The Semantic Structure of LLM Feature Spaces Turns Out to Be Highly Aligned with Human Psychology
A new study reveals that the geometric relationships of semantic features in the hidden states of large language models …
🏷️ AR
3363 articles about 'AR'
Safe but Useless? New Benchmark Exposes the LLM Alignment Dilemma
A research team has introduced CarryOnBench, the first benchmark to systematically evaluate whether large language model…
Exploring the Limits of Pruning: Task-Specific Neurons and Model Collapse
A new study systematically explores the limits of neuron pruning in large language models, revealing for the first time …
Study Reveals Claude's Cross-Language Response Consistency Performance
A new study based on the ILR (Interagency Language Roundtable) proficiency scale framework systematically evaluates Clau…
CL-bench Life: Can Language Models Learn from Real-Life Scenarios?
A new study introduces CL-bench Life, a benchmark that systematically evaluates the ability of large language models to …
A-Share Annual Cash Dividends Hit Record High of 2.42 Trillion Yuan
As the 2025 A-share annual report disclosure season concludes, over 3,700 listed companies have announced cash dividend …
2026 May Day Box Office Surpasses 100 Million Yuan as Three Films Lead the Pack
According to online platform data, the total box office (including presales) for China's 2026 May Day holiday period (Ma…
Fréchet Mean Redefines the Geometry of Model Merging
A new study proposes a universal framework for model merging based on the Fréchet mean, addressing the fragility of trad…
New Research Proposes Distributional Alignment Games Framework to Optimize Answer-Level Fine-Tuning for Large Language Models
A latest arXiv paper proposes the "Distributional Alignment Games" framework, leveraging game theory to tackle the compu…
New Method Enables Few-Step Alignment Guidance for Generative Models
A latest arXiv paper proposes Flow Map Reward Guidance, a method that reframes reward guidance for generative models as …
Study Reveals Learning Rate Transfer Flaws in nGPT and Proposes Systematic Fixes
A latest arXiv paper finds that while the Normalized Transformer (nGPT) delivers impressive training speedups, it fails …
Co-Evolving Policy Distillation: Cracking the Multi-Capability Fusion Challenge
A new study proposes the Co-Evolving Policy Distillation (CoPD) framework, offering a unified analysis of capability los…