1. X
  2. Chi Han
Log inSign up
Chi Han
55 posts
user avatar
Chi Han
@Glaciohound
CS PhD student at UIUC, interested in language models and their understanding.
Champaign, IL
glaciohound.github.io
Joined September 2021
266
Following
782
Followers
RepliesRepliesMediaMedia

New to X?

Sign up now to get your own personalized timeline!

Create account

By signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.

Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
Don't miss what's happening
People on X are the first to know.
Log inSign up
  • user avatar
    Chi Han
    @Glaciohound
    Aug 14, 2024
    🎖Excited that "LM-Steer: Word Embeddings Are Steers for Language Models" became my another 1st-authored Outstanding Paper #ACL2024 (besides LM-Infinite) We revealed steering roles of word embeddings for continuous, compositional, efficient, interpretable& transferrable control!
    Image
    Image
    Image
    This post is unavailable.
    25K
  • user avatar
    Chi Han
    @Glaciohound
    Mar 15, 2024
    Excited that LM-Infinite has been accepted into #NAACL2024 ! It is the first-of-its-kind zero-shot length generalizations for language models, with 200M length inference and downstream (Retrieval, Qasper) improvements! Great thanks to all my collaborators!
    arXiv logo
    arxiv.org
    LM-Infinite: Zero-Shot Extreme Length Generalization for Large...
    Today's large language models (LLMs) typically train on short text segments (e.g., <4K tokens) due to the quadratic complexity of their Transformer architectures. As a result, their performance...
    11K
  • user avatar
    Chi Han
    @Glaciohound
    Mar 19, 2025
    New Benchmark for LLM Multi-Turn-Instruct Following!🚀 Can LLMs handle multiple entangled instructions effectively? Our comprehensive Multi-Turn-Instruct dataset evaluates LLMs’ ability to 📜track, 🔄reason on, and ⚖️ resolve conflicts across multiple turns of instructions in
    12K
  • user avatar
    Chi Han
    @Glaciohound
    May 20, 2024
    Excited about LM-Steer accepted at ACL2024! It offered surprising insights: 1. What is the role of word embedding in LMs? They are generation steers! 2. A simple linear transformation enables flexible, transparent & compositional control! Find out more: arxiv.org/pdf/2305.12798
    Image
    Image
    Image
    Image
    21K
  • user avatar
    Chi Han
    @Glaciohound
    May 25, 2023
    Sharing our new work on explaining in-context learning (ICL) as kernel regression (sth like a weighted kNN): arxiv.org/abs/2305.12766 We provide both theoretical and empirical evidence, and give insights on many phenomena observed in ICL practice!
    arXiv logo
    arxiv.org
    Understanding Emergent In-Context Learning from a Kernel...
    Large language models (LLMs) have initiated a paradigm shift in transfer learning. In contrast to the classic pretraining-then-finetuning procedure, in order to use LLMs for downstream prediction...
    4.1K
  • user avatar
    Chi Han
    @Glaciohound
    Jun 20, 2024
    🎖 Excited to receive an outstanding paper award at NAACL2024 for LM-Infinite "Zero-Shot Extreme Length Generalization for Large Language Models" work! We extend to 200M length with no parameter updates, with downstream improvements arxiv.org/abs/2308.16137 github.com/Glaciohound/LM…
    You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
    14K
  • user avatar
    Chi Han
    @Glaciohound
    Oct 7, 2024
    🎮 Check out the live demo for LM-Steer “Word Embeddings Are Steers for LMs” (Outstanding Paper ACL 2024) at huggingface.co/spaces/Glacioh…, which can: 1.🕹️Steer model generation 2.🔬Discover word embedding dimensions 3.📊Profile sentences & identify keywords #ACL2024 #LLMs #Science4LM
    Image
    Image
    Image
    Image
    Image
    Image
    user avatar
    Chi Han
    @Glaciohound
    Aug 14, 2024
    🎖Excited that "LM-Steer: Word Embeddings Are Steers for Language Models" became my another 1st-authored Outstanding Paper #ACL2024 (besides LM-Infinite) We revealed steering roles of word embeddings for continuous, compositional, efficient, interpretable& transferrable control!
    8.9K
  • user avatar
    Chi Han
    @Glaciohound
    Feb 26, 2025
    Welcome to my #AAAI2025 Tutorial, "The Quest for A Science of LMs," today! Time: Feb 26, 2pm-3:45pm Location: Room 113A, Pennsylvania Convention Center Website: glaciohound.github.io/Science-of-LLM… Underline: underline.io/events/487/sch…
    glaciohound.github.io
    AAAI 2025 Tutorial: The Quest for A Science of Language Models
    3.5K
  • user avatar
    Chi Han
    @Glaciohound
    Mar 19, 2025
    Thanks, @hengjinlp, for sharing our recent work! In this paper, we investigate how LLMs achieve positional flexibility, such as understanding shuffled sentences and reading long context (a phenomenon also observed in humans). This paper uncovers the hidden computational
    You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
    9.8K
  • user avatar
    Chi Han
    @Glaciohound
    Jan 1, 2025
    Thank you so much, Heng @hengjinlp! I'm incredibly grateful for your guidance and support throughout this journey and honored to receive the fellowship to continue working on exciting projects. Looking forward to more collaborations with you and Avi Sil @aviaviavi__! 🙌
    You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
    6.4K
  • user avatar
    Chi Han
    @Glaciohound
    Jun 20, 2024
    Replying to @Glaciohound
    ❤ Thanks to my advisor @hengjinlp for the supports along the way, and to all authors (@sinongwang, Qifan Wang, @haopeng_nlp, @XiongWenhan, @chenyu_hugo)! I will be on job market next year, and always excited about deep researches on LLM understanding, adaptation and improvement!
    1.5K
  • user avatar
    Chi Han
    @Glaciohound
    May 20, 2024
    Replying to @Glaciohound
    Many thanks to my wonderful advisor @hengjinlp and lab @uiuc_nlp for their support. And also a big thank you to all my coauthors!
    729
  • user avatar
    Chi Han
    @Glaciohound
    Oct 9, 2023
    Replying to @Glaciohound
    2️⃣ Last week, @Guangxuan_Xiao’s “StreamingLLM” paper (arxiv.org/abs/2309.17453) offers empirical evidence for a similar approach (worded “attention sinks” + “within-cache position”) with alike results. They also added depth with ablation studies and pre-training!
    397
  • user avatar
    Chi Han
    @Glaciohound
    Oct 9, 2023
    Replying to @Glaciohound
    🔍 Both implementations bring unique value. StreamingLLM optimizes for simplicity, while for encoding or forwarding long docs, it forwards token-by-token. LM-Infinite implements a special sequence forwarding operation. Combining them is ideal for users with different resources.🤝
    373
Advertisement
Advertisement