Monday, September 14, 2026

The Swiss PostgreSQL Users Group and CERN have announced CERN PGDay 2027, scheduled for Friday, February 12, 2027. The single-track event will feature 6-7 sessions followed by a networking social, focusing on practical experiences and the future of PostgreSQL. It serves as the annual gathering for users and developers in western Switzerland and neighboring French regions.

• Event date is Feb 12, 2027, at the CERN Campus in Geneva.
• Co-organized by CERN and Swiss PostgreSQL Users Group (SwissPUG).
• Single-track format with 6-7 sessions plus a networking event.
• Target audience includes users, developers, and enthusiasts locally.
• Call for Papers is now open via the SwissPUG homepage.

Amazon RDS for Oracle now supports the Supplemental Patch Bundle (SPB) for the July 2026 Release Update, covering engine versions 19c and 26ai. Oracle renamed the Spatial Patch Bundle to SPB starting in April 2026 to include patches for Spatial, Data Pump, and GoldenGate. Practitioners can apply this update to new instances or upgrade existing ones to specific engine versions via the RDS console.

• SPB replaces the former Spatial Patch Bundle, expanding scope to Data Pump and GoldenGate.
• Supports Oracle Database 19c and 26ai engines on Amazon RDS.
• Apply via new instance creation or by upgrading existing instances to SPB engine versions.
• Target version strings are 19.0.0.0.ru-2026-07.spb-1.r1 and 26.0.0.0.ru-2026-07.spb-1.r1.
• Check the 'Supplemental Patch Bundle Engine Versions' box during upgrade workflows.

AWS introduces a zero-ETL integration that replicates Amazon DynamoDB tables into Apache Iceberg format on Amazon S3 Tables. An AWS Lambda function executes DuckDB to serve ad hoc SQL queries against this data, authorized via IAM. This setup allows practitioners to perform analytics on operational DynamoDB data without building custom ETL pipelines.

• Eliminates custom ETL code by using native DynamoDB-to-Iceberg replication.
• Leverages DuckDB for fast, in-process SQL analytics on S3-stored data.
• Lambda function exposes queries via an IAM-secured function URL.
• Data lands in Apache Iceberg format for broader ecosystem compatibility.

Dalibo has released PostgreSQL Migrator 1.0, a free open-source tool for moving databases from Oracle and MySQL/MariaDB to PostgreSQL. The tool offers an interactive web UI, offline catalog inspection, and complexity scoring to assess migration difficulty. It handles schema and data conversion, including datatype mapping and SQL expression transpilation.

• Zero proprietary dependencies: pure Go binary ensures easy installation.
• Offline workflow: extract source catalog once, then iterate conversions without connectivity.
• Complexity scoring helps prioritize and compare migration project risks.
• Interactive UI allows browsing original and converted catalogs to spot blockers.

The pg_vault_tde extension has reached version 1.7.1, extending transparent data encryption support to PostgreSQL 17 and 18. It utilizes an encrypted_heap table access method with AES-256-GCM, offloading key management to HashiCorp Vault, OpenBao, HSMs, or local wallets. This release corrects an Authenticated Encryption with Associated Data (AAD) derivation issue affecting out-of-line TOAST values.

• PostgreSQL 18 is now supported alongside version 17 for TDE capabilities.
• TOAST data encrypted by v1.7.0 or earlier will fail authentication; tables must be re-exported.
• Key rotation remains online and per-table, with keys stored externally from the DB.
• Requires OpenSSL 3.x and proper shared_preload_libraries configuration.

New research introduces Chopthin-Consensus Power Sampling (CCPS), an inference-time decoding method that enhances LLM reasoning without post-training. Unlike standard Sequential Monte Carlo approaches that use equal-weight resampling and aggressively prune low-probability paths, CCPS uses the Chopthin resampler to cap the weight ratio between particles. This technique carries unequal weights forward, maintaining genealogical diversity and preventing the loss of potentially correct reasoning trajectories.

• CCPS improves LLM reasoning at inference time without requiring model retraining or fine-tuning.
• It replaces aggressive equal-weight resampling with a bounded weight ratio strategy to preserve diversity.
• Unequal weights are carried forward, reducing the risk of discarding valid reasoning paths too early.
• The method mitigates the pruning problem inherent in standard Sequential Monte Carlo decoding.

This paper introduces EAR, an approach that replaces fixed-size text chunks with entity-aware partitioning for Retrieval-Augmented Generation. By extracting normalized surface anchors from questions, answers, and the corpus, EAR retrieves local windows around matching terms and optionally attaches a parent passage via extractive summary. Evaluated on a cleaned MMLU-style subset, the method aims to resolve the ambiguity often found in standard chunking strategies.

