Fengyu Cai

Claude Code for Academics "A gentle introduction in how to use Claude Code for Academics." presentation slides and github repo from Alessandro Spina link in reply

Almost there...

OpenAI's blog (openai.com/index/why-lang…) points out that today’s language models hallucinate because training and evaluation reward guessing instead of admitting uncertainty. This raises a natural question: can we reduce hallucination without hurting utility?🤔 On-policy RL with our Binary Retrieval-Augmented Reward (RAR) can improve factuality (40% reduction in hallucination) while preserving model utility (win rate and accuracy) of fully trained, capable LMs like Qwen3-8B. [1/n]

(1/7) 🧐Can we dynamically select and integrate the best retrievers for each query? We introduce ✨MoR✨: a zero-shot way to handle diverse queries with a weighted combination of heterogeneous retrievers – even including human information sources! We will present this paper at #EMNLP2025 today! 🗓️Welcome to our poster session at Hall C, Wednesday, Nov 5, 16:30 – 18:00 👋 We have more details in our paper! 📄Paper: arxiv.org/pdf/2506.15862 🧑‍💻Code: github.com/Josh1108/Mixtu… Please come to our poster session or reach out if you want to chat more! 🧵

🤖💡 LLMs are great at reasoning! But how confident are they in their answers? Our latest survey dives into the world of LLM confidence estimation and calibration. 📰 arxiv.org/abs/2311.08298 Learn more about our #NAACL2024 paper in this thread 🧵! (1/9) #NLProc #Survey

Our repo has been re-available now: github.com/XuezheMax/mega…

Thought-provoking 🤯 and funny 😂 talk by @YejinChoinka on modeling common sense, arguing for the importance of language and causality, and questioning GTP3's moral reasoning capabilities 🤔 at #NeuroHAI conference. hai.stanford.edu/events/2020-fa…

Really interesting presentation by Yejin Choi, on using language as a substrate for producing intuitive commonsense inferences, and treating this kind of reasoning as a generative problem. @StanfordHAI #NeuroHAI

Our controllable data-to-text generation dataset, ToTTo, has been accepted to #emnlp2020 ! We hope it can serve as a benchmark for high precision text generation. 1/5 Paper: arxiv.org/abs/2004.14373 Dataset (~135K examples): github.com/google-researc… #NLProc

Neural network models have made great advances recently on a wide range of tasks with state-of-the-art results, but these models also require large amounts of annotated data to be trained. Learn about new approaches to address the scarcity of labeled data: aka.ms/AA9sfgj

Slides up #eth2020" target="_blank" rel="nofollow noopener">sweichwald.de/slides.html#et… This was a super fun experience for me coming back to @ETH_en – great to see students intrigued and asking questions much smarter than my talk 🙃 Thanks Stefan Bauer & @bschoelkopf for inviting me – causal representation learning 🔥