Editorial Policy

What we optimize for

A lesson should leave the reader with usable understanding. Ideas are introduced from first principles, in prerequisite order, with small examples and visible computation. We introduce an abstraction after showing the problem, relationship, or calculation that it organizes.

We prefer a precise explanation over a broad survey. When a statement is an intuition, empirical observation, engineering rule of thumb, or open research question, the lesson should say so.

How the curriculum is planned

The LLM Primer team selects the subjects, required chapters and lessons, prerequisite order, learning goals, examples, projects, and expected depth. These decisions come from study, research, technical experience, and repeated review of the path as a whole. A lesson is judged not only on whether its statements are correct, but also on whether it prepares the reader for what follows.

The value of the publication lies in this selection and sequence, the connection between explanation and practice, and the effort spent checking each page in the context of the complete curriculum.

How tools assist the work

The team uses software tools, including coding agents and language models, during drafting and review. They may help prepare an initial draft from requirements set by the team, check calculations and code, refine an explanation, inspect consistency, find missing steps, suggest exercises or diagrams, and perform an initial critical review. They do not define the syllabus or decide what is true, sufficient, or ready to publish.

An initial draft is only a starting point. Tool output is treated as a proposal to inspect, not as an authority or finished lesson. Before material is marked Open, the editorial team reviews its explanation, prerequisite flow, notation, calculations, code, exercises, diagrams, references, and rendered page. Automated checks help find broken routes, invalid identifiers, code failures, missing metadata, and rendering problems, but they do not replace that review.

Sources and research references

Common Python and mathematical foundations may be taught without a citation for every standard fact. Research-based subjects, beginning with Deep Learning, include primary papers or official project sources when a research contribution, named method, historical claim, or implementation detail benefits from a traceable source.

References should appear near the relevant lesson rather than only in a distant bibliography. We prefer original papers and official project documentation. A citation provides a trail to the source; it does not replace the explanation or imply endorsement of every claim in that source.

Code, diagrams, and exercises

Code should make the idea inspectable and mirror the mathematics where practical. Examples are tested in proportion to their role, but they remain educational examples rather than production guarantees.

Diagrams must teach a relationship, computation, geometry, scale, or failure mode. Interactive figures should explain what to change and what to notice. Exercises are part of the teaching sequence; hints and solutions should explain the reasoning, not merely reveal an answer.

Corrections and revisions

Technical education is revisable. We correct factual, mathematical, code, accessibility, and presentation errors when they are confirmed. Larger revisions are reviewed in the context of neighboring lessons so a local fix does not create a new prerequisite gap.

Signed-in readers can use Report an issue on a lesson page. You can also use the Contact page. Include the page URL, the relevant passage, exercise, code sample, or figure, and a short explanation. Reports are evaluated before a change is made.

Publication status

The curriculum uses three reader-facing states. Open subjects are published for everyone. In review subjects are restricted to approved reviewers and may change substantially. Plannedsubjects describe intended coverage without promising content or a release date. Publication state is stored with the source and changed through repository review, not through a browser editor.

Independence and commercial boundaries

LLM Primer does not currently publish advertisements, sponsored lessons, or affiliate placements. If a financial relationship could affect a reader's judgment, it will be disclosed where it is relevant. Payment does not determine a technical conclusion. Essential explanations and accessibility features are not withheld merely to make the free material harder to use.

Scope of this policy

This policy describes the editorial standard we apply; it is not a promise that every page is error-free. The Terms of Service govern use of the site, and the Privacy Policy explains how information is handled.