• Replaces fixed-size chunks to improve retrieval precision for multiple-choice QA.
• Uses normalized surface anchors from queries and corpus to find relevant content.
• Attaches larger parent passages via extractive summary for broader context.
• Validates approach on a curated 153-question MMLU-style test set.

R2VC addresses the diagnostic opacity of end-to-end LLM fact-checking by decoupling retrieval, reasoning, and confidence estimation. The architecture combines hybrid sparse and dense Wikipedia retrieval with a supervised, DPO-aligned generator to produce structured verdict candidates. An external NLI cross-encoder selects the best candidate, while a lightweight calibrator estimates confidence to enable selective abstention when uncertainty is high.

• Modular design isolates retrieval, reasoning, and uncertainty for easier debugging and improvement.
• Hybrid sparse+dense retrieval over Wikipedia provides robust evidence grounding.
• DPO-aligned generator produces diverse, structured verdict candidates for better selection.
• External NLI cross-encoder ensures evidence-based candidate filtering.
• Lightweight calibrator enables selective abstention to prevent hallucinated confidence.

This study compares scaling trends between distilled byte and token models by introducing two methods to convert token logits to byte logits: an approximate approach called Marginalize-It and an exact method named End-Of-Token. The authors conduct a large-scale analysis of decoder-only dense transformers, varying tokenization schemes and training objectives across models ranging from 1 billion parameters to 1 trillion bytes of data. The work aims to determine if distilled byte models exhibit similar scaling behaviors to their tokenized counterparts as compute and dataset size increase.

• Introduces Marginalize-It for approximate token-to-byte logit conversion.
• Presents End-Of-Token as an exact alternative for logit conversion.
• Compares scaling trends of distilled byte vs. token models at scale.
• Analyzes decoder-only transformers up to 1 trillion bytes of data.
• Evaluates impact of distillation vs cross-entropy on byte-level training.

Researchers introduce SynthSentry, a method to identify synthetic data in training corpora without needing access to the generating model or its history. The tool calculates a contamination score by measuring distributional divergence across lexical diversity, n-gram tail truncation, and perplexity variance. This approach aims to prevent model collapse by screening data provenance before the training phase begins.

• SynthSentry flags synthetic data without requiring generating model access or labels.
• It measures lexical diversity collapse and n-gram tail truncation as key signals.
• Perplexity variance across reference models helps quantify contamination levels.
• Pre-training screening addresses model collapse before distributional tails deteriorate.

AWS Entity Resolution now assigns individual confidence scores to each matched record rather than applying a uniform group-level score. This change allows systems to distinguish high-quality matches from borderline ones, enabling differentiated thresholding for downstream activation. Previously, a single threshold limited the volume of resolved identities that could be safely processed.

• Use per-record scores to lower activation thresholds for high-confidence matches without increasing false positives
• Retire uniform group-level confidence checks in favor of granular record-level filtering logic
• Improve identity resolution throughput by activating borderline matches that previously failed strict group thresholds

Research shows that editing knowledge graph embeddings to promote a desired answer frequently displaces other correct answers from the result list. Standard locality tests based on parameter reuse fail to detect this ranking degradation. The study introduces a broader audit scope and identifies geometric conditions required to preserve existing scores during updates.

• Promoting a target answer in KGE models often harms other correct answers.
• Parameter-locality tests are insufficient for detecting ranking side-effects.
• Only ~23% of edits on FB15k-237 preserve other correct answer ranks.
• Geometric conditions exist to strictly preserve scores during updates.

Occamy-1.0 is a cost-efficient model built on the Qwen3.6-35B-A3B checkpoint, designed specifically for co-work agents that execute long-horizon workflows. It leverages execution-grounded data and replayable trajectories to optimize for state tracking, coordination, and recovery rather than just peak reasoning. The model uses staged post-training to balance capability with the latency and cost constraints of multi-step agent episodes.

• Optimized for multi-step agent workflows, not just single-turn reasoning tasks.
• Built on Qwen3.6-35B-A3B, focusing on efficiency and state management.
• Uses execution-grounded data to improve coordination and recovery skills.
• Staged post-training targets the Pareto frontier of cost vs. capability.

A new paper introduces GAUGE, an offline protocol evaluating the validity of persona-driven LLM user-simulators as judges for task-oriented agents. Testing 25 agents across six providers on the tau-2-bench and SimulatorArena benchmarks reveals a critical satisfaction-success gap. The study finds that high user satisfaction ratings provide essentially no information about whether the agent actually completed the task successfully.

