Category: Machine Learning

• The More Reasoning, The More Hallucinations? The "Hallucination Paradox" of Multimodal Reasoning Models
• Unveiling the "Thought Secrets" of Large Reasoning Models: Understanding Their "Aha! Moments" from a "Reasoning Graph" Perspective
• Think While Drawing! Multimodal Reasoning Achieves Significant Improvement!
• Sacrificing Sleep for a Blog Post Lands OpenAI Offer! Muon Author Angrily Reveals: "Almost All Optimizer Papers Are Fake"
• Midjourney Enters Video Generation, Image Model V7 Continually Updated, Visual King Confirmed
• ByteDance Seed's New Work DeltaFormer: An Attempt at Next-Generation Model Architecture
• LLMs Can Now Self-Update Weights, Significantly Boosting Adaptive and Knowledge Integration Capabilities. Is AI Waking Up?
• Kaiming He's New Work: Adding Regularization to Diffusion Models for Performance Improvement with No Pre-training or Data Augmentation, Simple to Implement
• Breaking! Meta Open-Sources Its Latest World Model
• SLOT: Sample-Specific Inference Optimization Tool Arrives, Boosting Accuracy by 10% Without SFT or RL
• After ZeroSearch, Tongyi's Latest Work MaskSearch Proposes a New Framework for Reasoning-Search Pre-training
• 35% Accuracy Evaporates! ByteDance & HUST's WildDoc Reveals Robustness Shortcomings in Multimodal Document Understanding
• Google Research Finds: Prompt Design is the Core of Multi-Agent Systems!
• The Sky Has Fallen! Apple Just Proved: DeepSeek, o3, Claude and Other "Reasoning" Models Lack True Reasoning Ability
• R1-like Training No Longer Just Focuses on Result Correctness! CUHK Launches SophiaVL-R1 Model
• Agent Zero: An Open-Source, Free, Evolving, and Learning Agent
• DeepMind's Latest Research: Agents Are World Models!
• Google Open-Sources Gemini-Level AI Research Capabilities: Is Deep Research Becoming Commoditized?
• Reviewing the Progress of RL-Reasoning
• OPA-DPO: An Efficient Solution for the Hallucination Problem in Multimodal Large Models
• AI Learns Reasoning Solely by "Confidence": Zhejiang University Alumnus Replicates DeepSeek's Long Chain-of-Thought Emergence, Reinforcement Learning Needs No External Reward Signals
• No Manual Annotation Needed! AI Self-Generates Training Data, Unlocking Reasoning Capabilities via "Deduction-Induction-Abduction"
• Sakana AI's New Research: The Birth of the Darwin-Gödel Machine with Self-Encoding Improvement and Self-Referential Open-Ended Evolution
• LLM + RL Questioned: Deliberately Using Incorrect Rewards Still Significantly Boosts Math Benchmarks, Causing a Stir in the AI Community
• Alibaba Open-Sources New Qwen Model: A Dragon Boat Festival Gift!