Do large multimodal models understand how to make dresses for your winter holiday party💃? We introduce AIpparel, a vision-language-garment model capable of generating and editing simulation-ready sewing patterns from text and images. Project page at georgenakayama.github.io/AIpparel/.

🇨🇦 Happy to present GrounDiT at #NeurIPS2024! Find out how we can obtain **precise spatial control** in DiT-based image generation! 📌 Poster: Fri 4:30PM - 7:30PM PST 💻 Our code is also released at: github.com/KAIST-Visual-A…

new paper! 🫡 we introduce 🪓AxBench, a scalable benchmark that evaluates interpretability techniques on two axes: concept detection and model steering. we find that: 🥇prompting and finetuning are still best 🥈supervised interp methods are effective 😮SAEs lag behind

Reasoning is all the rage these days. If you want to save some time and get to the crux of how to enable reasoning in LLMs, here’s a list of 10 recent papers that I find most informative, along with my notes: (Full thread in doc: docs.google.com/document/d/1TW…) 1/11

🏡Building realistic 3D scenes just got smarter! Introducing our #CVPR2025 work, 🔥FirePlace, a framework that enables Multimodal LLMs to automatically generate realistic and geometrically valid placements for objects into complex 3D scenes. How does it work?🧵👇

An LLM generates an article verbatim—did it “train on” the article? It’s complicated: under n-gram definitions of train-set inclusion, LLMs can complete “unseen” texts—both after data deletion and adding “gibberish” data. Our results impact unlearning, MIAs & data transparency🧵

Excited to share our work: Gaussian Mixture Flow Matching Models (GMFlow) github.com/lakonik/gmflow GMFlow generalizes diffusion models by predicting Gaussian mixture denoising distributions, enabling precise few-step sampling and high-quality generation.

Excited to share our #CVPR2025 Highlight work, BlenderGym🏋️‍♂️, on benchmarking multimodal LLMs on graphics editing! Curious how inference compute allocation impacts performance, or how your favorite MLLM stacks up? Check out our paper and benchmark!👇

Check out our recent work NVIDIA AI on a feedforward, open-world 3D part segmentation enabling various other downstream applications! 🚀 w/ Minghua Liu Donglai Xiang Hao Su Sanja Fidler Nick Sharp Jun Gao 🔗 Webpage: research.nvidia.com/labs/toronto-a…

Can you rotate a dice 🎲 in your head? Mental imagery plays a key role in perspective reasoning for humans - but can it help VLMs reason spatially? We show that Abstract Perspective Change significantly improves VLM reasoning from unseen views. Check out our preprint for more:

new paper! 🫡 why are state space models (SSMs) worse than Transformers at recall over their context? this is a question about the mechanisms underlying model behaviour: therefore, we propose using mechanistic evaluations to answer it!

We are presenting 17:00-19:00 today at Poster 267 in ExHall D for #CVPR25! Come and check out the first #VLM #3D #Graphics Benchmark! 📣📣📣

Are AI scientists already better than human researchers? We recruited 43 PhD students to spend 3 months executing research ideas proposed by an LLM agent vs human experts. Main finding: LLM ideas result in worse projects than human ideas.

We introduce MoVer, a Motion Verification DSL that automatically checks if AI-generated motion graphics animations match your text prompts! We make it easy for designers to specify and verify complex animations with LLM-powered iterative refinement. Catch our #SIGGRAPH2025 talk:

Does GPT-5 understand perspective change? GPT-5(-Thinking) still often struggles with simple *perspective changes*, a basic visual/spatial reasoning skill that could be essential for VLMs to reach *human-level intelligence* 👀

Calling all digital artists 🧑‍🎨 Have you ever forgotten to put objects on separate layers? Introducing InkLayer, a segmentation algorithm that makes scene sketches easy to edit. I’m presenting at #SIGGRAPH2025 on Wednesday, 11:45–11:55 am, West Building 118–120. See you there!

Soon, AI agents will act for us—collaborating, negotiating, and sharing data. But can they truly protect our privacy? We simulate privacy-critical scenarios, using alternating search to evolve attacks and defenses, uncovering severe vulnerabilities and building protections.

New paper! We explore a radical paradigm for AI evals: assessing LLMs on *unsolved* questions. Instead of contrived exams where progress ≠ value, we eval LLMs on organic, unsolved problems via reference-free LLM validation & community verification. LLMs solved ~10/500 so far:

What if your "For You" feed actually represented the values you care about? We introduce Alexandria, a browser extension that allows you to re-rank your Twitter feed in real time based on values you choose.

LLMs are powerful, but they don't know your world. This knowledge gap can lead to generic, unhelpful, or incorrect responses. In our #UIST2025 paper, we explore how users can fill these gaps through creating a community knowledge ecosystem, giving models access to more specific