Kaifeng Lyu
@vfleaking
Assistant Professor @ Tsinghua University
Excited to share our new method ✏️PENCIL! It decouples space complexity from time complexity in LLM reasoning, by allowing model to recursively erase and generate thoughts. Joint work w. my student @chenxiao_yang_ , along with @BartomNati and @McAllesterDavid.
I've discovered a truly marvelous idea for building AGI, but Twitter's space limit won't let me explain it! Damn! 😫 Introducing ✏️PENCIL, a new LLM reasoning paradigm that generates and erases thoughts, enabling longer and deeper thinking with shorter context. #ICML2025 🧵1/n…
Excited to present our paper this morning at ICLR 2025, revealing the gap in CoT reasoning between RNNs and Transformers! Poster Presentation: 🗓 Saturday, April 26 📷 10:00 AM – 12:30 PM 📍 Hall 2, Poster #640
Check out our new paper! We explore the representation gap between RNNs and Transformers. Theory: CoT improves RNNs but is insufficient to close the gap. Improving the capability of retrieving information from context is the key (e.g. +RAG / +1 attention). arxiv.org/abs/2402.18510
What's the optimal learning rate schedule for LLM pretraining? Come meet us this afternoon! Poster Presentation: 🗓 Friday, April 25 🕒 3:00 PM – 5:30 PM CST 📍 Hall 3 + Hall 2B, Poster #237
📢 Come meet us at #ICLR2025! We'll be presenting our Multi-Power Law — a new approach to predicting full pretraining loss curves across LR schedules — during the poster session: 🗓 Friday, April 25 🕒 3:00 PM – 5:30 PM CST 📍 Hall 3 + Hall 2B, Poster #237 Expect your feedback!
Thrilled to share that our paper “Safety Alignment Should be Made More Than Just a Few Tokens Deep” has received an ICLR 2025 Outstanding Paper Award! This project began as an effort to defend against fine-tuning attacks with constrained supervised fine-tuning (SFT). Along the…
Thrilled to know that our paper, `Safety Alignment Should be Made More Than Just a Few Tokens Deep`, received the ICLR 2025 Outstanding Paper Award. We sincerely thank the ICLR committee for awarding one of this year's Outstanding Paper Awards to AI Safety / Adversarial ML.…
We will present this paper at #ICLR2025! 1. 𝐎𝐫𝐚𝐥 𝐒𝐞𝐬𝐬𝐢𝐨𝐧 𝟏𝐃 (𝐓𝐡𝐮𝐫𝐬𝐝𝐚𝐲 𝟏𝟎:𝟒𝟐𝐚𝐦) @PandaAshwinee will give a talk 2. 𝐏𝐨𝐬𝐭𝐞𝐫 𝐒𝐞𝐬𝐬𝐢𝐨𝐧 𝟒 (𝐅𝐫𝐢𝐝𝐚𝐲 𝟑𝐩𝐦) Come to chat with @PandaAshwinee @vfleaking @infoxiao @abeirami Unfortunately, I…
Our recent paper shows: 1. Crrent LLM safety alignment is only a few tokens deep. 2. Deepening the safety alignment can make it more robust against multiple jailbreak attacks. 3. Protecting initial token positions can make the alignment more robust against fine-tuning attacks.
The success of RLHF depends heavily on the quality of the reward model (RM), but how should we measure this quality? 📰 We study what makes a good RM from an optimization perspective. Among other results, we formalize why more accurate RMs are not necessarily better teachers! 🧵
🚨Ever wonder why diffusion models generate nonsensical text? Our latest study at #ICLR2025 uncovers "Local Generation Bias"—a hidden training bias causing textual hallucinations! 🧠 Key finding: Diffusion models independently generate symbols locally without global context.
Kids use open textbooks for homework. Can LLM training benefit from "helpful textbooks" in context with no gradients computed on these tokens? We call this Context-Enhanced Learning – it can exponentially accelerate training while avoiding verbatim memorization of “textbooks”!…
Thanks to everyone who joined and supported the M3L workshop this year! It was so exciting to see so many inspiring ideas and discussions. Unfortunately, I got a fever one day before the workshop and couldn’t attend in person. Looking forward to seeing you all next year!
Hope everyone had fun at the 2nd workshop of M3L! Many thanks to the speakers, authors, reviewers, and participants for making this workshop a success. We had a full house again, and we hope to see you next year! 💡
We've extended the #M3L submission deadline to October 1st AoE to align with ICLR timelines. We look forward to your work!
📡Join us at the 2nd workshop on Mathematics of Modern Machine Learning (M3L) at #NeurIPS2024! sites.google.com/view/m3l-2024/ Submission deadline: September 29, 2024
💡 The Mathematics of Modern Machine Learning (M3L) workshop is back for its 2nd edition at NeurIPS 2024. Submit your work and share your perspectives on modern ML theory! 📅 Submission ddl: Sept 29, 2024 (2 days after ICLR abstract ddl) 🌐 sites.google.com/view/m3l-2024
📡Join us at the 2nd workshop on Mathematics of Modern Machine Learning (M3L) at #NeurIPS2024! sites.google.com/view/m3l-2024/ Submission deadline: September 29, 2024
1/ LLMs are often used to generate text new math questions. But can they generate challenging math questions? Current methods yield Qs that're either easy or too similar to existing questions. Our new paper "AI-Assisted Generation of Difficult Math Questions" shows how to…
Our recent paper shows: 1. Crrent LLM safety alignment is only a few tokens deep. 2. Deepening the safety alignment can make it more robust against multiple jailbreak attacks. 3. Protecting initial token positions can make the alignment more robust against fine-tuning attacks.
Our 12 scaling laws (for LLM knowledge capacity) are out: arxiv.org/abs/2404.05405. Took me 4mos to submit 50,000 jobs; took Meta 1mo for legal review; FAIR sponsored 4,200,000 GPU hrs. Hope this is a new direction to study scaling laws + help practitioners make informed decisions