Tuesday, July 21, 2026

The AWS Database Blog outlines methods for moving IBM Db2 instances from on-premises AIX or Windows environments to Amazon RDS for Db2. Primary migration steps utilize native IBM utilities like db2look and db2move, with AWS DMS offered as an alternative. The guide emphasizes verification procedures for data integrity and strategies to reduce cutover duration.

• Use native db2look and db2move tools for schema and data extraction from legacy AIX/Windows hosts.
• Consider AWS DMS as an alternative migration path depending on downtime tolerance.
• Implement strict data integrity checks before final cutover to ensure consistency.
• Plan cutover procedures carefully to minimize service interruption during migration.

SelKV tackles the linear memory growth of LLM KV caches by selectively merging or dropping tokens based on value-vector similarity. The method uses a soft cosine gate to modulate these decisions, preventing the representation degradation and attention sag common in indiscriminate aggregation approaches. This training-free framework aims to maintain model fidelity while significantly reducing the memory footprint during autoregressive generation.

• Eliminates need for retraining or fine-tuning via a training-free approach.
• Reduces KV cache memory usage by merging similar tokens and dropping dissimilar ones.
• Mitigates attention sag by compensating for softmax mass distribution mismatches.
• Uses soft cosine gating to make adaptive, per-token merge-or-drop decisions.

A new arXiv paper identifies a structured confound in RLHF where pairwise preference labels reflect the annotator's mental state rather than just output quality. Under stress, raters' preferences shift in ways that create systematic bias in reward models. This state-dependent error propagates through policy optimization, distinct from ordinary disagreement or random noise. The authors propose an audit framework to detect and mitigate this specific source of bias.

• RLHF preference data may encode annotator stress levels, not just response quality.
• State-dependent bias is systematic and propagates through reward modeling.
• Distinguish rater state shifts from random label noise or normal disagreement.
• New audit framework proposed to test for and mitigate this structured confound.

The paper identifies that self-generated rollouts in RL with verifiable rewards often suffer from semantic redundancy, causing models to converge on erroneous reasoning basins with low reward contrast. To fix this, the authors propose W2SPO, an off-policy method that leverages a weaker, computationally efficient auxiliary model to inform the stronger policy's exploration. This weak-to-strong paradigm helps break out of local optima by introducing diverse trajectories that the primary model might miss.

• Solves semantic redundancy in self-generated rollouts for reasoning tasks.
• Uses a weaker auxiliary model to guide stronger model exploration.
• Off-policy RL approach avoids converging on erroneous reasoning basins.
• Improves reward contrast for more effective policy updates.

LaCache accelerates Diffusion-based Large Language Models by eliminating operator-level redundancy in Semi-Autoregressive decoding. The framework employs Lossless State Memoization to cache embedding outputs, RoPE pre-attention states, and FlashAttention softmax statistics. By reusing these invariant components during denoising, it achieves training-free acceleration without sacrificing generation quality.

• Targets Diffusion LLMs using Semi-Autoregressive decoding for parallel text generation.
• Caches EmbedCache, RoPECache, and FACache to avoid recomputing invariant sequence blocks.
• Training-free approach requiring no model retraining or architectural changes.
• Reduces redundancy by recognizing prefix and masked suffix invariance within denoising steps.

SpecLA introduces a speculative decoding runtime tailored for stateful linear-attention models, which use recurrent states instead of growing KV caches. Existing speculative systems fail here because verification must respect recurrent dependencies across chains and branches. The approach ensures acceptance updates only the accepted state trajectory and prevents the drafter from wasting work on invalid candidates.

• Linear-attention models use recurrent states, not KV caches, requiring new verification logic.
• SpecLA handles topology-aware verification for chains and trees in stateful targets.
• Acceptance logic updates only the accepted state trajectory to maintain correctness.
• Drafter constraints prevent submitting candidates that waste stateful verification work.

OpenLanguageModel (OLM) is an open-source PyTorch library designed for transparent pretraining of small language models. It structures model code to mirror architecture diagrams using explicit wiring modules like Block and Parallel, enabling seamless transitions from educational notebooks to full-scale research ablations. The library integrates tokenizers, streaming datasets, optimization, mixed precision, and hardware-aware execution across CPU and single-node GPU setups.

• Code readability mirrors architecture diagrams, easing debugging and teaching.
• Reusable components allow moving from notebooks to production pretraining unchanged.
• Built-in support for FineWeb-Edu and other streaming datasets simplifies data pipelines.
• Hardware-aware execution covers CPU, single-GPU, and single-node multi-GPU modes.

