- AI
- LLM
- MultimodalAI
- Innovation
- OpenSource
- MachineLearning
- Research
•
•
•
•
•
•
-
Native Sparse Attention ; Hardware-Aligned Breakthrough for Long-Context LLMs 🤖✨
DeepSeek’s Native Sparse Attention sets a new bar for efficient, hardware-optimized long-context modeling in LLMs—combining dynamic sparsity with end-to-end trainability.
-
🏅📐 AlphaGeometry2 ; AI Reaching Gold Medal Level in IMO Geometry!
AlphaGeometry2 by Google DeepMind sets a new benchmark in mathematical reasoning—surpassing IMO gold medalists and showcasing next-level symbolic intelligence.
-
EvalPlanner ; Meta’s Transparent & Accurate LLM Evaluation Approach 🌟
Meta’s EvalPlanner trains LLMs to evaluate each other with synthetic data, planning, and transparent reasoning—setting new benchmarks for LLM evaluation accuracy.
-
SFT vs RL ; Generalization Power in Foundation Models 🚀🤖
Google DeepMind’s research reveals how Supervised Fine-Tuning and Reinforcement Learning together shape generalization in foundation models—balancing stability and adaptability.
-
Open Deep Research ; Hugging Face’s Transparent Alternative to OpenAI’s Tool 💥
Open Deep Research from Hugging Face is a transparent, open-source answer to OpenAI’s deep research agent—unlocking autonomous web research for all.