Amazon Science

15 articles in the TruthFoundry index. Each links out to the original.

• Amazon Science optimizes travel LLMs with Booking.com for speed and cost · Amazon and Booking.com combined LLM optimizations deliver 20x speed at 1/3 operating cost.
• Amazon Audible develops improved controllable text-to-speech for audiobooks · Amazon Audible introduced a controllable TTS system to generate natural consistent audiobook character voices.
• Amazon releases TorchNeuron PyTorch backend for AWS Trainium AI chips · Amazon launches open-source TorchNeuron to run PyTorch code on AWS Trainium hardware.
• Amazon partners with AVERI for independent AI safety evaluation review · Amazon engaged independent nonprofit AVERI to audit its internal AI safety testing processes.
• Amazon AWS introduces priming to convert Transformers without full retraining · Amazon AWS researchers present priming, cutting hybrid AI model training cost over 99%.
• Amazon Science outlines graph-centric agentic AI for network fault analysis · Amazon Science details graph-based agentic intelligence for rapid network root cause resolution.
• AWS and Reactor Enable Real-Time Video Generation on Trainium · AWS Neuron Science and Reactor collaborate to deploy autoregressive diffusion models for real-time video on Trainium chips.
• Amazon launches formal research partnership with Stanford for AI advancement · Amazon and Stanford University launch a joint research initiative for AI, energy, and healthcare.
• Amazon Science publishes three AI papers for therapeutic antibody design · Amazon Bio Discovery releases three AI research papers advancing computational antibody development.
• Amazon Science Links Compression To ML Research Agent Overfitting Resistance · Amazon Science research explains why machine learning research agents avoid overfitting.
• Amazon uses Verus verifier to prove correctness of critical Rust code · Amazon employs open-source Verus tool to formally verify critical Rust infrastructure code.
• Amazon Science presents correlation-aware LLM judge aggregation at ICML · Amazon researchers introduce method to correct correlated bias in LLM judge panels.
• Amazon Science releases SOP-Bench benchmark for real-world AI agent testing · Amazon Science introduces SOP-Bench, an open benchmark for evaluating AI agents on business SOPs.
• Amazon's Automated Reasoning Group proves cloud systems with math over a decade · Amazon's Automated Reasoning Group uses mathematical logic to verify AWS security and reliability at scale.
• AWS Launches Trainium Frontier Competition for Hardware-Native AI Models · AWS invites researchers to compete in a challenge optimizing LLMs for its purpose-built Trainium chips.