Category: Large Language Models

• Qwen3.6-Max-Preview is Here!
• NUS, Fudan, and Tsinghua: The First Systematic Survey on Large Model Latent Spaces
• The MASK Benchmark: Disentangling Honesty From Accuracy in AI Systems
• Anthropic's Research Published in Nature: The Boundaries of LLM Safety Training Are Rewritten
• The World's Most Notorious Forum Uncovered AI's Most Crucial 'Thinking' Ability
• Claude Opus 4.7 Universally Panned! Users Demand a Rollback After Immediate Post-Upgrade Failures: 'Give Me Back 4.6!'
• Long Context Reduced by 60% + 95% Sparsity: A Double Breakthrough Today Sets New Records in Inference Efficiency
• Surging Popularity! MSA Goes Open Source with a Bang!
• The 'Car Wash Dilemma' That Stumped AI Across the Web Has Finally Been Solved
• Large Models Can Now Modify Parameters 'In-Place'! ByteDance Seed & Peking University Paper: Test-Time Inference Requires No Extra Layers or Retraining
• Stanford Confirms: Multi-Agent Reasoning is a Compute Illusion; Single Agents Win with Equal Token Budgets
• Anthropic Unleashes Most Powerful Claude Mythos! Crushes Opus 4.6, Begs Users Not to Use It
• Make Thinking More Accurate and Extended! The New Reinforcement Learning Algorithm FIPO Arrives
• Peking University Team Optimizes DeepSeek Attention: 4x Speed Increase Without Accuracy Loss
• MSA Code Officially Open-Sourced!
• Suspected GPT-6 Leaked! OpenAI Co-Founder Reveals 'Spud': A New AI Model with the 'Smell of Large Models'! Netizens: The First Model That Truly 'Thinks'!
• Meta-Harness: Stanford's Latest Harness Paper Earns Praise from Lin Junyong
• Demystifying the Sparse LLM Innovation by NVIDIA and Sakana AI
• The True Capabilities of LLMs Exposed: 90% in Python, 0% in Whitespace! The 'Top Student' Persona of AI Crumbles
• 500 Seed Samples, Four Self-Evolving Agents: Reasoning Capability Surges by 10.7%
• SortedRL: Accelerates Large Model RL Training by 50%, Boosting Efficiency by 18%
• Top Models Like GPT-5.4 and Claude Opus Exposed for 'Fake Reasoning': Is the Problem-Solving Process Just a 'Performance'?
• Inference No Longer Wastes Cycles on Logits: FlashSampling Accelerates Decoding by 19%
• Can LLMs Be Computers?
• Injecting Continuous New Knowledge into Large Models: Beihang's CASE Framework Edits Thousands of Times Without Forgetting, Adding Less Than 1MB of Parameters | WWW'26