Justin Cui

UCLA PhD Student

[Justin Cui @cuijiaxingfb]

@HaoZhang623 very inspiring.

[Liner @search_liner]

Liner is partnering with @spoticlr at #ICLR2026 — supporting Best Paper and Travel Awards for LLM research.

And to celebrate, we're giving away:
✈️ Round-trip flights + hotel to #ICML2026 in Seoul
🎁 $300 Liner Credits

Follow @search_liner + repost to enter by 4/27.

Liner is built for research workflows. Find papers, verify sources, and write with citations in one place.

See you in 🇧🇷 and 🇰🇷!

@iclr_conf @icmlconf

[Victor.Kai Wang @VictorKaiWang1]

very happy to release this parameter generation work.
from P-diff (2024), RPG (2025), DnD (2025) to HY-WU, parameter generation becomes more and more practical. thx @TencentHunyuan and happy to work with @oahzxl @mmbronstein @ZiqiaoWang63428 @cindy_x_wu @YangYou1991 @VITAGroupUT

[Victor.Kai Wang @VictorKaiWang1]

Customizing Your LLMs in seconds using prompts🥳!
Excited to share our latest work with @HPCAILab, @VITAGroupUT, @k_schuerholt, @YangYou1991, @mmbronstein, @damianborth : Drag-and-Drop LLMs(DnD). 2 features: tuning-free, comparable or even better than full-shot tuning.(🧵1/8)

[Justin Cui @cuijiaxingfb]

@JieWu_10 The results look amazing!!!!

[JIE WU @JieWu_10]

💡Let me introduce Seedance 1.0 Video Foundation model🎬, a model that supports multi-shot video generation from both text and images.

🥇Seedance 1.0 ranked #1 in both text-to-video & image-to-video on Artificial Analysis’s third-party benchmarks. (Note: Veo 3 Preview’s audio was not available for fairness.

🍹It achieves breakthroughs in semantic understanding and prompt following, and can create 1080p videos with smooth motion, rich detail, and cinematic aesthetics.

Key Features:

🍭Smooth & Stable Motion: Seedance 1.0 has a wide dynamic range, generating fluid, large-scale movements. From subtle expressions to active scenes, it maintains a high level of stability and physical realism.

🍦Native Multi-Shot Storytelling:Natively supports the generation of narrative videos with multiple cohesive shots. It maintains consistency in the main subject, visual style, and atmosphere across shot transitions and spatio-temporal shifts.

🔥Diverse Stylistic Expression: From photorealism and cyberpunk to traditional Chinese animation and claymation, Seedance 1.0 can accurately interpret diverse stylistic prompts to support a wide range of creative needs.

👑Precise Semantic & Prompt Following: Accurately parses natural language prompts, enabling stable control over multi-agent interactions, complex action sequences, and a rich variety of camera movements to precisely translate your textual concepts into videos.

Technical Report: https://t.co/EG1MdT9Rs4
Official Website: https://t.co/rA80AuTRKe

#seed #seedance #bytedance #video-generation #LLM #ai

[Justin Cui @cuijiaxingfb]

@JieWu_10 Amazing performance, great work!!!

[JIE WU @JieWu_10]

🌟 Let me introduce Seedream 3.0, the ultimate Image Generation Foundation Model brought to you by ByteDance’s Seed Team.

🚀 This game-changer landed on multiple platforms, like Doubao and Jimeng in early April 2025.

🙌Seedream3.0 has the capabilities of 2K resolution direct output, small text layout, and high generation efficiency, which greatly reduces the threshold for visual creativity in posters and covers.

Key Features:
🌟 Comprehensive capability upgrades—everything is better, faster, and more powerful.
✍️ Enhanced text rendering—perfect for complex Chinese characters and typography.
🎨 Aesthetic improvements—your visuals will look amazing next-level. 😍
📸 Native high-resolution output—up to 2K for stunning details.
💡 Efficient inference cost—speed and performance optimized.