ColGraphRAG addresses retrieval accuracy issues in graph-grounded multimodal QA by swapping single-vector visual ranking for late-interaction MaxSim scoring. This ColBERT/ColPali-inspired approach preserves patch and token-level structure often lost in bi-encoder similarity, ensuring fine-grained alignment of graph-linked images. The method keeps offline graph construction and text retrieval unchanged while improving retrieval-stage performance on MultimodalQA.

• Late-interaction scoring retains token-level detail for better image-text alignment in GraphRAG
• Visual ranking is upgraded to MaxSim-style multi-vector scoring without changing graph construction
• Retrieval accuracy improves on MultimodalQA compared to standard bi-encoder approaches

Shapley Context Pruning introduces a framework that models RAG context retrieval as a cooperative game to determine token importance. It replaces heuristic loss functions with Shapley values for interpretable attribution. The method utilizes a Deep Sets architecture to balance fine-grained and coarse-grained context representations effectively.

• Replaces heuristic losses with cooperative game theory for context importance
• Uses Shapley values to provide interpretable attribution for retrieved tokens
• Employs Deep Sets architecture to balance representation granularity
• Offers a unified framework for context reranking and pruning

Researchers propose Generative Ontology Induction (GOI), a domain-agnostic framework that extracts structured ontologies from document corpora using large language models. The system produces a generative blueprint of entities, relationships, and constraints, exporting them as typed graphs in YAML or JSON. To evaluate structural completeness, the authors introduce the Node Coverage Score, which measures how many ontology classes and properties appear in the generated outputs.

• GOI automates schema discovery without requiring predefined schemas or narrow domain constraints.
• Outputs are structured as typed graphs (6 node types, 7 edge types) in YAML/JSON for easy integration.
• New Node Coverage Score metric evaluates the fraction of structural ontology nodes in generated outputs.
• Addresses the bottleneck of manual ontology engineering for knowledge-intensive AI systems.

PPO-HSC is a reinforcement learning framework that addresses mode collapse in LLM fine-tuning by optimizing wide-area policy coverage. It introduces a High-order Sampling Coverage reward that encourages the discovery of unique, high-validity reasoning patterns rather than over-optimizing known solutions. The method maintains a dynamic library of verified unique trajectories to sustain exploration during training.

• Prevents mode collapse by rewarding low-similarity, high-validity reasoning paths
• Uses dynamic trajectory libraries to track and incentivize unique solutions
• Shifts RLVR focus from pure reward maximization to broader exploration
• Aims to preserve model curiosity and solution manifold diversity
• Targets LLM fine-tuning stability in complex reasoning tasks

Researchers introduce masked diffusion language models as a solution to the left-to-right bias inherent in autoregressive world models. This approach enables better conditioning on globally interdependent state anchors like tool schemas and expected outcomes. The result is a steerable text-based world model that supports diverse, on-demand training environments for reinforcement learning agents.

• Addresses mode collapse in RL caused by sparse rewards and fixed task difficulties.
• Overcomes autoregressive limitations by conditioning on global state anchors.
• Enables on-demand diversity scaling for specialized agentic training environments.
• Formalizes text-based world modeling as a steerable transition dynamic.

Researchers identify the planning phase in multi-agent LLM systems as a critical vulnerability where a single prompt injection can cascade through downstream executors. The PlanFlip framework demonstrates four attack vectors—GoalSubstitution, PriorityInversion, ContextPollution, and RoleConfusion—that mimic legitimate tool outputs to bypass filters. Testing across nine frontier models reveals that higher capability often correlates with increased susceptibility to these specific planning-phase attacks.

• Multi-agent architectures are vulnerable at the Planner stage, not just execution.
• Attacks disguise malicious prompts as plausible tool outputs to evade keyword filters.
• Higher model capability does not guarantee immunity; it may increase risk.
• Four distinct attack types (PF-1 to PF-4) systematically corrupt sub-task sequences.
• Defenders must inspect Planner context integrity, not just final outputs.

AI agent systems combining LLMs with external tools suffer from inherent non-determinism due to sampling variance and external state. Existing observability tools capture logs but cannot reproduce these runs in isolation. The authors introduce agrepl, a CLI framework that uses a MITM proxy to intercept and serialize external interactions at the transport layer. This enables strict replay in an isolated environment with zero outbound network access.

