Discover AI Courses

Improved DeepSeek R1-32B & R1-14B: NEW Light-R1

Discover the latest optimization improvements for DeepSeek R1 models with Light-R1 versions for 32B and 14B sizes, outperforming their predecessors with enhanced reasoning capabilities and performance.

• YouTube
• 26 minutes
• Self-Paced
• Free Video

Topological AI for Reasoning - SOLAR by CMU

Discover how topological AI reasoning models like SOLAR can enhance performance through Graph-of-Thought complexities instead of traditional linear Chain-of-Thoughts approaches.

• YouTube
• 27 minutes
• Self-Paced
• Free Video

Length Controlled Policy Optimization for Scaling Reinforcement Learning - CMU Research

Explore Length Controlled Policy Optimization (LCPO), a reinforcement learning method that trains language models to produce accurate outputs while adhering to specified length constraints, as demonstrated by CMU's L1 model.

• YouTube
• 30 minutes
• Self-Paced
• Free Video

Meta Reinforcement Fine-Tuning - Optimizing Test-Time Compute vs GRPO

Discover how Meta Reinforcement Fine-Tuning (MRT) optimizes AI models through continuous evaluation of reasoning steps, balancing exploration and exploitation for more efficient decision-making compared to traditional approaches.

• YouTube
• 24 minutes
• Self-Paced
• Free Video

Guided Thought Reinforcement to Prevent Thought Collapse in AI Robotics

Explore how Guided Thought Reinforcement (GTR) prevents thought collapse in AI robotics by supervising reasoning processes during RL training, creating more coherent and interpretable AI agents.

• YouTube
• 20 minutes
• Self-Paced
• Free Video

Chain-of-Thought to Sketch-of-Thought: Redefining Abstract AI Reasoning

Discover how Sketch-of-Thought (SoT) offers a more efficient AI reasoning paradigm than Chain-of-Thought, exploring abstract logic capabilities and associative abstractions in AI systems.

• YouTube
• 20 minutes
• Self-Paced
• Free Video

AI 2.0: The New Core of AI is General Reasoning

Discover how the next generation of AI is being designed with optimizations and new ideas from current LLM systems, building on advances in reasoning models.

• YouTube
• 32 minutes
• Self-Paced
• Free Video

Scaling RL: 3B AI with Long Chain-of-Thought and 4 Patterns

Explore how structured cognitive behaviors and extended chain-of-thought reasoning via Scaling RL can build AI systems that solve complex problems while explaining their thought processes clearly.

• YouTube
• 34 minutes
• Self-Paced
• Free Video

Beyond CoT: ReasonFLUX - Hierarchical LLM Reasoning via Scaling Thought Templates

Explore ReasonFlux, a hierarchical LLM reasoning approach that builds on Chain-of-Thought by using flexible, retrievable reasoning templates to optimize complex problem-solving strategies.

• YouTube
• 22 minutes
• Self-Paced
• Free Video

NVIDIA World Foundation Models and Physical AI: REASON 1 Explained

Explore NVIDIA's World Foundation Models and Physical AI technologies, focusing on Cosmos-Reason1 for embodied reasoning and physical common sense applications.

• YouTube
• 33 minutes
• Self-Paced
• Free Video

NVIDIA: New AI Model N1 Explained in Detail

Delve into the technical details, mathematical mappings, and architecture of NVIDIA's N1 AI model designed for generalist humanoid robots.

• YouTube
• 33 minutes
• Self-Paced
• Free Video

In a Network of AI Agents: Pure CHAOS - Why Multi-Agent LLM Systems Fail

Explore the paradoxical world of Multi-Agent Systems where adding more AI agents often leads to worse outcomes, examining research on why collaborative AI systems fail despite individual intelligence.

• YouTube
• 25 minutes
• Self-Paced
• Free Video

DocETL: AI Agents for Complex Document Processing and Data Transformation

Discover how DocETL framework leverages AI agents and LLMs to transform complex document processing, featuring innovative operators and optimization techniques for enhanced data extraction and analysis.

• YouTube
• 30 minutes
• Self-Paced
• Free Video

Collaborative Multi-Agent LLMs for Mixed-Autonomy Traffic - CoMAL Framework

Explore how CoMAL framework enables autonomous vehicles to collaborate through LLM-powered reasoning, role assignment, and real-time coordination for safer and more efficient mixed-autonomy traffic systems.

• YouTube
• 25 minutes
• Self-Paced
• Free Video

Task-Optimized Multi-Agent Communication Protocols with G-Designer

Explore the G-Designer framework for optimizing multi-agent AI communication through Graph Neural Networks, focusing on efficiency, scalability, and robust topology design.

• YouTube
• 49 minutes
• Self-Paced
• Free Video