• LLM judge satisfaction scores are poor proxies for actual task completion success.
• GAUGE protocol separates ranking validity from construct validity in agent evaluation.
• Blind panel ratings of satisfaction do not correlate with verifiable rewards.
• Offline simulation gates may promote agents that are polite but ineffective.

AMD introduces AMDKernelVault, an open dataset and training framework specifically for optimizing kernels on recent CDNA GPUs. The project addresses the NVIDIA-centric bias in current LLM-based kernel agents by providing HIP and Triton corpora with 62,153 verified HIP samples and nearly 40,000 Triton kernels. It features agent-driven pipelines, HIPKernelGen and TritonKernelGen, which convert PyTorch references into optimized kernels validated under ROCm. The team also fine-tunes Qwen3-8B using supervised learning and execution-aware reinforcement learning to improve agent performance.

• Fills the gap for AMD-specific LLM kernel agents, moving beyond CUDA/NVIDIA bias.
• Provides a large, execution-verified corpus of 62k+ HIP and 39k+ Triton kernels.
• Uses agentic pipelines to transform PyTorch code into ROCm-validated HIP/Triton kernels.
• Fine-tunes Qwen3-8B with execution-aware RL to enhance kernel generation quality.

A new arXiv paper isolates the 'harness effect' by testing identical large language models against different agentic toolkits using a private, contamination-controlled suite of 256 tasks. The study ran paired contrasts, such as Claude Opus 4.8 with both the native SDK and DeepAgents, and GPT-5.5 with OpenAI Codex and DeepAgents, grading 792 runs via an isolated oracle. Results indicate that switching harnesses does not yield a statistically significant average advantage in task resolution compared to vendor-native configurations.

• Vendor-native model-harness pairings are not inherently superior to third-party or open harnesses like DeepAgents.
• Performance gains from agentic systems may stem from the model itself rather than the surrounding tooling infrastructure.
• The study uses a contamination-controlled private suite, mitigating data leakage concerns common in public benchmarks.
• Practitioners should evaluate harness flexibility and integration costs rather than assuming vendor lock-in improves outcomes.

WinSyn addresses the gap between existing benchmarks and real-world enterprise complexity by generating synthetic datasets based on realistic workplace scenarios. The pipeline creates synthetic emails to support Retrieval-Augmented Generation and Deep Research agents, providing both long-form and short-form questions with gold answers. This approach aims to capture the challenges of information spread across evolving and potentially conflicting enterprise artifacts.

• Generates synthetic enterprise emails to simulate complex, realistic QA scenarios.
• Provides both long- and short-form questions with gold answers for evaluation.
• Targets limitations of current benchmarks that lack real-world data complexity.
• Supports testing of RAG and Deep Research agents in enterprise settings.

The AIM framework addresses the limitation of LLMs being confined to single-session contexts by enabling persistent memory management across multi-agent, multi-user environments. It dynamically classifies information into private scopes for individual users and public scopes accessible to all, allowing shared knowledge to improve downstream responses. This approach ensures that user preferences evolve over time while maintaining strict privacy boundaries between personal and community data.

• Shifts LLM memory from ephemeral sessions to persistent, cross-user knowledge bases.
• Dynamically enforces privacy by separating private user data from public shared memory.
• Enables multi-agent systems to leverage collective insights without compromising individual privacy.

Alibaba released Open Code Review, an open-source CLI tool that combines deterministic pipelines with LLM agents to review code. Trained on internal data at massive scale, it provides precise line-level comments and detects issues like NPE, SQL injection, and thread-safety violations. The tool supports OpenAI and Anthropic model endpoints and is designed for fast, efficient integration into developer workflows.

• Hybrid architecture merges deterministic rules with LLM agents for balanced accuracy and speed.
• Built-in multi-language rulesets cover critical defects like XSS, SQL injection, and thread safety.
• CLI-based design allows easy integration into existing Git workflows via diff parsing.
• Compatible with major LLM providers including OpenAI and Anthropic for flexible model selection.
• Battle-tested at Alibaba scale, having identified millions of defects for thousands of developers.

AWS Private CA now offers its Kubernetes EKS add-on and Active Directory connector in AWS GovCloud US-East and US-West regions. The EKS add-on simplifies installation and lifecycle management via the console, CLI, and API. This integration automates certificate requests, distribution to Kubernetes secrets, and renewal for TLS and service mesh security.

