Rahul Gupta
@rahul1987iit
Responsible AI @Amazon AGI (Nova models) | Organizer @TrustNLP | Organizer @UnlearningSEM
Excited to share our safety framework evaluations with the community! See full thread from the first author below. @AmazonScience #AmazonNovaModels #ResponsibleAI
🚨 Excited to share the first frontier risk evaluation report for Amazon Nova Premier: “Evaluating the Critical Risks of Amazon’s Nova Premier under the Frontier Model Safety Framework”! This is the first comprehensive evaluation of Nova Premier’s frontier safety, aligned with…
Looking forward to the winner announcements!
A fun look at the Nova AI Challenge finals. Winners announced tomorrow! 🏆
Nova Act is now⚡️ enterprise ready ⚡️ and we've added new capabilities to our preview to help you take your prototype to production—with 90%+ reliability across our early enterprise customer use cases!
More details on Big Sleep and our latest security work: blog.google/technology/saf…
🎉 Thrilled to be presenting my first paper at @icmlconf! "Just Enough Shifts: Mitigating Over-Refusal in Aligned Language Models with Targeted Representation Fine-Tuning" We introduce ACTOR—a lightweight, activation-based training method that reduces over-refusal without…
Yet another fruitful collaboration between LIME@USC lab and NovaRAI team! @jieyuzhao11 , @linxins2
Our paper just got accepted by COLM2025! See you at Montreal for further discussion on how to discover LLM deficiencies! #colm2025
As AI agents near real-world use, how do we know what they can actually do? Reliable benchmarks are critical but agentic benchmarks are broken! Example: WebArena marks "45+8 minutes" on a duration calculation task as correct (real answer: "63 minutes"). Other benchmarks…
If you're building a benchmark for AI agents, you absolutely need this groundbreaking paper from UIUC, Stanford University, and other leading institutions. The researchers introduce ABC—the Agentic Benchmark Checklist—a practical, actionable guide for creating reliable and…
Our research on embodied AI agents that can perceive, learn, act and interact in the virtual and physical worlds. #metaAI #AIAgent #embodied #worldmodel #superintelligemce arxiv.org/abs/2506.22355
Another launch from the Nova team!
Just launched virtual try-on in Amazon Nova Canvas! This gives retailers a powerful way to help shoppers see products as they’d really look – from trying on different shirt colors to previewing how a coffee table fits in their living room. Learn more: aws.amazon.com/blogs/aws/amaz…
First author's take on kaleidoscopic teaming!
Second update of the week: Check out our recent work where we coin the term kaleidoscopic teaming. What is kaleidoscopic teaming you might ask.