• Solves the non-determinism problem inherent in LLM-agent interactions with external APIs.
• Uses a MITM proxy to serialize external interactions into structured execution traces.
• Enables faithful reproduction of agent runs in an isolated, zero-network environment.
• Provides a developer-first CLI for debugging and testing agent behavior deterministically.

MOSAIC addresses two critical failures in current long-term memory systems: the loss of relational context in flat storage and the high latency of LLM-based classification. By implementing an entity-typed graph structure, it preserves the temporal and multi-hop relationships necessary for accurate reasoning. The framework also introduces conflict-aware mechanisms to validate new data against existing knowledge, preventing the silent accumulation of contradictions.

• Replaces flat storage with entity-typed graphs to retain relational context for complex queries.
• Eliminates expensive LLM-based classification steps to reduce latency for real-time agents.
• Detects and resolves contradictions between new inputs and stored knowledge to maintain accuracy.
• Enables reliable multi-hop and temporal reasoning without hallucinated or lost context.

AWS has introduced a preview of KNFSD File Cache, an open-source Apache-2.0 tool designed to cache NFS exports from diverse sources like on-premises filers, Amazon FSx, or other clouds. By storing frequently accessed data in memory and local NVMe, it serves large compute fleets at local VPC speeds, avoiding repeated high-latency trips to the source. The solution supports NFS v3 and v4.1/v4.2 protocols and can front multiple source servers simultaneously.

• Reduces latency for large compute fleets by serving cached files at local VPC speed
• Aggregates exports from on-prem, AWS FSx, and multicloud sources over Interconnect
• Leverages local NVMe and memory caching to minimize cross-link data transfers
• Open-source Apache-2.0 license allows flexible deployment within AWS environments

AWS Control Tower Account Factory for Terraform (AFT) now supports automatic re-application of account customizations when accounts are moved between Organizational Units. This update eliminates the previous manual overhead and reduces configuration drift risks associated with OU transitions. The feature is enabled by setting aft_customization_triggers to account_move, skipping bootstrap and provisioning phases for faster execution of global and account-level customizations.

• Set aft_customization_triggers = ["account_move"] to enable auto-reapplication.
• Reduces operational overhead by removing manual triggers for OU moves.
• Skips bootstrap/provisioning, running only global and account-level customizations.
• Prevents configuration drift by ensuring accounts stay consistent with OU policies.
• Available immediately in AFT deployments with the updated configuration.

Amazon CloudWatch introduces coding agent insights to help engineering leaders measure the value of AI coding tools. The feature integrates with the Claude Apps Gateway for AWS to automatically collect telemetry from Claude Code, Codex, and GitHub Copilot without extra instrumentation. It leverages OpenTelemetry metrics to present agent performance alongside existing operational data in CloudWatch.

• Automated telemetry collection from Claude Code, Codex, and GitHub Copilot via CloudWatch.
• No additional instrumentation required; leverages existing OpenTelemetry metrics.
• Enables leadership to assess ROI, identify high-impact teams, and right-size token budgets.
• Provides visibility into delivery acceleration and access expansion needs.

AWS has updated the EC2 describe-ami response to include any AWS Systems Manager Parameter Store parameters linked to a public AMI. This eliminates the previous need to manually search namespaces to find configuration aliases. The feature allows engineers to easily discover and reference parameters that resolve to the latest AMI version.

• Public AMI metadata now exposes associated SSM parameters directly.
• No manual namespace search required to find configuration aliases.
• Simplifies infrastructure updates by referencing version-resolving parameters.
• Available to all customers at no extra cost in all regions.

AWS Backup has extended its logically air-gapped vault capability to six additional regions, including Taipei, Malaysia, New Zealand, Thailand, Mexico Central, and Canada West. This feature allows for the storage of immutable, isolated backups that are locked by default and encrypted with either AWS or customer-managed keys. Operators can now back up directly to these vaults, copy data across accounts and regions, and share access via Resource Access Manager.

• Logically air-gapped vaults now support six new regions for enhanced geographic redundancy.
• Backups are immutable and locked by default to prevent unauthorized modification or deletion.
• Encryption supports both AWS-owned keys and customer-managed keys for compliance control.
• Vault access can be protected during account compromise using Multi-party approval workflows.

Max Körbächer argues that internal development platforms fail when teams focus solely on infrastructure rather than treating the platform as a product. He highlights the critical need for a product mindset to drive adoption and align with user needs. The presentation outlines how to measure success using DevEx and SPACE metrics while managing technical debt and community engagement.

