Zae Myung Kim

🚨 New Paper Alert! 🚨 How can we align language models without drowning in prompt engineering or falling into reward hacking traps? We introduce Meta Policy Optimization (MPO)—a new reinforcement learning framework that evolves its own reward model rubrics through meta-level reflection. Inspired by metacognition and evaluative thinking, MPO trains models to think about how they evaluate, not just what they generate. 🔥 Why it matters: ✔️ Boosts stability and robustness in RLAIF ✔️ Reduces human labor in prompt crafting ✔️ Generalizes across tasks: essays, summarization, ethical and mathematical reasoning Check it out: huggingface.co/papers/2504.20… Big thanks to co-authors @chanwoopark20 (MIT), @_vipulraheja (Grammarly), and @dongyeopkang (UMN)! #AI #LLMs #ReinforcementLearning #MetaLearning #NLP #Alignment #RLHF #RLAIF #EvaluativeThinking #PromptEngineering

@chanwoopark20 What's your impression on coding?

codex 5.3 is just super wild -- very different with codex 5.2...

This work is done with my amazing collaborators Kwang Hee Lee, Preston Zhu, @_vipulraheja, and @dongyeopkang If you're interested, you can read all about it here!: arxiv.org/pdf/2402.10586… Data and code will be released soon!

Our empirical findings indicate that robustness against paraphrasing attacks arises from the preservation of higher-level discourse structures, despite variations at the sentence level.

"Paraphrasing attacks" can compromise the effectiveness of AI content detectors. 🙀 Can hierarchical structures in texts help build a more robust detector? Our research reveals a resounding💡YES!💡Delighted to share our work on merging discourse frameworks with graph analysis.

🚀Excited to share MinnesotaNLP's FIRST lab-wide paper (15+ team) on artifacts present in LLM-generated data! We explore the diverse world of LLM-generated text content and its impact on the artificial data ecosystem. #NLProc #syntheticdata #LLM ArXiV: arxiv.org/abs/2401.14698

LLMs have proven to outperform humans on a multitude of tasks. Does this also mean they are more biased too? In our work, we benchmark several different LLMs as automatic evaluators for various cognitive biases. arxiv.org/abs/2309.17012

I deeply appreciate my amazing collaborators for their efforts and guidance: @WanyuDu @_vipulraheja @ddhruvkr @dongyeopkang 🙏 and ACK to @Grammarly @UVA_ILP @UMNComputerSci!

We also augmented the training dataset with datasets from other relevant tasks, for example, "Lang-8" for improving fluency. We found out that many of these data were "meaning-changed" edits that were more toward generation than revision, and thus we filtered them accordingly.

We are excited to present our recent improvements on the iterative text revision task at #EMNLP2022! Come check it out in Poster Session 2 at 11:00AM on Dec. 9, 2022. Paper: arxiv.org/abs/2212.01350 Code: github.com/vipulraheja/it… #EMNLP2022 #nlpproc

@dongyeopkang 🙀

@zaemyung is teaching us how to make a Korean bomb shot (soju+beer)

Our Minnesota NLP group ia getting bigger and diverse! ❤️❤️

If you are interested in human-machine interactive text revision, please check out our paper, which won 🎉the best paper award🎉 at #In2Writing workshop at #ACL2022! We attach more details below. Paper: arxiv.org/abs/2204.03685 Demo: youtube.com/watch?v=lK08tI…

Please consider submitting your work to #in2writing workshop at ACL!

Excited to share DCQA (Discourse Comprehension by Question Answering), a scalable data collection framework + 22K training data capturing semantic and discursive relationships between sentences via free-form questions and their answers arxiv.org/abs/2111.00701 (1/2)