\( \newcommand{\vect}[1]{\mathbf{#1}} \newcommand{\mat}[1]{\mathbf{#1}} \newcommand{\R}{\mathbb{R}} \newcommand{\E}{\mathbb{E}} \newcommand{\Prob}{\mathbb{P}} \newcommand{\T}{^{\top}} \newcommand{\norm}[1]{\lVert #1 \rVert} \newcommand{\abs}[1]{\lvert #1 \rvert} \newcommand{\ihat}{\mkern1mu\hat{\boldsymbol{\imath}}\mkern2mu} \newcommand{\jhat}{\mkern1mu\hat{\boldsymbol{\jmath}}\mkern2mu} \newcommand{\pd}[2]{\frac{\partial #1}{\partial #2}} \newcommand{\argmax}{\operatorname*{arg\,max}} \newcommand{\argmin}{\operatorname*{arg\,min}} \newcommand{\cdata}[1]{\class{c-data}{#1}} \newcommand{\cmodel}[1]{\class{c-model}{#1}} \newcommand{\ctruth}[1]{\class{c-truth}{#1}} \newcommand{\cgrad}[1]{\class{c-grad}{#1}} \newcommand{\closs}[1]{\class{c-loss}{#1}} \newcommand{\cparam}[1]{\class{c-param}{#1}} \newcommand{\csecond}[1]{\class{c-second}{#1}} \renewcommand{\,}{\mkern3mu} \)

Changelog

Every release and every correction to Applied AI Mastery, newest first.

Every release and every correction is recorded here, newest first. Lessons are named by their codes.

0.7.0 · 14 November 2026

Lessons

Math Foundations is complete: M01.3 Functions and graphs and M01.9 Complex numbers join the seven modules already live.
Calculus I is complete: M04.2 to M04.4 and M04.6 to M04.10 join M04.1 and M04.5, from the derivative and the chain rule to Taylor approximation, first differential equations and a first look at partial derivatives.

Every one of these modules passed an independent second check before it was published.

Site and tooling

Lessons written from now on carry at least five figures each.

0.6.0 · 14 November 2026

Lessons

Math Foundations opens with seven modules: M01.1 Numbers, algebra and notation, M01.2 Coordinates and a first look at vectors, M01.4 Exponentials and logarithms, M01.5 Trigonometry and the unit circle, M01.6 Sets, logic and proof techniques, M01.7 Sums, products and sequences, and M01.8 Counting and combinatorics.
Calculus I opens with two modules: M04.1 Limits and continuity and M04.5 Integrals as accumulated area.

Every one of these modules passed an independent second check before it was published.

Site and tooling

The start page offers Math Foundations as the place to begin for anyone whose school maths is rusty.
The page check opens four pages at a time, so it stays quick as the site grows.

0.5.0 · 12 November 2026

Lessons

Machine Learning Foundations is complete. New: ML01.3 k-nearest neighbors, ML01.6 Logistic regression and classification, ML01.7 Evaluation metrics, ML01.8 Overfitting, bias and variance, ML01.9 Regularization and validation, and ML01.10 A full project from raw data to report.

Corrections, from an independent second check of ML01.4

ML01.4.1: the mean absolute error is lowest where neither side holds more than half of the targets; the two sides can never balance exactly when the number of targets is odd. The solution of exercise 3 now measures how far the median moves correctly.
ML01.4.3: how far the weights computed through the inverse are off depends on the computer, and the lesson now says so. “An exact copy of another column” is now “an exact multiple”.
ML01.4.4: a model exactly as good as the mean has R² = 0, and a worse one has a negative R². The interactive no longer calls a line “too steep” when the best slope is negative.
All four lessons: equations and tables fit a phone screen.

Site and tooling

A new logo: a wave-shaped surface plotted in three dimensions. It is also the browser-tab icon.
The page check fails on any displayed equation or table wider than a phone’s column, and it measures only once the maths and the site’s font have loaded.
Maths is typeset once the site’s font has loaded, so a formula has the same width on a first visit as on a later one.
Table cells are narrower on a phone, so that worked tables fit.

0.4.0 · 12 November 2026

Lessons

Linear Algebra I is complete. New: M02.7 Inverses, rank and the four subspaces, and M02.10 Orthogonality, Gram-Schmidt and QR, each four lessons and one interactive.

Corrections, from an independent second check of five more modules

M02.6.1: what NumPy’s solver does when the first column starts with zeros, and why it refuses a singular matrix whatever the target, are now stated correctly. Interactive M02.6-X1: after repeated division, rounding could make a true zero look like a pivot; elimination now always finishes, and the read-out no longer claims a value for an unknown that is still free.
M02.8.2: a minor is the determinant of what is left, not the leftover matrix itself. M02.8.4 now shows why the determinant of a triangular matrix of any size is the product of its diagonal. Interactive M02.8-X1: an equation printed from rounded entries now uses ≈, not =.
M02.9.3: the two matrices of one map agree in two of their entries, so the opening question no longer says they differ in all four. M02.9.4: Axler’s result 7.29 is quoted for what it states.
ML01.2.3: a larger test set does not always make the marks bunch together; the lesson and its interactive now say what really happens. ML01.2.4: two divisions by rounded standard deviations are now exact, and “a validation set removes the leak” is now “keeps it off the test score”.
ML01.1: figures ML01.1.2-F4 and ML01.1.3-F4 are redrawn so that no dot or sign is hidden.
All five modules: equations fit a phone screen, inline formulas no longer break inside a bracket or a product, and the labels of the interactives stay clear of handles and arrows.

