Edan Meyer Courses

Steps to AGI

1 rating

An overview of the Alberta Plan’s 12-step research roadmap toward AGI, centered on continual learning, learned world models, and intelligence amplification.

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

Learning From Passive Data Explained

Explore reinforcement learning algorithm using passive data to learn without direct interaction, based on successor features. Covers offline RL, algorithm details, results, and critical analysis.

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

RL Foundation Models Are Coming!

How AdA combines meta-learning, memory, distillation, and auto-curriculum learning to scale reinforcement learning toward foundation models.

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

Agent Learns to Do Reinforcement Learning

A research walkthrough of how a Transformer learns reinforcement-learning algorithms from histories of prior learning episodes.

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

Stable Diffusion - What, Why, How?

Explore Stable Diffusion’s mechanics and code, extending a starter pipeline to create diffusion videos, similar images, and image-to-image transformations.

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

This AI Learns from YouTube

An overview of Video Pre-Training, showing how unlabeled Minecraft videos, behavior cloning, and reinforcement learning teach an AI agent to play.

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

Chinchilla Explained - Compute-Optimal Massive Language Models

A walkthrough of DeepMind’s Chinchilla paper and its compute-optimal scaling laws for choosing language-model size and training data under a fixed budget.

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

Learning Forever, Backprop Is Insufficient

Explores why backpropagation loses learning ability over time and how Continual Backprop uses persistent randomness to restore adaptation.

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

Learning Fast with No Goals - VISR Explained

A walkthrough of VISR, combining successor features and diverse policies for rapid adaptation to new reinforcement-learning tasks.

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

AlphaCode Explained - AI Code Generation

An analysis of AlphaCode’s language-model approach to generating competitive-programming solutions, including pretraining, sampling, ensembles, and limitations.

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

Reinforcement Learning in RTS Games

Examines an ablation study of deep reinforcement learning agents in real-time strategy games, including action-space design, invalid-action masking, and reward shaping.

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

Can OpenAI Codex Compete in an ML Competition?

A hands-on test of OpenAI Codex on a Kaggle ventilator-pressure competition, from generating an LSTM model to submitting predictions.

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

A New Theory for How the Brain Learns

An accessible paper walkthrough on burst-dependent synaptic plasticity and biologically plausible learning in hierarchical neural circuits.

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

Trillion Parameter Models Are Here

How ZeRO-Infinity uses memory offloading and parallelism to make trillion-parameter model training feasible on modest GPU clusters.

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

Recreate Google Translate - Model Training

Train and test a multilingual neural machine translation model with mT5, tune its optimization settings, and use it to translate text across languages.

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