Seedream 3.0 (formerly known as Mogao) entered the Artificial Analysis rankings and landed in the first tier, right alongside GPT-4o. 💪 It crushed other models like Recraft V3, HiDream, Reve Image, Imagen 3 (v002), FLUX1.1 Pro, and Midjourney v6.1. 🔥

📄 Technical Report: https://t.co/gkzGZZJwqY 🌐 Official Website: https://t.co/x9cxdR0pfk

#seedream_3_0 , #genAI , #text_to_image_model , #AI #image_generation

[JIE WU @JieWu_10]

🚀 Introducing Seedream 2.0: A Native Chinese-English Bilingual Image Generation Foundation Model by ByteDance. Currently powering Dreamina and Doubao apps.
✨ Key Features:
1. Powerful General Capability
2. Native Bilingual Comprehension Ability
3. Excellent Text Rendering
4. Deep Understanding of Chinese Characteristics
📢 For the first time, we’re sharing the technical details of our text-to-image model.
🙌Through extensive experimentation, we demonstrate that Seedream 2.0 achieves SOTA performance across multiple aspects, including prompt-following, aesthetics, text rendering, and structural correctness.
💡 Compared to Flux 1.1 Pro, MJv6.1, DALL·E,Hunyuan, and 4o, Seedream 2.0 delivers outstanding overall performance.
ArXiv: https://t.co/5wKW3SjJDD
Website: https://t.co/aAhgdcsCRW

[Justin Cui @cuijiaxingfb]

@JieWu_10 Great work, congrats on the release!

[TurningPoint AI @TurningPointAI]

🚀 We’re excited to share our latest work! Welcome to the first successful "aha moment" on multimodal reasoning.

"Aha moment" is featured by improved response length & performance. It emerges during RL of an unaligned base model on multimodal tasks. Aha moment for language reasoning was originally observed on DeepSeek-R1-Zero.

🔍 Key Findings:
1. Directly applying GRPO on an unaligned 2B base model could elicit the multimodal “aha moment”: thinking capability marked by spontaneous reasoning strategy and increased reasoning length
2. Visual-centric task could benefit from long Chain-of-Thoughts

💻 Discover more on our notion blog and project page!

Detailed Research Blog: Follow our complete journey and technical insights at our Notion Blog:
🔗https://t.co/DY3UgYpAwr

Reproduce Our Results: Access and build upon our implementation at GitHub:
🔗https://t.co/mEyMh1BS6b

Presented by: TurningPointAI Team
🔗https://t.co/utzaQrQCro

#turningpointai #Smallmodel #MultimodalR1 #DeepseekR1 #R1 #Deepseek #AI #MultimodalReasoning #Qwen #QwenVL #DeepSeekR1zero

[Zekai Li @Richard91316073]

🚀 We’re thrilled to introduce DD-Ranking: Rethinking the Evaluation of Dataset Distillation—a unified, user-friendly, and long-term maintained benchmark for dataset distillation (DD)!
📢 Together with 20 institutions worldwide, we’re releasing our Repo, API Documentation, and Leaderboard:
🔗 Repo: https://t.co/n2WT5ziaJJ
📖 Documentation: https://t.co/9nqogEM350
🏆 Leaderboard: https://t.co/tAHZKXOLH1 🧵1/7

[Justin Cui @cuijiaxingfb]

@VictorKaiWang1 the work is very inspiring!

[Victor.Kai Wang @VictorKaiWang1]

Generating ~200 million parameters in just minutes! 🥳
Excited to share our work with @MTDovent , @heisejiasuo96 , and @YangYou1991: 'Recurrent Diffusion for Large-Scale Parameter Generation' (RPG for short).
Example: Obtain customized models using prompts (see below). (🧵1/8)

[Justin Cui @cuijiaxingfb]

@cindy_x_wu Great work!

[Justin Cui @cuijiaxingfb]

@YUYANG_UCLA @baharanm Thanks for the great work!

[Justin Cui @cuijiaxingfb]

@furongh congrats!

[Justin Cui @cuijiaxingfb]

@furongh This is really inspiring. AI has indeed transformed our daily life so much.

[Justin Cui @cuijiaxingfb]

Claude 3.5’s over-refusal rate has significantly decreased while maintaining safety standards. Gemma-2-9b/27b models are comparable to Gemini-1.5, both of which are much safer than their earlier versions. Please see our leaderboard for the updated rankings.

Leaderboard: https://t.co/mjThIXjXW0

[Justin Cui @cuijiaxingfb]

@furongh Congrats!!!