- AI
- LLM
- MultimodalAI
- Innovation
- OpenSource
- MachineLearning
- Research
•
•
•
•
•
•
-
Tiny Recursive Model ; Small, Simple… and Surprisingly Strong 🤖🧩
The 7M-parameter Tiny Recursive Model (TRM) outperforms larger networks on reasoning puzzles like Sudoku, Maze, and ARC-AGI through clever recursion.
-
GLM-4.5 ; A Unified Open-Source Powerhouse for Agents, Reasoning & Coding 🤖✨
GLM-4.5 from Zhipu AI & Tsinghua unifies agents, reasoning, and coding in a massively efficient Mixture-of-Experts model—advancing open-source AI.
-
AlphaGo Moment for Model Architecture Discovery ; The Rise of Autonomous AI Scientists 🤖🚀
An autonomous, agentic AI system discovers novel architectures and accelerates scientific progress—breaking the human creativity bottleneck.
-
Reinforcement pre-training - baking the cherry into the cake
exploring microsoft research's revolutionary approach to training language models with reinforcement learning from the ground up
-
Group Sequence Policy Optimization (GSPO); A Smarter Approach to RL for LLMs and MoE Models
Review and notes on the GSPO paper from the Qwen team at Alibaba — improving reinforcement learning stability in LLMs, especially for Mixture-of-Experts architectures.