• GovCloud customers can now use managed EKS add-ons for Private CA.
• Active Directory connector availability expands to both GovCloud regions.
• Automates cert requests, distribution, and renewal for Kubernetes workloads.
• Supports TLS for ingress controllers and service-to-service mesh security.

The langchain-anthropic package version 1.7.2 addresses a bug where invalid tool use blocks were not being preserved correctly. This fix ensures that malformed tool usage responses from Anthropic models are retained rather than discarded. The update resolves the issue introduced in previous versions to improve robustness in tool-calling workflows.

• langchain-anthropic 1.7.2 is now available via pip.
• Fixes preservation of invalid tool use blocks from Anthropic.
• Prevents silent dropping of malformed tool responses.
• Upgrade recommended for stable tool-calling integration.

AWS Systems Manager now identifies six additional root causes for EC2 and hybrid nodes failing to register as managed. Previously limited to network connectivity, the feature now checks IAM permissions, SSM Agent versions, instance status checks, OS configuration, Default Host Management Configuration, and hybrid activation. This update replaces vague errors with specific, actionable diagnostics to streamline remediation.

• Diagnosis coverage expanded from network issues to include IAM, agent version, and OS config errors
• Hybrid-activated nodes are now included in the expanded diagnostic scope
• Specific error messages replace unidentified results, speeding up remediation workflows
• Managed status is critical for patching, Session Manager, and inventory collection

AWS Lambda now allows functions to explicitly choose between Amazon S3 Files high-performance storage and standard S3 buckets for input data. This configuration enables teams to optimize read throughput and latency based on specific application needs, such as low-latency file system access or high-throughput large reads. S3 Files automatically routes operations to the storage tier best suited for the request, combining performance with S3 durability and cost benefits.

• Lambda functions can now explicitly target S3 Files or S3 buckets for input data.
• Optimize read latency with S3 Files or high throughput with standard S3 buckets.
• S3 Files routes operations automatically to the best storage tier for performance.
• Enables scalable data pipelines and stateful agentic workloads with file system simplicity.
• Default behavior remains unchanged unless explicitly configured by the user.

AWS Graviton-powered Redshift RG instances have launched in the Zurich region, offering up to 2.4x better performance than RA3 instances at a 30% lower cost per vCPU. These instances utilize a custom vectorized query engine to process Apache Iceberg and Parquet data directly on cluster nodes. Existing RA3 clusters can be upgraded to RG using Snapshot & Restore, Elastic Resize, or Classic Resize methods.

• RG instances in Zurich deliver 2.4x speedup and 30% cost reduction vs RA3.
• Unified engine handles Redshift warehouse and data lake workloads via Iceberg/Parquet.
• Available in rg.large, rg.xlarge, rg.4xlarge, and rg.12xlarge sizes.
• Upgrade paths include Snapshot & Restore, Elastic Resize, or Classic Resize.

AWS Lambda Managed Instances now support a 90-minute function timeout for asynchronous and event source mapping invocations, up from the previous 15-minute limit. This sixfold increase allows developers to run data-intensive workloads like media transcoding, financial simulations, and AI inference without architectural workarounds. The change eliminates the need to re-architect applications for jobs requiring continuous execution within the serverless model.

• LMI timeout increased 6x to 90 minutes for async and ESM invocations
• Eliminates need for workarounds in long-running batch or AI inference tasks
• Supports data-intensive jobs like Monte Carlo simulations and media transcoding
• No re-architecture required for existing serverless apps needing longer execution

Amazon EBS Volume Clones now allows copying volumes between AWS accounts while re-encrypting them using a KMS key in the destination account. This feature enables organizations to securely move production data into isolated development accounts for safe experimentation. It also supports compliance requirements that mandate separate encryption keys for production and non-production environments.

• Clone production DB volumes directly into dev accounts without manual export/import steps.
• Enforce key separation by re-encrypting data with target account KMS keys during copy.
• Simplifies secure data sharing across account boundaries for testing and analysis.
• Reduces risk of exposing production data in non-isolated environments.

AWS has introduced a native data source connector for Confluence Data Center within Amazon Bedrock Managed Knowledge Base. This fully managed RAG service now automatically crawls, extracts metadata, and performs incremental syncs from self-hosted Confluence instances. Users can ingest wiki pages and blog posts by simply providing instance credentials, eliminating the need for custom ingestion pipelines.

• Eliminates custom ETL pipelines for self-hosted Confluence Data Center integration
• Automates crawling, metadata extraction, and incremental syncs via native connector
• Enables AI agents to access institutional knowledge from Confluence spaces and blogs
• Requires only instance credentials to start ingesting content into Bedrock Knowledge Base