reliability.
▸ StickyMoE introduces routing consistency losses to minimize expensive memory swaps in Mixture-of-Experts models during inference.
▸ Netflix adopts stateful key-value abstractions via CloudStream to safely process terabytes of bulk data at scale.

Summaries, takeaways and sources → https://offnetchat.com/news/2026-07-13/brief.html

#Linux #PostgreSQL #LLM #MoE #DataEngineering

Automation / DevOps / IaC NEW Hacker News (100+ points)

GhostLock: 15-year-old stack-UAF vulnerability in Linux kernel

Researchers have identified GhostLock, a use-after-free flaw rooted in the Linux kernel's stack handling that has persisted across distributions for 15 years. The vulnerability allows for potential privilege escalation or system instability by accessing freed stack memory. This finding highlights deep-seated issues in long-standing kernel memory management patterns that remain unpatched in many environments.

GhostLock is a stack-based Use-After-Free (UAF) bug in the Linux kernel.

Database Technology NEW PostgreSQL News

PoWA Archivist 5.2.0 Adds PostgreSQL 19 Support

The PoWA team released version 5.2.0 of powa-archivist, the core extension for the PostgreSQL Workload Analyzer. This update introduces compatibility with PostgreSQL 19, including new system views for recovery and lock statistics. The tool continues to provide real-time metrics, charts, and index optimization suggestions across supported PostgreSQL versions.

PoWA now supports PostgreSQL 19, essential for early adopters of the latest release.

Agentic AI NEW arXiv cs.AI

GATS replaces LLM inference with layered world models for efficient agent planning

The GATS framework addresses the high cost and stochastic behavior of LLM-based planning tools like LATS and ReAct by eliminating LLM calls during inference. It combines systematic UCB1-based tree search with a three-layer world model that handles exact symbolic matching, execution log statistics, and LLM predictions only for unknown actions. This approach aims to deliver superior planning performance on synthetic tasks with branching paths and dead-ends while reducing computational overhead.

GATS removes LLM inference during the planning phase, significantly cutting computational costs.

LLMs NEW arXiv cs.AI

CogniConsole externalizes LLM inference control to reduce failures

This paper argues that LLM reliability depends less on model capability and more on inference-time control mechanisms like task framing and context selection. It introduces CogniConsole, an architecture that externalizes this control into a structured interface combining programmatic coordination with bounded prompt reasoning. Probes in a multi-step interactive environment show that adding structural scaffolding systematically reduces output variance and failure rates.

Reliability is driven by inference-time control layers, not just model size or capability.

AI / ML NEW arXiv cs.CL

StickyMoE trains MoE routers to reduce memory swaps during inference

Mixture-of-Experts models often trigger frequent expert switching between consecutive tokens, causing costly weight swaps between storage and fast memory. This work introduces StickyMoE, a differentiable routing consistency loss that penalizes abrupt switches, encouraging the router to maintain assignments across semantically coherent spans. Unlike post-hoc fine-tuning, this method integrates directly into pretraining without architectural changes.

StickyMoE adds a single hyperparameter lambda to penalize abrupt expert switches during training.

AWS NEW InfoQ

Netflix shifts key-value abstractions from stateless to stateful for bulk data

Netflix engineers detail their pivot to CloudStream, a reusable framework for capturing, converting, and deploying data at scale. The architecture moves key-value abstractions from stateless to stateful models to safely handle terabytes of bulk data. This approach leverages specific data access patterns and Pathfinder prototypes to streamline operations. The result is a deployment process that is 99% faster than previous methods.

Adopt stateful key-value abstractions to safely migrate terabytes of bulk data