From AI Foundations to Machine Learning: Learning Paths through CS50 AI, CS181, and CS182

TL;DRCS50 AI is Harvard's most complete public entry point, but the Summer 2026 course still uses 2020 recordings and assignment assets while the rolling OCW projects have moved to other editions. CS181 Spring 2026 exposes current homework and notes without current recordings; CS182 Fall 2026 has not yet completed an offering.

#harvard #ai-course #machine-learning #learning-path #open-course

Series: Global AI/CS Course Map (5 / 8)

Berkeley AI/ML Course Guide: From CS61A to CS288, What Can You Actually Study Online? | Global AI/CS Course Map Connecting Courses by Hung-yi Lee, Hsuan-Tien Lin, and Yun-Nung Chen | Global AI/CS Course Map

🌏 中文版

Video status: Official entry or recording index only. Source details

Search for a public Harvard AI course and CS50’s Introduction to Artificial Intelligence with Python appears first. Seven weeks of video, lecture notes, Python projects, and check50 make it look like a single, current course. The year printed on the course shell, however, is not necessarily the recording year or the assignment edition.

CSCI S-80 Summer 2026 was a formal Harvard Summer School offering, but that does not mean its underlying assets were recorded in 2026. Lecture 0 still points to Spring 2020 recordings, slides, and transcripts, and the enrolled course's Degrees project downloads a Spring 2020 distribution.

The undated CS50 AI OpenCourseWare uses a 2023 distribution for the same project and a 2024 check50 slug. All access judgments below reflect an audit performed on August 22, 2026.

This guide therefore separates the offering, lecture assets, assignment assets, and feedback system before reconstructing Harvard's internal AI/ML route.

Harvard's formal foundation is not CS50 AI

Harvard College calls a major a concentration. The current CS concentration requirements cover programming, formal reasoning, systems, computation and the world, linear algebra, and probability; the honors route also requires an AI-tagged course. A common route is:

CS50/CS32 → CS51/CS61
    + linear algebra + probability + formal reasoning
                         ↓
              CS1810 Machine Learning
              CS1820 Planning and Learning Methods in AI
                         ↓
         CS1840 RL / CS2831 Vision / advanced topics

Harvard CS advising lists CS1810 and CS1820 among the core courses guaranteed at least once per year. They are parallel foundations, not consecutive difficulty levels. CS1810 builds machine learning from probability; CS1820 treats search, planning, games, uncertainty, and learning as a broader AI problem. CS50 AI is a useful implementation-first preview, but it does not replace CS1810's mathematics or non-trivial Python expectations.

Three entry points, three different kinds of course

Course and edition	Editorial grade	What an anonymous learner gets	Main gap
CS50 AI OCW, rolling edition	A3	Seven weeks of video, audio, slides, transcripts, notes, project specs, distributions, and parts of the `check50` workflow	Lecture and project assets span editions; this is not current Harvard College enrollment
CSCI S-80, Summer 2026	A3*	A formal seven-week schedule, 2020 recordings, 2020 project distributions, and public specifications	`*` The material is usable, but Gradescope, sections, quizzes, office hours, and formal feedback require enrollment
CS1810, Spring 2026	A3	Current syllabus, seven homework directories, course notes, and section material	The syllabus says all learning is in person; current complete recordings are not public, while Gradescope and Ed are restricted
CS1820, Fall 2026	A0	Catalog entry, topic scope, instructor, and preview site	The course begins September 2; no public syllabus or delivered material yet
CS182, Fall 2022 archive	A2	Historical lecture and section notes plus exams	Not the current edition; no complete, consistent video/starter/grader chain
CS1840 RL, Fall 2026	A0	Catalog and topic description	The offering is not complete yet
CS2831 Advanced Computer Vision, Fall 2026	A0	Catalog, instructor, and topic description	The current material set is not yet an auditable public course
CS 2881R AI Safety, Fall 2025	A3 (seminar standard)	Reading lists for all 12 lectures, 11 lecture recordings on YouTube, a runnable HW0 GitHub repo, midterm and final specs with rubrics, student project papers and posters	No traditional problem sets; slides for only about half the lectures; no lecture recording for L5, only the opening of L8

The A0–A3 scale comes from the Global AI and CS Course Map. A0 covers catalog, offering, or topic descriptions; A1 requires an accessible syllabus; A2 exposes substantive partial material; and A3 is coherent enough for self-study. These are not Harvard ratings and do not promise credit, instructor feedback, or free compute.

CS50 AI: complete enough to study, not a wholly new 2026 edition

CSCI S-80 Summer 2026 covers Search, Knowledge, Uncertainty, Optimization, Learning, Neural Networks, and Language, each paired with a quiz, section, and project. At the curriculum level, the lectures and assignments follow the same seven-week sequence.

At the file level, editions diverge. Summer 2026 Lecture 0 links directly into cdn.cs50.net/ai/2020/spring/; the enrolled Degrees project also downloads a 2020/spring distribution. The OCW Degrees project downloads 2023/x and runs check50 ai50/projects/2024/x/degrees.

The accurate description is: Summer 2026 is a new formal offering built from classic recordings and a pinned assignment snapshot; OCW is a rolling public edition. The concepts align closely, but learners should not mix OCW starter files with Summer 2026 Gradescope instructions. Use OCW end to end for independent study; use only the linked files for enrolled CSCI S-80 work. This site now has a full deep dive on the course, expanding the version differences and self-study routes there.

The quickest version check is to open every project's download URL and check50 command and record the embedded years. If they differ, treat the assets as conceptually aligned but operationally distinct.

CS1810: current 2026 homework is public; current lectures are not recordings

The CS1810 Spring 2026 syllabus defines a rigorous ML foundation: students derive algorithms mathematically and apply them to real data. Expected preparation includes Python beyond CS50, probability, calculus, and linear algebra. The course explicitly presents Homework 0 as a readiness check.