• Avoid infrastructure-first thinking; treat internal platforms as products with users.
• Adopt a clear product mindset to ensure platform adoption and value delivery.
• Measure platform success using DevEx and SPACE metrics, not just uptime.
• Actively manage technical debt and foster a community for long-term health.
• Align team incentives with platform goals to drive sustainable engineering practices.

DoorDash deployed Entity Cache, a transparent proxy caching layer within its service mesh, to eliminate redundant service-to-service calls. Built on Envoy and Valkey, the system handles over 1.5 million requests per second while maintaining 99.99999% availability. The architecture relies on event-driven invalidation and robust failure handling to sustain high performance.

• Envoy and Valkey form the core stack for high-throughput microservice caching
• Event-driven invalidation ensures data consistency without polling overhead
• Transparent proxy design reduces application-level caching complexity
• 99.99999% availability achieved through rigorous failure handling mechanisms

AWS Data Exports now includes structured product metadata for Amazon Bedrock within Cost and Usage Reports. This update provides consistent attributes like model provider, inference type, and pricing unit to simplify cost attribution. Teams can now query this data via Athena or load it into warehouses without custom parsing logic.

• Eliminates need for custom parsing of varied Bedrock metadata in CUR 2.0 exports.
• Standardized fields include model provider, name, pricing unit, and inference type.
• Enables precise cost attribution by separating input vs output token charges.
• Supports FinOps by unifying Bedrock spend under a single product family name.

AWS CloudTrail now supports selective logging of network activity events generated by VPC endpoints based on the IAM user identity making the API call. This enhancement allows engineers to configure selectors that capture specific events, such as access denied attempts from untrusted identities, while excluding routine traffic from safe lists. The feature helps reduce logging costs and noise by focusing data perimeter strategies on high-value security scenarios.

• Filter VPC endpoint network logs by IAM user identity to reduce noise.
• Log only access denied events for untrusted users to save costs.
• Exclude routine traffic from trusted identities to improve signal-to-noise ratio.
• Supports data perimeter strategies by focusing on critical security events.

Amazon Managed Service for Apache Flink has updated to support Apache Flink version 2.3. The release introduces adaptive partition selection to manage backpressure more effectively under uneven load conditions. It also enhances change data capture pipelines by improving the handling of out-of-order updates and adding new SQL functions for stream conversion.

• Adaptive partition selection reduces backpressure issues during load spikes.
• CDC pipelines gain better correctness for out-of-order update handling.
• New SQL functions simplify conversion between changelog and standard streams.
• Managed service simplifies setup, operation, and scaling of Flink apps.

AWS has made R8i and R8i-flex instances available in the Europe (Stockholm, Zurich) regions. These instances utilize custom Intel Xeon 6 processors to deliver up to 20% higher performance than R7i instances. Specific workloads see significant gains, including 30% faster PostgreSQL performance and 60% faster NGINX web application speed.

• R8i instances are now deployed in Stockholm and Zurich regions for low-latency access.
• PostgreSQL databases can see up to 30% performance improvement over R7i instances.
• NGINX web applications benefit from up to 60% speed gains on this new hardware.
• Custom Intel Xeon 6 processors provide 2.5x more memory bandwidth than previous generations.
• R8i-flex offers a flexible alternative for varied workload requirements.

AWS has launched EC2 I8ge instances in GovCloud US-East and US-West, bringing storage-optimized compute powered by Graviton4 processors. These instances provide up to 60% better compute performance than Graviton2-based predecessors and leverage third-generation Nitro SSDs for faster local NVMe storage. The configuration supports up to 120TB of local storage with significantly reduced I/O latency and variability compared to Im4gn instances.

• I8ge uses Graviton4 for up to 60% compute improvement over Graviton2 instances.
• Up to 120TB local NVMe storage with third-gen Nitro SSD technology.
• Delivers 55% better storage performance per TB compared to Im4gn instances.
• Reduces storage I/O latency by 60% and variability by 75% versus Im4gn.

Amazon OpenSearch Service now allows users to migrate from legacy OpenSearch Dashboards to the new OpenSearch UI with a single click. This feature supports both managed domains and serverless collections, automatically transferring tenants and saved objects without manual recreation. The new interface is designed as a zero-downtime, serverless tool for unified observability across multiple data sources. This reduces operational complexity by preserving existing configurations during the transition.

• One-click migration preserves tenants and saved objects from legacy Dashboards.
• Supports migration for both OpenSearch domains and serverless collections.
• Eliminates manual recreation of thousands of saved objects.
• New OpenSearch UI offers zero-downtime, serverless unified observability.
• Reduces operational overhead when switching to the new interface.