Amazon Science

The latest news and research from Amazon's science community. #AmazonScience

[Amazon Science @AmazonScience]

Attending #CVPR2026 in Denver? We'll be at booth 349 showcasing our latest research with live demos throughout the conference. Stop by to meet our scientists and see their work in action.

Check out the full booth schedule:

[Amazon Science @AmazonScience]

Next stop: Denver. 🏔️

Amazon will be at @CVPR with workshops, live demos, and new computer vision research. Come find us at booth 349. #CVPR2026 https://t.co/kx1oBeH48t

[Amazon Science @AmazonScience]

A passive optical device called a ShuffleBox and a routing protocol called Spraypoint just made flat datacenter networks as practical as fat trees, with 69% fewer routers and up to 33% better throughput. https://t.co/M9mEzCWfrx

[Amazon Science @AmazonScience]

Announcing the #AmazonResearchAwards fall 2025 recipients:

🔍 68 researchers
🏫 49 universities
🌏 11 countries

Each gains access to 800+ Amazon public datasets and AWS AI/ML tools. Meet the cohort: https://t.co/47jUdPuRrV

[Amazon Science @AmazonScience]

Current LLMs overallocate to MLP layers. A new scaling law presented at #ICLR2026 identifies architectures with up to 47% higher throughput and equivalent accuracy. https://t.co/BP0zxQ5zJp

[Amazon @amazon]

If a model is 50% reliable, it's 0% useful.

Michael Giannangeli, Head of Agentic AI at Amazon Nova, breaks down one of the hardest problems in agentic AI: reliability.

Models need to match or beat human performance, or they won't get adopted. To close the gap, teams are building an ecosystem of reinforcement learning "gyms" where AI agents can hone their skills. Think of it like a virtual training ground where agents practice tasks through trial and error, building reliability one rep at a time. 💪

[Amazon Science @AmazonScience]

The next generation of AI researchers is using Amazon's Trainium chip to build the next AI breakthrough. Teams at @UCBerkeley, @MIT, @CarnegieMellon, and @UCLA are seeing results across ML, healthcare, and quantum computing. https://t.co/h678Pshfry

[Amazon Science @AmazonScience]

Promptimus, a new automated prompt-engineering framework from Amazon, improves already good LLM prompts by focusing on specific failure points. It achieves the best result on 16 of 20 benchmarks. https://t.co/11YxqEuatn

[Amazon Science @AmazonScience]

https://t.co/YQJs2PLqOd

[Amazon News @amazonnews]

https://t.co/aF37lO75pX

[Amazon Science @AmazonScience]

🎙️ Reinforcement learning gyms, agentic shopping, and why 50% reliability = 0% utility. A panel at SXSW breaks down what it takes to ship AI agents that work at scale: https://t.co/U3KgpZO2wF

[Amazon Science @AmazonScience]

What keeps a delivery network running when a viral product spikes or a snowstorm hits? Amazon scientists designed a system where shocks trigger effective adaptation rather than crises. https://t.co/OceyfKViu2

[Amazon Science @AmazonScience]

From auction theory to supply chain coordination. Amazon's Flo Pro system achieves optimal plans where Amazon and its vendors keep their private cost data secret.
https://t.co/wzk5Jcyoze

[Amazon Science @AmazonScience]

✅ 70+ RAI tools
✅ 500+ research papers
✅ tens of thousands of hours of training

How Amazon embeds responsible AI across the model lifecycle: https://t.co/e9DTyIghcK

[Amazon Science @AmazonScience]

📣 Deadline extended to May 16. The Amazon Research Awards spring 2026 call for proposals is still open for submissions. Successful applicants receive unrestricted funds, AWS promotional credits, and training resources. https://t.co/scxlMSM7WZ

[Penn Engineering AI @PennEngAI]

AI is reshaping mathematical research. In @AmazonScience, @PennEngineers faculty and Amazon Scholars Michael Kearns and Aaron Roth describe how AI tools can generate formal proofs and accelerate discovery. Read more: https://t.co/8UmVRzqQGg #AIandMath

[Amazon Science @AmazonScience]

Amazon researchers reproduced three attacks that extract private training data from AI models and the cryptographic defenses that stop them: https://t.co/KnbDUzPMhv

[Matt Lease @mattlease]

Work presented today @ Amazon Research Day: w/ Houjiang Liu, Yujin Choi, Sanjana Gautam, Gabriel Jaffe, & Soo Young Rieh. Effects of Varying Control vs. Automation in Human-in-the-loop Research Ideation. Paper: https://t.co/fmcZ4a7qLl

@CosmicAI_Inst @UTGoodSystems @UTiSchool

[Amazon @amazon]

Amazon is proud to be named to @TIME's TIME100 Companies AI list—recognized for providing the foundational compute, powered by our custom Trainium AI chips, that the AI industry is being built on.

[Amazon Science @AmazonScience]

Another inspiring ICLR in the books. Thank you to everyone who connected with our team in Rio de Janeiro. Great to be among the global research community advancing deep learning. #ICLR2026