Its access profile is almost the inverse of CS50 AI. The Spring 2026 homework repository exposes HW0 through HW6, and the course site links notes and sections. The syllabus also states that all learning will be in person. Gradescope handles submission and grades; Ed carries course discussion; solution feedback belongs to the enrolled workflow.

CS1810 therefore earns A3 because its current notes, sections, textbook, and assignments form a coherent route—not because it has a public current-semester playlist. An independent learner should use the homework as the clock: attempt HW0, repair gaps in probability or linear algebra, then read and solve toward each later assignment. An old playlist should not be relabeled as Spring 2026. This site's Harvard CS181 weekly guide follows exactly that clock: one post per Spring 2026 assignment from HW0 through HW6, plus midterm and final review checkpoints.

CS1820: the newest offering is not complete yet

The Harvard SEAS course listing schedules CS1820 for Fall 2026 with Stephanie Gil. It covers search and planning, optimization and games, uncertainty and learning, plus AI ethics and social applications. That is the current official design, but the first class is still eleven days away on this article's audit date.

The catalog and preview site are not enough to judge a semester's recordings, assignments, starter code, or solutions, so the current course remains A0 (Locator 2026 Fall says This Canvas course site has not been published yet; Locator 2025 Spring is a bare listing without a public syllabus or homework repo). Ariel Procaccia's Fall 2022 archive preserves 22 lectures + pset0–4 + section1–12 + midterm/final (60+ PDFs) on problem solving, multi-agent systems, uncertainty, machine learning, and ethics, and the Fall 2018 syllabus (Procaccia / Radanovic / Xu) plus Harvard-CS182-F18 P0–P5 are from the same lineage. They are a historical A2, not a Fall 2026 substitute. 2023/2024 show the same gap (only Embedded EthiCS snapshots; no public pset repo — and Berkeley's CS182 with the same number is a different course).

Series disclaimer: A weekly CS182 guide cannot be promised as A3 for the current 2026/2025 offering (A0). The site will use the Fall 2022 (or Fall 2018) historical A2 as the dated source and publish a separate current-edition comparison once Fall 2026 material exists. The CS181 weekly guide uses the 2026 hw0–6 A3 assignments as its spine (all learning will be in-person, no public recordings, Google Sheet schedule deleted — week = homework number), plus midterm and final checkpoints.

The most stable external route

1. Build working intuition with CS50 AI OCW

This fits learners with roughly a year of Python who want a project-driven overview of search, logic, probability, optimization, ML, neural networks, and language. Stay inside the OCW download and check50 paths instead of crossing into Summer 2026.

2. Make the mathematics real with CS1810 Spring 2026

Attempt HW0 first. Repair only the gaps it exposes, then use the six main homeworks as milestones. Without recordings, notes, section problems, and assignments preserve edition consistency better than an unrelated video playlist.

3. Add broad AI or a specialist branch

For planning, games, and multi-agent systems, start now with the first problem-solving notes in the dated CS182 Fall 2022 archive. CS1840 and CS2831 are future tracking items, not current self-study routes; do not promise their material before a syllabus and assignments exist.

For AI safety, the course usable today is CS 2881R AI Safety, the graduate seminar Boaz Barak first taught in Fall 2025. That offering is complete: reading lists for all 12 lectures are public, YouTube has 11 lecture recordings, HW0 is a GitHub repo you can run yourself, and the midterm and final specs are posted, so by seminar standards it rates A3. The prerequisite is CS 181-level background, so it comes after CS1810. The site lives on Boaz's GitHub Pages; the root now shows Fall 2026, and Fall 2025 moved to /fall2025/. Fall 2026 is in session and serves only as a preview. There are no traditional problem sets, so a self-study rhythm has to come from the reading lists and the project specs. This site's CS 2881R guide series follows Fall 2025.

Harvard's clearest lesson is that a polished public page still needs an asset-level version audit. CS50 AI shows how lecture and assignment years can diverge; CS1810 shows that a course without public recordings can still be A3; CS1820 shows that the newest catalog entry is not yet a public course. Separating those states produces a route learners can start now and update honestly later.

Changelog

2026-10-10: Added explicit video status and checked recording sources and access notes.
2026-10-01: Added CS 2881R AI Safety (Fall 2025 complete; reading lists, 11 lecture recordings, HW0 repo, and project specs public; A3 by seminar standards) to the entry-point table and the broad-AI/specialist route, with a link to this site's guide; added the Fall 2025 course site to References.
2026-09-30: The CS1810 section and the series disclaimer now link to this site's Harvard CS181 weekly guide (Spring 2026 HW0–HW6 plus midterm/final checkpoints); added it to the references.
2026-08-27: Added a six-round audit disclaimer — CS182 2026/2025/2024/2023 are all A0 for the current offering (Locator Canvas unpublished, no public homework repo; only F22 22-lecture + F18 P0–5 as historical A2); CS181 four-year comparison (two midterms vs midterm+final, hw3–5 reshuffle, practical lifecycle) and the deleted Google Sheet note; weekly guides will use homework numbers as the clock.
2026-08-26: Backfilled an internal link — the CS50 AI deep dive is now live; linked it from the version-comparison passage.
2026-08-22: Standardized the A0/A1 boundary, re-rated catalog-only Fall 2026 courses as A0, and separated future tracking from current self-study routes.

Course video sources

This is a course overview or resource map with no single corresponding lecture. Use the official course entries and playlists to find recordings.

Course and recording entries:

From AI Foundations to Machine Learning: Learning Paths through CS50 AI, CS181, and CS182 | Global AI/CS Course Map