RLHF 구조적 결함인 Sycophancy 해결을 위한 검증 Gate 설계
Sycophancy in AI Is the Safety Problem That Looks Like Politeness
Sycophancy in AI Is the Safety Problem That Looks Like Politeness
Open Weight LLM, 2026년 폐쇄형 모델 성능 격차 제로 예측
How Claude AI Actually Works: The Technical Story Behind the Scenes
VibeThinker: A 3B-Parameter Model Just Beat Opus 4.5 on Reasoning — Here is How
GLM-5.2, GPT-5.5 대비 환각률 28% 달성 및 추론 효율성 증명
You don't pick the RL algorithm — SIA's Feedback loop does
AI 중심 재편으로 인한 Meta 엔지니어링 조직의 구조적 붕괴와 보안 리스크
RLAIF Is Eating RLHF — Here Are the Four Places Human Feedback Still Wins
RLHF vs DPO vs IPO vs KTO: which alignment method should you use
The Whole Paper Fits in One Sigmoid: Implementing the SDAR Gate
We Built a 'Grovel Index' to Measure LLM Sycophancy —Here's What We Found
Open Reproduction of DeepSeek-R1
CS336: Language Modeling from Scratch
Understanding Reinforcement Learning with Human Feedback Part 6: How the Reward Model Trains the Original Model
Stop Being Nice, Start Being Right": The Day My User Reconfigured My Reward Function
The Cursor 3 Features Nobody Is Talking About Yet
Understanding Reinforcement Learning with Human Feedback Part 2: Aligning Pretrained Models
Understanding Reinforcement Learning with Human Feedback Part 1: Pre-Training Large Language Models
RLHF trained Claude to be verbose. Here's the proof
The Man Who Summoned Ghosts | Chapter 2: The Training Stack Is Not a Secret