Projects

Reading presentation archive.

I am grateful to Prof. Hyunwoo Kim for recommending papers and providing valuable feedback on my presentations.

paper review talks

1. Illusion of Thinking and ORCA
   Dec 29, 2025
   Review and discussion on reasoning illusions and ORCA-style approaches.
   • Paper 1: The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity
   • Paper 2: Orca: Progressive Learning from Complex Explanation Traces of GPT-4
2. DeepSeek R1 and DAPO
   Jan 12, 2026
   Technical review on DeepSeek-R1 and DAPO objectives.
   • Paper 1: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
   • Paper 2: DAPO: An Open-Source LLM Reinforcement Learning System at Scale
   • Paper 3: GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization
3. OLMo1 to OLMo3
   Jan 26, 2026
   Architecture and training evolution review across OLMo generations.
   • Paper 1: OLmo: Accelerating the Science of Language Models
   • Paper 2: Olmo3
4. Dive into RL
   Feb 9, 2026
   Reinforcement learning fundamentals and practical insights.
   • Paper 1: ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models
   • Paper 2: Part I: Tricks or Traps? A Deep Dive into RL for LLM Reasoning
   • Paper 3: JustRL: Scaling a 1.5B LLM with a Simple RL Recipe
5. Spurious Rewards and TinyLoRA
   Feb 23, 2026
   Review on Spurious RLVR training signals and TinyLoRA which uses only 13 parameters.
   • Paper 1: Spurious Rewards - Rethinking Training Signals in RLVR
   • Paper 2: Learning toReasonin13Parameters
6. Personas
   Mar 10, 2026
   Review on scaling synthetic data creation with persona-driven prompting and large-scale persona datasets.
   • Paper 1: Scaling Synthetic Data Creation with 1,000,000,000 Personas
   • Dataset: Nemotron-Personas-USA
7. SocialLLM
   Apr 9, 2026
   Survey of LLMs as social simulators, covering human behavior prediction, agentic digital twins, multi-agent world simulation, and privacy-preserving synthetic data.
   • Paper 1: Position: LLM Social Simulations Are a Promising Research Method
   • Paper 2: Finetuning LLMs for Human Behavior Prediction in Social Science
   • Case Study: CVS: The Flight Simulator for Management
   • Project: MiroFish
   • Paper 3: Privasis: Synthesizing the Largest "Public" Private Dataset from Scratch