Site and tooling

On a phone, the tables of the curriculum, the roadmap and the track pages become labelled cards.
The floating “Back to top” button no longer covers the footer links on a phone.
The lesson check fails on any line that Markdown would turn into a list by accident.
The start page begins with the first module of each course.

0.3.0 · 12 November 2026

Curriculum

Math Foundations gains M01.9 Complex numbers, which later courses on matrices, signals and time series rely on. Coordinates move up to M01.2, so that graphs and the unit circle can use them.
Calculus I gains M04.10 Partial derivatives, a first look, so that a mathematics course teaches partial derivatives before ML01.4 and ML01.5 use them. ML01.4.2 and ML01.5.2 now list it as a prerequisite.
Python from Zero now teaches modules, packages and environments fifth (P01.5), before the modules that import libraries. Exploratory Data Analysis now teaches sampling and bias second (S01.2), before the n − 1 divisor and the kinds of missing data. Lessons that named the old codes now name the new ones.
The plan now has 73 courses and 671 modules.

Corrections, from an independent second check of seven modules

M02.1.1: the proof that two arrows are the same vector exactly when their run and rise agree now covers arrows of length 0. M02.1.3: the pitfall on equal lengths now allows for the zero vector.
M02.2.1: a pitfall said that 2u + v + (1, 1) is not a combination of u and v. It is; the pitfall now makes its point with the rule a u + b v + (1, 1). M02.2.4: the flattened sheet in figure F4 is drawn true to shape.
M02.3.2 now recaps length and distance from M02.1 instead of teaching them again. M02.3.3: the cosine of a vector with itself is checked to within 10⁻¹² of 1, because its last digit can vary from one machine to another.
M02.4.1: parallel lines can land on one and the same line. M02.4.2: Axler’s result 3.71 is in Section 3D, not 3C. M02.4.4: a pitfall about 3 × 2 matrices held only for the lesson’s matrix and now says so.
M02.5.4: the condition for exploding gradients from Pascanu, Mikolov and Bengio is on the absolute value of the largest eigenvalue.
ML01.5.3 now derives the eigenvalue equation it uses. ML01.5.4: the long-run mean of the parameters with shuffled epochs is (1.380, 0.528), from exact arithmetic.
RL01.6.3 and RL01.6.4: “acting greedily on the optimal values is optimal” and the convergence of repeated backups now state the condition γ < 1 that their proofs use. A note in figure RL01.6.4-F5 that was false for the first two sweeps is corrected.
Interactives M02.1-X1, M02.4-X1, M02.5-X1 and RL01.6-X1: labels no longer sit on arrows or handles, and every read-out describes the state on screen.
All 28 lessons of these modules: displayed equations and tables fit a phone screen without scrolling sideways, and no source entry turns into a stray numbered list.

Site and tooling

Inline formulas no longer break across two lines at a thin space.
The label of an arrow in a figure takes a darker or lighter shade of the arrow’s colour where it needs one, so that it reaches a contrast of 4.5 to 1 in both themes.
The notation guide adds complex numbers and τ, a threshold on a probability.
Lesson code runs each lesson in a process of its own, as the site build does, and must run on Python 3.12 and every later version.

0.2.0 · 12 November 2026

Lessons

Linear Algebra I: seven more modules, each four lessons and one interactive. M02.1 Vectors as arrows and as data, M02.2 Linear combinations, span and basis, M02.3 Dot product, norms, angles and projections, M02.5 Matrix multiplication as composition, M02.6 Linear systems and elimination, M02.8 Determinants as volume scaling and M02.9 Change of basis.
Machine Learning Foundations: three more modules. ML01.1 What learning from data means, ML01.2 Data splits and the generalization gap and ML01.4 Linear regression as a learning algorithm.

Corrections

ML01.5.1: the claim that the contour lines are ellipses is now derived, and “the bottom lies between w = 1.5 and w = 2” is argued from the table and the slope separately.
ML01.5.2: the note on backpropagation now cites its source.
ML01.5.3: the statement about a rising loss above the stability limit now names its exception (a start exactly on the valley floor), and “a hundred times smaller” is replaced by the measured figures.

Site and tooling

Each figure is now one small file that serves both the light and the dark theme.
The build stops if any text in a figure overflows its box, runs into the next panel or overlaps other text, and if any page or interactive has text that does not fit at phone width.
Dragging a handle in an interactive no longer scrolls the page on a phone.

0.1.0 · 12 November 2026 · Pilot

The first public release.

Lessons

M02.4 Matrices as linear maps: four lessons (M02.4.1 to M02.4.4) and the interactive M02.4-X1, Matrix playground.
ML01.5 Gradient descent from scratch: four lessons (ML01.5.1 to ML01.5.4) and the interactive ML01.5-X1, Gradient descent explorer.
RL01.6 Bellman equations: four lessons (RL01.6.1 to RL01.6.4) and the interactive RL01.6-X1, Bellman backup gridworld.

All twelve lessons are marked in review.

Curriculum

The full plan is published: 73 courses and 669 modules in ten tracks.
Added A09 Classical AI: Search, Logic and Planning, which closes the gap in logic and knowledge representation.

Site and tooling

Style guide 0.9 and the notation guide.
The shared figure style (style/aimstyle), with light and dark versions of every figure.
Automated checks on every change: numeric tests for derivations, the lesson blueprint, figure regeneration, code cells, colour contrast and internal links.