Wenhao Zhu
114 posts
Wenhao Zhu
@Wenhao_NLP
AI researcher@ByteDance Seed | prev. @EdinburghNLP | Multilingual LLM & machine translation
Edinburgh, Scotland Katılım Ekim 2019
794 Takip Edilen588 Takipçiler
In Beijing and Hangzhou this week — want to talk to more AI researchers! Reach out.
English
Life update at the end of 2025: I’ve completed my PhD at @MITEECS and joined @thinkymachines to work on LLM archs
English
Calling for papers! 📢 Join us at the Multilingual Multicultural Evaluation (MME) workshop co-located at EACL.
Vilém Zouhar @ EACL@zouharvi
Do you have work on resources, metrics & methodologies for evaluating multilingual systems? Share it at the MME workshop🕵️co-located at EACL. Direct submission deadline in 10 days! (December 19th)! …al-multicultural-evaluation.github.io
English
Does anyone else feel the same? Gemini 3 Pro hasn’t been great for my daily use… feels worse than Gemini 2.5 Pro.
Guess I had my expectations way too high...
English
🥳🥳
Fuli Luo@_LuoFuli
Intelligence will inevitably evolve from language to the physical world, unlocking spatial intelligence for multi-modal perception, reasoning, generation, and action—essential for true AGI. I'm working on building this at @XiaomiMiMo, spearheading a creative and talented team!
ART
Wenhao Zhu retweetledi
📢 Announcing the First Workshop on Multilingual and Multicultural Evaluation (MME) — co-located with #EACL2026 🇲🇦
📅 Mar 24–29, 2026 | Rabat, Morocco
MME focuses on resources, metrics & methodologies for evaluating multilingual systems! …al-multicultural-evaluation.github.io
🗓️ Submit by Dec 19, 2025
English
Exciting to see the great potential of DuPO in scenarios where supervision label is scarce.
Shuaijie She ✈️ ICLR26@kevinprossj
🔥 Thrilled to introduce DuPO (Dual Learning-based Preference Optimization) - DuPO enables LLMs to get reliable and scalable self-supervision through duality-derived rewards. - General application in various tasks, eg, math reasoning and multingual translation. - Strong performance on various backbones, excelling both as a reward for training and as a reranker for inference. 🤗 Paper: huggingface.co/papers/2508.14… 📝 Blog: shesj-note.notion.site/dupo
English
Wenhao Zhu retweetledi
DuPO: Enabling Reliable LLM Self-Verification via Dual Preference Optimization
"We present DuPO, a dual learning-based preference optimization framework that generates annotation-free feedback via a generalized duality"
"DuPO decomposes a primal task’s input into known and unknown components, then constructs its dual task to reconstruct the unknown part using the primal output and known information (e.g., reversing math solutions to recover hidden variables"
English
@sarahookr @Cohere_Labs @cohere You’ve done an incredible job leading Cohere and empowering the multilingual community. Wishing you the best in your next adventure!
English
It has been an incredible honor to spend the past few years leading @Cohere_Labs @cohere .
This has been the adventure of a lifetime. However, after much deliberation, I made a tough decision 2 months ago it is time to say goodbye.
English
Wenhao Zhu retweetledi
The video in the link will surprise you. Trust me!
Shanbo Cheng@cshanbo
Not a social media/ X person, but still glad to announce Seed LiveInterpret 2.0. In short, it is an end-to-end, full duplex speech-to-speech simultaneous interpretation model. Achieves high-quality, ultra-low latency S2S translation. Website: seed.bytedance.com/en/seed_livein…
English
@devoto_alessio @PMinervini @simeng_ssun I'd like to recommend our LongBioBench as well! It supports infinite-length evaluation and enables controllable examination. arxiv.org/pdf/2506.02921
English
@PMinervini @simeng_ssun Good point! We support LongBench v1 and v2 in KVPress! As for MMLongBench, we are not planning to include it as we don't have KV Cache compression methods for VLMs (for now)
English
🏆 Our @nvidia KV Cache Compression Leaderboard is now live!
Compare state-of-the-art compression methods side-by-side with KVPress. See which techniques are leading in efficiency and performance. 🥇
huggingface.co/spaces/nvidia/…
English
Could multi-turn interaction the next promising direction for scaling?
Multi-Turn Interaction LLM Workshop @ NeurIPS 2025@mti_neurips
🚀 Call for Papers — @NeurIPSConf 2025 Workshop Multi-Turn Interactions in LLMs 📅 December 6/7 · 📍 San Diego Convention Center Join us to shape the future of interactive AI. Topics include but are not limited to: 🧠 Multi-Turn RL for Agentic Tasks (e.g., web & GUI agents, tool use) 🤝 Human-AI Interaction over time 🛡️ Alignment across extended interactions 📏 Evaluation of long-horizon tasks 🧩 Social learning, Open-Endedness, trust, and more 🌟 Featuring an all-star speaker lineup: Dawn Song @dawnsongtweets (UC Berkeley) Jason Weston @jaseweston (Meta FAIR) Natasha Jaques @natashajaques (University of Washington & Google DeepMind) Tim Rocktäschel @_rockt (UCL & DeepMind, tentative) Diyi Yang @Diyi_Yang (Stanford) Peter Henderson @PeterHndrsn (Princeton) Yu Su @ysu_nlp (OSU) Hannah Rose Kirk @hannahrosekirk (Oxford) 📣 Updates: Follow us here & spread the word! #NeurIPS2025 #LLMs #AIAlignment #MultiAgent #ReinforcementLearning #LanguageAgents #InteractiveAI
English
Wenhao Zhu retweetledi
🚀 Introducing Prefix-RFT to blend SFT and RFT!
SFT can learn more complex problems by mimicking, but can have poor generalization. RFT has better overall performance but is limited by the initial policy. Our method, Prefix-RFT, makes the best of both worlds!
GIF
English
Bytedance Seed is insane
they have so much stuff to publish it's going out of order
English
@Elaina43114880 @teortaxesTex 晚些时候可以进行在线测试，敬请期待！
The service will be available soon. Stay tuned!
Currently, we recommend deploying Seed-X on your own device with our released weights.
中文
ByteDance Seed released Seed-X, a Mistral-7B shaped LLM specialized for translation, apparently pretrained on ≈6.4B tokens, equaling the likes or R1 and 2.5-Pro in human evaluation.
«We deliberately exclude STEM, coding, and reasoning-focused data» lol
unexpected data paper
English