Sophia S. Han
237 posts
Sophia S. Han
@HanSineng
Joined February 2019
- We’re thrilled to welcome you to New England NLP 2025 at Yale University on April 11th in New Haven, CT 🎉 nenlp.github.io/spr2025/! Join us for a full day of exciting talks and sparkling discussions with NLP researchers across the New England region and beyond. 👉 Register now
- I’ll be visiting Stanford/UCB next week — if you’re working on language model reasoning and see overlap with my research, I’d love to connect and chat! DMs open. sophiahan6.github.io
- Excited to be joining Meta NYC this summer as a Research Scientist Intern! If you’re also in NYC or at Meta and working on reasoning or related topics, I’d love to connect - DM me!
- Can LLM really do logical reasoning even if it generates the correct truth value 👀？ Presenting our follow-up (and no less interesting) work on FOLIO: “P-FOLIO” this morning during the “Resource and Evaluation” poster session 📢📖🗞️! #EMNLP2024 #LLMs P-FOLIO is the first
- Traveling to EMNLP and will be presenting our work “FOLIO: Natural Language Reasoning with First-Order Logic” tomorrow at 4pm during the “Resource and Evaluation” poster session 👩🏫. FOLIO is the first human-annotated, logically and semantically complex and diverse dataset for
- Join us for the Complex Reasoning with LLMs BoF session at #NAACL2025 @naaclmeeting — 11 AM this Friday! Open to all. Excited to dive into new ideas at the intersection of reasoning, language, and learning and reflect on old ones!
- In 2019, Drago helped me edit the first paper I wrote on summarization. Ever since I came to the U.S. for my Ph.D. and embarked on a new research topic, Drago has provided me with unwavering support and guidance, for which I will forever be grateful.The #AI community, the #computerscience community, the @YaleSEAS community, and humanity have suddenly lost a remarkable person, @dragomir_radev - kind and brilliant, devoted to his family and friends... gone too soon. A sad day @Yale @YINSedge @YaleCompsci #NLP2023
- Who wants to build a benchmark for evaluating street smartness while actually learning it ourselves 🙂？Gonna be fun lol.i hate the fact that llms are book smart, but not street smart.
- Had a great time speaking at @GoogleDeepMind’s language team reading group on “From First Principles to Real-World Impacts: Advancing Reasoning from Logic to Law”. Such a thought-provoking discussion with an inspiring group!
- Excited to see more investigation into LLM creativity. We have some pioneering work on this topic as well: Creativity or Brute Force? Using Brainteasers as a Window into the Problem-Solving Abilities of Large Language Models. arxiv.org/pdf/2505.10844.🚨 New study on LLM's reasoning boundary! Can LLMs really think out of the box? We introduce OMEGA—a benchmark probing how they generalize: 🔹 RL boosts accuracy on slightly harder problems with familiar strategies, 🔹 but struggles with creative leaps & strategy composition. 👇
- Besides natural language and formal language, truth table is also a great media for logical reasoning with a synergistic effect. Check out this cool idea from @LichangChen2!Learn to Reason via Mixture-of-Thought Interesting paper to improve LLM reasoning utilizing multiple reasoning modalities: - code - natural language - symbolic (truth-table) representations Cool idea and nice results. My notes below:
- Best thread I have read about neurosymbolic systems for a while!Replying to @AndrewLampinenFor humans, mathematical symbols (and formal systems like lean) are *tools* we learn how to use, not a structure that wraps around us. I think that's the right role for formal still manipulation: a tool that can be employed by an intelligent system if/when it supports a goal.
- After a bunch of interviews with legal practitioners and educators to build a legal domain benchmark, it feels like although certain expert domain is under-studied by LLM experts, it might already be well-studied by the domain experts themselves 😧.
