October 2025

Our latest newsletter features Chronos-2, a foundation model for zero-shot time series forecasting, insights into how agentic AI systems work, a novel benchmark for visual defect detection, along with the Amazon Research Awards fall 2025 call for proposals.

Deep dives

Introducing Chronos-2: From univariate to universal forecasting: Amazon's Chronos-2, a 120M-parameter time series model, achieves state-of-the-art zero-shot performance for multiple forecasting types. It excels on benchmarks like fev-bench and GIFT-Eval, particularly when using exogenous features, and outperforms its predecessor Chronos-Bolt in over 90% of comparisons.

Demystifying AI agents: Agentic AI systems can autonomously make online purchases, build software, and research business trends. But how do they actually work? Amazon VP and distinguished engineer Marc Brooker breaks down the core components of agents and explains how AWS’s new Bedrock AgentCore framework implements them.

Novel “Kaputt” dataset sets new benchmark for large-scale visual defect detection: With 238,421 high-resolution images of 48,376 unique items, Kaputt is 40 times as large as current state-of-the-art benchmark datasets and captures the real-world complexities of detecting defects across a vast range of products.

News and updates

Amazon launches $68 million AI PhD Fellowship program: The program will provide two years of funding for more than 100 PhD students at nine universities who are pursuing research in machine learning, computer vision, and natural-language processing. With annual investments of $10 million in student funding and $24 million in AWS credits, the program's goal is to drive innovations that will underwrite the next step in the evolution of practical AI.

Amazon Research Awards issues fall 2025 call for proposals: Successful applicants will receive unrestricted funds, AWS promotional credits, and training resources, including AWS tutorials and hands-on sessions with Amazon scientists and engineers. The deadline for submissions has been extended to November 12.

Amazon and Carnegie Mellon University launch strategic AI Innovation Hub: Amazon will fund research projects, doctoral fellowships, and community-building initiatives designed to accelerate innovation in foundational and emerging technologies across generative AI, robotics, natural language processing, and cloud computing.

Conference roundup

• EMNLP 2025, November 4-9
• NeurIPS 2025, December 2-7

INFORMS names 12 Leaders and Innovators in Operations Research and Analytics for Class of 2025 Fellows: Stefan Karisch, Amazon director of air science & technology, is recognized as 2025 INFORMS Fellows for his contributions to operations research and optimization-based aircraft operations planning.

Featured publications

• Context length alone hurts LLM performance despite perfect retrieval
• Kaputt: A large-scale dataset for visual defect detection
• MURPHY: Reflective multi-turn reinforcement learning for self-correcting code generation in large language models
• Rethinking LLM uncertainty: A multi-agent approach to estimating black-box model uncertainty

LinkedIn | X/Twitter | Facebook | Instagram | GitHub | RSS

© 1996-2025 Amazon.com, Inc. or its affiliates | Privacy | Conditions of Use