Blog
Post-Training for Reasoning Is Not Post-Training for Agents
Most new post-training methods are evaluated on math and reasoning benchmarks, not on genuinely agentic tasks. Why we should be more skeptical about claims that they transfer to long-horizon agentic settings.
Welcome to My Blog
Hello! This is my first blog post. I'll be sharing research notes, technical insights, and thoughts on NLP and LLMs here.
Revisiting GraphRAG in the Era of Agents
Personal reflections on GraphRAG's role today — why it still matters for global queries over large knowledge bases, and why search agents may eventually subsume it.