Shengyi Qian

Vision isn't an "add-on"—and we have the data to prove it. 👁️⚡️ Thrilled to share our new work on Transfusion-style models. We explored treating visual data as a first-class citizen from day one, from architecture to scaling behavior. Check it out: 🔗 beyond-llms.github.io

@DJiafei Congrats!

4 years have been simply amazing! I’m happy to share that I have successfully defended my PhD! Thank you to everyone who came to support me, and most importantly, to my thesis committee, advisors, collaborators, friends, and family for being part of this journey.

@TongPetersb @ylecun @sainingxie @mengyer @YiMaTweets @LukeZettlemoyer @liuzhuang1234 Congrats!

I defended my thesis today! Sincere thanks to my advisors @sainingxie @ylecun and committee members: @mengyer @YiMaTweets @LukeZettlemoyer @liuzhuang1234. I could not have wished for a better PhD life, and I want to thank everyone who was part of this journey. Slides Link: tsb0601.github.io/data/defense_s…

Excited to share what we’ve been building at Meta Superintelligence Labs! We just released Muse Spark, our first AI model. It's a natively multimodal reasoning model and the first step on our path to personal superintelligence. We've overhauled our entire stack to support scaling, and this is just the beginning. ai.meta.com/blog/introduci…

1/ today we're releasing muse spark, the first model from MSL. nine months ago we rebuilt our ai stack from scratch. new infrastructure, new architecture, new data pipelines. muse spark is the result of that work, and now it powers meta ai. 🧵

@iamsashasax @AnthropicAI Congrats Sasha!

In a couple weeks I'm joining @AnthropicAI to work on pretraining after nearly 3 years at FAIR, developing post-training flywheels for physical intelligence (like SAM 3D) I'm stoked to build new capabilities for a model I personally love, with such thoughtful people

Beyond Language Modeling FAIR Meta and NYU present a deep dive into native multimodal pretraining. They show RAEs unify visual understanding/generation, vision/language data are complementary, world modeling emerges naturally, and MoE harmonizes vision's higher data hunger—paving the way for truly unified models.

Humans communicate through language and interact with the world through vision, yet most multimodal models are language-first. What happens when we go beyond language? 🤔 Beyond Language Modeling: a deep dive into the design space of truly native multimodal models Paper: arxiv.org/abs/2603.03276 Project: beyond-llms.github.io

Train Beyond Language. We bet on the visual world as the critical next step alongside and beyond language modeling. So, we studied building foundation models from scratch with vision. We share our exploration: visual representations, data, world modeling, architecture, and scaling behavior! [1/9]

Excited to share our latest work from Meta Superintelligence Labs! 🚀 We’re moving beyond static AI to agents that actually evolve with you. Our PAHF framework solves "Alignment Drift" through a continuous feedback loop. Check out the paper!

We are excited to host the 2nd 3D-LLM / VLA Workshop at CVPR this June! If your research explores the synergy between spatial intelligence, robotics, and language grounding, we invite you to submit your work. We also have an incredible lineup of speakers. Join us!

Have arrived in Suzhou! I will present DISCO paper in EMNLP 2025 Thursday’s noon poster session. Feel free to reach out and discuss! If you’re interested in Meta’s current position for both FTE or internships, also let me know! #EMNLP2025

Glad to have my PhD’s last work accepted by EMNLP 2025! Thank all my collaborators for their efforts. Expect to see you all in Soochow! #EMNLP25

Reasoning can be made much, much faster—with fundamental changes in neural architecture. 😮 Introducing Phi4-mini-Flash-Reasoning: a 3.8B model that surpasses Phi4-mini-Reasoning on major reasoning tasks (AIME24/25, MATH500, GPQA-D), while delivering up-to 10× higher throughput at 32K generation length with vLLM. 🤯 Model: huggingface.co/microsoft/Phi-… Codebase: github.com/microsoft/Arch… Blog: aka.ms/flashreasoning… Paper: aka.ms/flashreasoning… (1/8)

Interactable Digital Twins hold great promises. It allows us to train in sim and test in real. But can we go a step further? Can we deploy a robot w/o training? Key idea: simulate the outcome of each action with Digital Twins and use VLM as critic to select the best action.

We’re presenting 3D-MVP at CVPR poster #140 right now! #CVPR2025 #ComputerVision #MachineLearning #DeepLearning #3DVision #Robotics #AI #Research #PaperPresentation

One of the biggest bottlenecks in deploying visual AI and computer vision is annotation, which can be both costly and time-consuming. Today, we’re introducing Verified Auto Labeling, a new approach to AI-assisted annotation that achieves up to 95% of human-level performance while cutting labeling costs by up to 100,000x and time by 5,000x. Read the full paper: arxiv.org/abs/2506.02359

3️⃣ 3D-GRAND: Towards Better Grounding & Less Hallucination for 3D-LLMs. A large-scale dataset & models for improved 3D visual grounding. Project: 3d-grand.github.io #3DLLM #AI DM me if you're at #CVPR or want to chat about these! Looking forward to it!

2️⃣ Mosaic of Modalities: A Comprehensive Benchmark for Multimodal Graph Learning. Introducing a new benchmark for #MultimodalGraphLearning. Project: mm-graph-benchmark.github.io #MachineLearning

Thrilled to be heading to Nashville next week for #CVPR2025! Can't wait to connect with the community & dive into the latest in computer vision.