📑LG AI: "750B 모델, 공짜로 풀어줄게!"

이 호에서 다룬 논문

• LG AI

  K-EXAONE 2.0 Technical Report

  LG AI Research · Mixture-of-Experts, Open-Weight, Long-Context

• AI RESEARCH

  The Personalization Mirage: How LLMs Fabricate User Profiles, and Why Self-Monitoring Misleads

  HKUST · Personalization, Over-Inference, Faithfulness

• AGENTS

  Recursive Synthesis for Long-Horizon Terminal Tasks

  Independent · Synthetic Data, Long-Horizon Agent, Recursive Verification

• MULTIMODAL

  Towards Physics of Multimodal Pretraining: Knowledge Flow, Modality Synergy, Early Unification, and Recipes

  Meta, University of Oxford, New York University · Multimodal Pretraining, Modality Synergy, MoE

• WORLD MODEL

  WorldClaw: Agentic 3D Open-World Generation at Scale

  Tencent · 3D World Generation, Agentic Pipeline, Editable Assets

• WORLD MODEL

  EnvACE: Internalizing Environment Dynamics via World Rehearsal for Agentic Reinforcement Learning

  Huawei, Shanghai Jiao Tong University · Agentic RL, World Rehearsal, Agent World Model

• REASONING

  On-Policy Delta Distillation for Multilingual Math Reasoning

  Naver · On-Policy Distillation, Multilingual Reasoning, Math

• AGENTS

  AgentOPSD: Recursive Self-Distillation for Agentic Reinforcement Learning

  Meituan, Zhejiang University · Credit Assignment, Self-Distillation, Agentic RL

• AGENTS

  FinEvo-Bench: A Longitudinal Benchmark for Self-Evolving Agents in Professional Financial Workflows

  Alibaba, Beihang University · Self-Evolving Agent, Financial Workflow, Benchmark

• GENERATIVE

  Energy-Guided Flow Matching

  JD.com, CASIA, Xi'an Jiaotong University · Flow Matching, Pixel-Space Generation, Coarse-to-Fine