Apple MLThe P-Completeness of Inverted Index Traversal: On the Complexity of Evaluating Boolean Query DAGs
Analyzes the P-Completeness of evaluating nested Boolean query DAGs on inverted indices and introduces ComputePN, a sparsity-aware evaluator with a Positive-Negative dual representation that bounds time to O(|Q| · |U_active|) and avoids exponential blowups and the universal scan.
Apple MLExamining Human-Like Behaviors in LLMs: A Multi-Dimensional Analysis of Model Behaviors, User Factors, and System Prompts
A concise multi-dimensional analysis of human-like behaviors in LLMs, examining model behaviors, user factors, and system prompts, backed by large-scale evaluation across multiple models to illuminate controllability and guide responsible LLM design and evaluation.
How we knew COVID was over (and what our models had to unlearn)
Decision framework for when to refit, respecify, or hold production forecasting models in the face of drift and shocks, illustrated by COVID-era disruptions at Airbnb.
How we knew COVID was over (and what our models had to unlearn)
A practical framework for deciding whether to refit, respecify, or hold a production forecasting model, using COVID-era lessons from Airbnb to balance data drift, structural change, and risk.
AWS MLKnowledgeForge: mining gold from the ITSM ticket graveyard
KnowledgeForge demonstrates an AI-driven pipeline that mines resolved ITSM tickets to automatically generate and curate high-quality knowledge-base articles, with deduplication, quality scoring, and a closed loop that feeds back into generation, all orchestrated on AWS Bedrock, S3 Vectors, and Step Functions.
AWS MLHow Fanatics Betting and Gaming built a multi-agent customer support system
Fanatics Betting and Gaming’s scalable multi-agent AI system on AWS orchestrates specialized agents via EKS and Bedrock, with Guardrails, MCP tool integration, and a custom RAG pipeline to deliver fast, compliant, state-specific answers and seamless human escalation during live sports events.
AWS MLAsynchronous patterns for calling Amazon Bedrock AgentCore agents in serverless pipelines
Explores asynchronous invocation patterns for Amazon Bedrock AgentCore agents in serverless pipelines, illustrating how task-token callbacks, direct Step Functions integration, and durable functions reduce idle Lambda costs while coordinating long-running AI reasoning.
AWS MLAutomate Document Processing with Quick Automate and the IDP Accelerator
A scalable, serverless blueprint that automates mortgage document intake end-to-end—classifying, extracting, and validating data with the AWS GAIC IDP Accelerator and Quick Automate to orchestrate downstream workflows, delivering faster time-to-close and lower costs.
AWS MLDomain and publish date filters for Web Search on AgentCore
Explains runtime domain and published-date filtering for Web Search on Amazon Bedrock AgentCore, detailing the admin-plus-runtime policy model, per-call API usage, and regional availability.
DatabricksDesigning effective Genie Agents from a single prompt
Turn trusted business context into production-ready, domain-specific AI agents from a single prompt by grounding them in Unity Catalog data and documents, then scope, benchmark, and expand with evolving knowledge and workflows.
OpenAIAsana cleared 5 years of engineering work in 2 weeks with Codex
Asana uses OpenAI Codex and frontier models to replace an outdated Enzyme testing system, delivering in about two weeks what was expected to take five years, at a fraction of the cost through AI agents and automated codebase migrations.
OpenAIHow NVIDIA scales expertise with ChatGPT Work
NVIDIA scales expertise by turning ChatGPT Work into automated, reusable workflows that surface actionable AI signals, cut manual tasks, and accelerate prototypes and global planning across GTM and solutions architecture.