Discover AI Courses

LLM Knows The Future - Uncovering Multi-Token Prediction Potential

Discover how Apple researchers enable language models to predict multiple tokens simultaneously, breaking sequential constraints to dramatically improve inference speed and parallelism.

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

True AI Reasoning - Graph-Based CPT

Discover GraphPile, a breakthrough corpus for training LLMs with graph problem reasoning, achieving up to 21.2% improvement in logical and commonsense reasoning tasks.

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

The End of RL - GEPA: New Genetic AI Algorithm from MIT and UC Berkeley

Discover how GEPA, a revolutionary genetic AI algorithm from MIT and UC Berkeley, outperforms traditional reinforcement learning methods in prompt optimization and AI agent development.

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

Scaling GraphRAG to Planetary-wide Knowledge Graph

Discover how to scale GraphRAG systems to handle planetary-wide knowledge graphs, exploring advanced techniques for processing millions of documents and multi-path fusion approaches.

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

VLMs Are Almost Blind - Visual Reasoning in Vision-Language Models

Discover how Vision Language Models struggle with true visual reasoning, relying on semantic patterns rather than genuine visual understanding based on Princeton and Harvard research.

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

256K Context Window? Forget It! - Understanding the Limitations of Long Context Reasoning in Large Language Models

Discover why 256K context windows fail at reasoning tasks and explore new benchmarks revealing LLM limitations in long-context performance beyond simple retrieval.

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

AI Thinking, Fast and Slow - Intelligence is Not the Key

Explore how AI systems can decouple knowledge from reasoning using cognitive dual-system theory, challenging traditional views on intelligence in large language models.

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

Can We Trust Deep Research? New Data from ByteDance

Explore ByteDance's evaluation of deep research agents and LLMs with web tools, comparing their trustworthiness against OpenAI and Google's research modes through academic survey tasks.

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

Dream to Chat - Model-based Reinforcement Learning on Dialogues with User Belief Modeling and LSD-3D - Large-Scale 3D Driving Scene Generation with Geometry Grounding

Discover cutting-edge AI research from Geely and Mercedes-Benz on dialogue systems with user belief modeling and large-scale 3D driving scene generation for autonomous vehicles.

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

DeepScholar AI - Automated Scientific Literature Review with Multi-Agent Systems

Discover how Stanford and UC Berkeley's DeepScholar AI automates scientific literature reviews using multi-agent systems and re-ranking RAG for enhanced research synthesis.

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

CORE of AI is EXPLODING - 10 New Papers

Discover 10 groundbreaking AI research papers from ArXiv's September 2025 collection, exploring cutting-edge developments that are reshaping artificial intelligence's core foundations.

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

AI Agents Evolve - New Topology for Multi-Agents

Explore revolutionary multi-agent systems using semantic-topological evolution algorithms that model agentic workflows as self-organized graphs with textual gradients.

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

RAG is BACK - NEW MSRS by Yale University

Discover Yale's MSRS framework for multi-source retrieval and synthesis, advancing RAG systems to integrate information across distinct sources and generate comprehensive long-form responses.

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

Emergent Communicative AI World Models - Multi-Agent

Discover how CORAL framework enables AI agents to develop emergent communication protocols, dramatically improving sample efficiency and zero-shot generalization in multi-agent reinforcement learning.

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

Why GPT-5 Fails with Complex Tasks - Simple Explanation

Discover why GPT-5 struggles with complex tasks through research from Harvard, MIT, and Carnegie Mellon on RCR-Router and Logic-Augmented Generation solutions.

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