OffNet Newsroom

Daily topic roundup

AI / ML

Thursday, September 17, 2026 · 5 stories, curated & summarized — click any story for the source.

Share this roundup on LinkedIn Full edition
arXiv cs.AI researchai

NVIDIA NeMo Data Designer offers declarative multimodal synthetic data generation

NVIDIA has released NeMo Data Designer (NDD), an open-source framework for generating multimodal synthetic data using a declarative configuration format. The tool supports diverse column types including text, code, images, and embeddings, allowing users to explicitly steer dataset diversity through statistical samplers. A core feature is its built-in preview-and-revision loop, which treats configuration as an inspectable artifact to ensure reproducibility and workflow sharing during the iterative generation process.

  • Declarative config enables reproducible, shareable multimodal synthetic data pipelines.
  • Built-in preview-and-revision loop supports iterative dataset refinement and quality control.
  • Plugin architecture allows custom column types and functionality beyond standard offerings.
  • Supports text, code, images, embeddings, and statistical samplers for controlled diversity.
HOW IT WORKSNeMo Data Designer Pipeline1Define declarative configuration2Generate multimodal data3Preview and inspect4Revise and refine
Read at source →
InfoQ generaldevops

OpenAI Classifies GPT-6 Astra as Critical for Cybersecurity

OpenAI has designated GPT-6 Astra as the first model meeting the Critical threshold in its Preparedness Framework. Expert-led testing revealed the model identified previously unknown vulnerabilities in a browser and OS kernel, successfully constructing working exploits. The accompanying system card highlights a significant reduction in the monitorability of the model's chain-of-thought processes.

  • GPT-6 Astra is the first model OpenAI deems Critical under its cybersecurity framework.
  • The model autonomously discovered and exploited zero-days in a browser and OS kernel.
  • Chain-of-thought monitorability has declined substantially, complicating auditability.
  • Rapidly escalating offensive AI capabilities demand stricter safety controls.
Read at source →
InfoQ generaldevops

Pinterest cuts memory in Manas search using quantized SPANN and SSDs

Pinterest Engineering upgraded its Manas search platform to handle massive datasets more efficiently. The team replaced memory-intensive HNSW indexes with quantized SPANN, leveraging scalar and product quantization to drastically reduce RAM consumption. This shift maintains high recall rates while optimizing performance through SSD utilization and preparing for multi-vector relevance matching.

  • Quantized SPANN replaces HNSW to lower memory footprint without sacrificing recall.
  • SSDs are used to optimize I/O performance for large-scale vector searches.
  • Platform is evolving toward multi-vector models for better relevance matching.
  • Significant memory savings achieved through scalar and product quantization techniques.
Read at source →
arXiv cs.AI researchai

Outcome-Based Calibration Pruning Targets Reasoning Model Efficiency

Large reasoning models suffer from inference overhead due to long chain-of-thought traces. Recent pruning methods calibrate on model rollouts but treat all tokens uniformly, failing to distinguish between correct and erroneous reasoning paths. This new approach, OBC-Prune, evaluates parameter importance based on whether the computation leads to a correct outcome, ensuring pruned weights support valid reasoning rather than statistical salience alone.

  • Standard pruning protects weights by statistical salience, not reasoning correctness.
  • Erroneous computation patterns survive pruning, degrading final model quality.
  • OBC-Prune calibrates importance based on successful reasoning outcomes.
  • This method reduces inference overhead while preserving logical accuracy.
TRADE-OFFStandard vs Outcome-Based PruningStandard PruningRelies on statistical salienceIgnores reasoning correctnessErroneous patterns surviveOBC-PruneCalibrates on outcomesEnsures valid reasoningPreserves logical accuracyvs
Read at source →
AWS What's New awsdatabase ↺ since 09-16

New Granite Speech, Kanana 2, and OpenFold3 models on SageMaker JumpStart

Amazon SageMaker JumpStart now hosts three new foundation models: IBM's granite-speech-4.1-2b, Kakao's kanana-2-30b-a3b-instruct, and OpenFold3. These additions target specific enterprise needs, including multilingual speech recognition, bilingual agentic AI, and biomolecular structure prediction. The models allow customers to deploy specialized AI workloads directly on AWS infrastructure.

  • IBM's granite-speech-4.1-2b supports ASR and translation for six languages in a compact 2B parameter size.
  • Kakao's kanana-2-30b-a3b-instruct enables bilingual agentic AI capabilities for complex reasoning tasks.
  • OpenFold3 provides high-performance biomolecular structure prediction for scientific and healthcare use cases.
  • All models are pre-configured on SageMaker JumpStart for immediate deployment and scalability.
BY THE NUMBERSThree New Models Added3New foundation models on SageMakerGranite Speech, Kanana 2, OpenFold3 now available
Read at source →
↑