01
David J. Malan (Harvard CS50) · 2024 · YouTube
Why?
Ten live lectures with a graded problem set each: functions, files, and classes by writing Python, not reading about it.
Loading…
One moment while we fetch this page.
From Python and the maths the models rest on, through the courses, to a GPT you trained yourself, reasoning models, and serving at scale.
Start here: the first 7 items in the community’s reading order.
01
David J. Malan (Harvard CS50) · 2024 · YouTube
Why?
Ten live lectures with a graded problem set each: functions, files, and classes by writing Python, not reading about it.
02
Grant Sanderson (3Blue1Brown) · 2016 · YouTube
Why?
Sixteen short videos on vectors, matrices, and eigenvectors: enough to read a model's notation. Follow with his probability series.
03
CampusX (Nitish Singh) · 2021 · YouTube
Why?
134 Hinglish lectures, about 63 hours, deriving each classical model before calling sklearn. Watch at 1.25x and type the notebooks.
04
CampusX (Nitish Singh) · 2022 · YouTube
Why?
84 lectures from perceptron to transformers, backprop derived by hand before Keras appears. Almost nothing repeats the ML playlist.
05
TensorTonic · 2026 · tensortonic.com
Why?
Write a conv, an LSTM cell, and multi-head attention against real tests. Do it beside the DL playlist, not after it.
06
Percy Liang, Tatsunori Hashimoto · 2026 · YouTube
Why?
The whole stack in one course: tokenizer, transformer, Triton kernels, distributed training, scaling laws. Assignments assume real code.
07
Sebastian Raschka · 2024 · Manning
Why?
The gentle version: a GPT-2-sized model on a laptop, every tensor printed. Read it beside CS336, not instead of it.
For
Someone who writes a little Python and wants one unbroken path from the basics of machine learning to serving a model you understand from the weights up.
Covers
Basics of Machine Learning and Deep Learning through to the latest Generative AI topics, from training to evaluation.
Leaves out
Frontier-scale training runs, and reproducing research papers from scratch. Both need GPUs and weeks rather than a reading list; this path gets you to the point where they are worth attempting.