AI, end to end
Twenty-two classes from the perceptron to a production agent. Watch in order; each one collects on the debt the last one left.
22 videos
Watch in order
01How we got here: from the perceptron to the transformer
11:17
02What deep learning actually is: backprop, and why language was hard
10:46
03Neural networks end to end, and the PyTorch that runs them
10:43
04RNNs and LSTMs: the memory that almost worked
10:22
05CNNs: how a machine learned to see
11:18
06Attention from scratch: the mechanism that ate the field
10:46
07Attention in production: KV cache, MQA, GQA and MLA
10:46
08Hugging Face end to end: models, datasets, spaces
11:08
09Instrumenting LLM calls: traces, spans and what to log
11:11
10Vector databases and RAG, end to end
11:07
11Context engineering: the window is a budget
10:34
12Agents from first principles: build the loop yourself
10:10
13Agent frameworks: what they add and what they hide
10:27
14Memory: what an agent should remember, and where
10:40
15MCP: one protocol for every tool
11:00
16Computer use and multimodal agents
10:42
17What fine-tuning actually is: prompting, retrieval, LoRA and the order to try them in
10:31
18Fine-tuning a model for one real use case
10:51
19RL fine-tuning: reward models, RLHF and DPO
10:42
20Evals: how you know an agent is good
10:36
21Mixture of experts, reasoning, and test-time compute
11:03
22Voice, image and video: diffusion and the other modalities
10:35