AI Briefing
Amazon Science news
The latest AI and developer news from Amazon Science, with short summaries.
Visit website →AWS uses Tiros automated reasoning to validate network segmentation exhaustively
The method analyzes network topology statically every hour to prove isolation rules hold across all modeled paths, replacing impractical dynamic packet sampling.
Mitra-v2: A tabular foundation model leading TabArena and ScoringBench
Mitra-v2, a second-generation tabular foundation model, is pretrained on tables with 10x more rows and 3x more columns than its predecessor, leading TabArena and ScoringBench.
Hamed Karimi Wins 2026 ECML PKDD Test-of-Time Award for 2016 Optimization Paper
The 2016 paper by Karimi, Nutini, and Schmidt established that the Polyak–Łojasiewicz condition guarantees linear convergence for standard training methods, bridging a gap between theory and practice.
AWS and Booking.com combine knowledge distillation and Medusa to cut LLM costs by two-thirds while speeding up inference up to 20x
The combined approach matches the quality of a 70-billion-parameter teacher model while running 10 to 20 times faster and at less than one-third the cost.
Audible uses controllable TTS and BookLLM to improve audiobook character differentiation
The system uses BookLLM to profile characters with 97% accuracy in English and 94% in Spanish, enabling distinct and consistent voices across full-length books.
Amazon releases TorchNeuron, an open-source PyTorch backend for AWS Trainium
The open-source backend uses AI agents to benchmark 841 representative models, ensuring accurate execution and reducing time to the first training step by about a third.
Amazon partners with AVERI for independent review of AI safety evaluation datasets
The review confirmed comprehensive topical coverage and low contamination risk while identifying gaps in linguistic diversity and attack sophistication.
AWS introduces priming to convert pre-trained Transformers into hybrid state space models with 99% less pretraining cost
The method reduces pretraining costs by over 99% and enables controlled comparisons of hybrid architectures like Gated KalmaNet, Gated DeltaNet, and Mamba-2.
Graph-Centric Agentic Intelligence for Network Root Cause Analysis
Demonstrated with NTT DOCOMO, the approach reduces root cause analysis from hours to minutes on commercial networks.
Reactor and Amazon Optimize Real-Time Video Generation on Trainium Using NKI
The collaboration reduced 3D-RoPE kernel latency from 5 seconds to 1.8 milliseconds, enabling real-time generation above 16 fps.
Amazon Launches AI, Energy, and Healthcare Research Initiative with Stanford
Amazon has officially launched an initiative for joint research in AI, energy, and healthcare with Stanford University.
Amazon Releases Three Papers on AI-Based Antibody Design and Evaluation
The Amazon Bio Discovery team released three AI-based antibody research papers covering binding strength prediction, developability assessment, and novel drug design.
Why LLM Research Agents Generalize Without Benchmark Overfitting: 'Compressibility' Is Key
LLM research agents generalize without benchmark overfitting because their successful strategies possess high compressibility.
Verifying Rust Code with Verus: Amazon Highlights 'Proven Correctness'
Amazon uses Verus to mathematically prove the correctness of Rust code, providing guarantees that go beyond Rust's existing safety features.
Can We Trust LLM Judges When They Agree?
A new aggregation method accounting for correlations among LLM judges improved accuracy by 9-14% over existing methods.
SOP-Bench: A New Benchmark for Evaluating AI Agents on Real-World Business Procedures
SOP-Bench, a benchmark combining real-world corporate SOPs and tools to verify AI agents' task execution capabilities, has been released.
A Decade of Mathematical Certainty: A Retrospective on the Automated Reasoning Group
Amazon's Automated Reasoning Group reflects on its past decade of research and achievements.
AWS Trainium's Frontier Competition: Co-designing Models and Kernels with Purpose-Built AI Chips
AWS optimizes AI models and kernels alongside chip design to strengthen Trainium's competitiveness.
Amazon Announces 34 Winners of Build on Trainium Research Grants
Amazon is providing Trainium research support to 34 AI research teams from 30 universities.
A Method for Coordinating Multitask Machine Learning Using Industrial Control Systems
ControlG applies industrial PID control principles to efficiently coordinate conflicting multi-task learning objectives.