Susung Hong

@SusungHong

PhD student @uwcse | Intern @Google | Generative simulation and video/3D

Seattle, United States

susunghong.github.io

Joined October 2022

129 Following

243 Followers

Replies Reposts Media

Pinned
Susung Hong
@SusungHong
Mar 12
🎬Introducing COMIC — fully automated AI comedy Can AI be funny? We've seen AI solve math and code. Comedy is the opposite extreme, whose success is hard to define. 🍿Watch full videos here: susunghong.github.io/COMIC 📄See how we tackle open-ended comedy: arxiv.org/abs/2603.11048
00:00
1
Susung Hong
@SusungHong
Mar 22
Love this observation. LLMs struggle most where we can't verify their outputs (e.g., humor). One approach that works: arxiv.org/abs/2603.11048
sarah guo
@saranormous
Mar 20
Caught up with @karpathy for a new @NoPriorsPod: on the phase shift in engineering, AI psychosis, claws, AutoResearch, the opportunity for a SETI-at-Home like movement in AI, the model landscape, and second order effects 02:55 - What Capability Limits Remain? 06:15 - What
00:00
Susung Hong
@SusungHong
Mar 12
🎉 MusicInfuser has been accepted to #CVPR2026! We plug music perception into silent video diffusion and make it dance! 🎶 Adding a new sensor like audio to a pretrained diffusion model can be destructive. Check out how we tackle this: arxiv.org/abs/2503.14505
Susung Hong
@SusungHong
Mar 19, 2025
Text-to-video models are silent🔇, but does that mean they don't know music, beat, and tempo🎶? I'm excited to present MusicInfuser🎹, an adapter network which aligns silent dancing videos to music. Check out our paper, examples, code, and weights here: susunghong.github.io/MusicInfuser
00:00
Susung Hong
@SusungHong
Mar 30, 2025
My working guess is that they auto-regress image tokens, with some sort of previewing mechanism (like they can foresee remaining future tokens). At least the fact is that they're sending us a puzzle, which is pretty interesting🙂
Jie Liu
@jie_liu1
Mar 28, 2025
After hacking GPT-4o's frontend, I made amazing discoveries: 💡The line-by-line image generation effect users see is just a browser-side animation (pure frontend trick) 🔦OpenAI's server sends only 5 intermediate images per generation, captured at different stages 🎾Patch size=8
GIF
Susung Hong
@SusungHong
Mar 25, 2025
🚀#CVPR2025 Code and camera-ready version for "Perturb-and-Revise: Flexible 3D Editing with Generative Trajectories" has been dropped! @CVPR 📝 Paper: arxiv.org/abs/2412.05279 🧑🏻‍💻 GitHub: github.com/SusungHong/Per… 🌐 Project Page: susunghong.github.io/Perturb-and-Re…
Susung Hong
@SusungHong
Dec 9, 2024
✨ Exciting paper: "Perturb-and-Revise" brings the power of noise injection (like SDEdit) to 3D model parameters! By perturbing parameters before editing, it enables dramatic geometric changes to 3D objects using just text prompts.
00:00