Discover AI Courses

Two-Phase Pretraining - Enhancing LLM Accuracy and Scalability

Discover cutting-edge techniques for enhancing Large Language Model training through a two-phase approach, focusing on core functionalities to improve accuracy and scalability.

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

Byte Latent Transformer: Token-Free Architecture Using Entropy-Based Byte Prediction

Dive into the innovative Byte Latent Transformer architecture, exploring entropy-based byte prediction, causal local attention mechanisms, and cross-attention for latent patches in tokenizer-free models.

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

Code Chain of Thought with Self-Evolution LLM: rStar-Math Mathematical Reasoning Framework

Delve into Microsoft's rStar-Math framework, exploring how small language models can achieve advanced mathematical reasoning through self-evolution and Monte Carlo Tree Search techniques.

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

ACE Agentic Context Engineering - Self-Learning AI

Explore Stanford's breakthrough ACE framework combining Early Experience with strategic context engineering for autonomous AI self-improvement and continuous learning.

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

AgentFlow - In-the-Flow Agentic System Optimization for Effective Planning and Tool Use

Discover how Stanford's AgentFlow enables a 7B parameter agent to outperform 200B LLMs through innovative agentic system optimization for planning and tool use.

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

First Level Complexity of an AI Agent Explained

Discover the fundamental complexity layers of AI agents and explore Gödel's logic loops that define intelligence limitations in current systems.

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

Agent Builder with ChatKit - West Coast Disaster

Explore OpenAI's Agent Builder and ChatKit for multi-agent system design through a West Coast disaster scenario using GPT-5 PRO intelligence.

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

Neurosymbolic 80M AI from Princeton Beats GPT - GraphMERT Framework for Knowledge Graph Distillation

Explore Princeton's GraphMERT framework that combines neural learning with symbolic reasoning to create reliable knowledge graphs, outperforming GPT with superior accuracy.

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

Base LLM Can Reason - Activation Switch Found

Discover how base LLMs already possess dormant reasoning abilities - backtracking, verification, and computation - that can be activated without additional training.

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

STOP Agent - Teaching LLM Agents to Know When They Know Enough with CaRT

Discover how to teach AI agents when to stop thinking and make decisions using CaRT's groundbreaking counterfactual reasoning method for optimal timing in autonomous systems.

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

Everything That Is Wrong with AI Today - 10 Papers

Uncover critical flaws in current AI systems through 10 cutting-edge research papers, exploring inconsistent reasoning, shutdown resistance, and new methods for RL, RAG, and knowledge graphs.

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

MIT Invents Neuro-Symbolic LLM Fusion

Discover MIT's breakthrough neuro-symbolic approach that teaches LLMs logical chain-of-thought reasoning for enhanced symbolic planning and decision-making capabilities.

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

NEW AI Coding Agents for Software Engineering - ENTROPY

Explore cutting-edge AI coding agents using entropy regularization and multi-turn preference optimization to enhance software engineering performance and efficiency.

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

Bottom-up Domain-specific AI Superintelligence - A Reliable Knowledge Graph Approach

Discover Princeton's breakthrough approach to AI superintelligence using symbolic logic and knowledge graphs instead of traditional token prediction methods.

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

Qwen3-2507 vs Kimi K2 - Independent Benchmark Testing of Non-Reasoning Models

Discover how Qwen3-2507 and Kimi K2 non-reasoning models perform on reasoning benchmarks through live testing and independent evaluation of AI model capabilities.

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