Puyin Li

CS PhD student @Stanford

[Puyin Li @PuyinLi]

QuantiPhy challenges VLMs to reason about the physical world with numerical accuracy, estimating size, velocity, and acceleration directly from video.
👉 https://t.co/PhwqXEu4eF

#NeurIPS2026 #VLM #MultimodalAI #PhysicalAI #WorldModels

[Puyin Li @PuyinLi]

💰 Announcing prizes for the QuantiPhy Challenge @ NeurIPS 2026!

🏆 $3,500 total cash prizes
Main Track
🥇 $1,000 · 🥈 $500 · 🥉 $250
Open-Weight Track
🥇 $1,000 · 🥈 $500 · 🥉 $250

⏰ Team registration: Oct 9, 23:59 AoE
⏰ Submission deadline: Oct 23, 23:59 AoE

[Puyin Li @PuyinLi]

Whether you're working on VLMs, multimodal reasoning, or world models, we'd love to see what your models can do.

Looking forward to seeing everyone on the leaderboard! 🚀🔥

[Puyin Li @PuyinLi]

🚀 The QuantiPhy Challenge @ NeurIPS 2026 is officially LIVE!

Can AI truly reason about physics quantitatively?

Team registration and submissions are now open! 🏆

👉 https://t.co/3fnHC3RuFQ

[Puyin Li @PuyinLi]

We're excited to invite the community to tackle one of today's most challenging problems in multimodal AI.
✅ Official NeurIPS 2026 Competition Track
✅ Team registration open
✅ Submission portal live
✅ Official leaderboard

[Tiange Xiang @xxtiange]

Can you identify AI-generated videos by watching how human figures move in the video?

We introduce 🏃‍♂️HumanScore💃 — a benchmark for evaluating human motions in generated videos through biomechanics, not just visual realism.

Check out our project page: https://t.co/D9LCSgoAq5

#AI #ComputerVision #VideoGeneration #StanfordAI

[Stanford HAI @StanfordHAI]

Current models struggle with basic physics estimates, limiting their use in robotics and autonomous systems. Stanford scholars developed QuantiPhy, a benchmark to evaluate and improve AI's ability to reason about physical properties: https://t.co/4hPbCbD1yL

[Tiange Xiang @xxtiange]

‼️VLMs/MLLMs do NOT yet understand the physical world from videos‼️

In our recent work, we found that even the most advanced AI models still lag behind humans in one key aspect: reasoning about the kinematic properties of objects from videos.

Takeaways:
1. ChatGPT 5.1 leads overall among 21 advanced VLMs, followed by Gemini 2.5 Pro/Flash.
2. Grok 4.1 delivers impressive performance at the lowest API cost.
3. Qwen3-VL is the top-performing open-source model.

Read here: https://t.co/5lagvLNE37
🧵1/N