Category: Large Language Models

• A New Revolution in Reward Models! SWIFT Reads "Inner Voice" Instead of Text, Creating a Faster, Stronger, and More Cost-Effective AI Judge
• The "Mirage" of Chain-of-Thought Reasoning: An In-depth Look at LLM Generalization
• GPT-5 vs Claude Opus 4.1: Coding Capability Assessment
• In-depth Dissection of Large Models: From DeepSeek-V3 to Kimi K2, Understanding Mainstream LLM Architectures
• ARPO: Agentic Reinforced Policy Optimization, Enabling Agents to Explore One Step Further at Critical Moments
• Open-Sourcing the Largest High-Quality Scientific Reasoning Post-Training Dataset to Quickly Turn Qwen3 and Others into "Scientists"
• Wang Mengdi's Team Review of "Self-Evolving Agents": From Static LLMs to Artificial Superintelligence (ASI)
• Anthropic Team Uncovers 'Persona Variables' to Control Large Language Model Behavior, Cracking the Black Box of AI Madness
• Is Your Model's Attention Drifting? RUC and Tsinghua University Introduce LeaF: Pruning Distracting Tokens for Focused Learning
• Can Models Truly "Reflect on Code"? Beihang University Releases Repository-Level Understanding and Generation Benchmark, Refreshing the LLM Understanding Evaluation Paradigm
• ReaGAN: Empowering Each Node as an Intelligent Reasoning Expert in Graphs
• Google's Challenge: DeepSeek, Kimi and More to Compete in First Large Model Showdown Starting Tomorrow
• RAG Revolution! Graph-R1, the First RL-driven Graph Reasoning Agent
• RAG Can Also Reason! Thoroughly Solving the Multi-Source Heterogeneous Knowledge Challenge
• Beyond Human Annotation: Meta Introduces CoT-Self-Instruct – Reshaping LLM Training with 'Reasoning-Driven Self-Evolution'
• A Deep Dive: Where Does Large Model Training Time Go?
• Revisiting Qwen3's Abandoned Mixed Inference Mode
• DeepSeek R2's Secret Weapon Revealed! The Technology Just Awarded a Top Prize to Liang Wen-feng Allows AI to Read Long Texts 11 Times Faster
• Qwen Updates Overnight: Runs on RTX 3090, 3B Parameters Activated Rival GPT-4o
• Dualformer: Controllable Fast and Slow Thinking by Learning with Randomized Reasoning Traces
• Do Multimodal Large Language Models Truly 'Understand' the World? — Unveiling Core Knowledge Deficits in MLLMs
• Hierarchical Reasoning Model
• DeepSeek-GRPO Importance Weight Design Flaw? Explaining Qwen3's New Reinforcement Learning Algorithm GSPO
• Must-Read: In-depth Comparison of Mainstream LLM Architectures, Covering Llama, Qwen, DeepSeek, and Six Other Models
• Kimi K2's Key Training Technique: QK-Clip!