🌏 中文版
Video status: Official entry or recording index only. Source details
Search for Stanford CS229, Berkeley CS188, or MIT deep learning and you will quickly find course sites, YouTube videos, and community notes. The hard question comes next: do those resources belong to the same semester? Do they still open without a university account? Does the assignment link include starter code and required assets, or only a prompt?
This map covers Stanford, Carnegie Mellon University (CMU), MIT, UC Berkeley, Harvard, and National Taiwan University (NTU), spanning 2025–2026. A complete 2026 edition takes priority. If the new semester has only a schedule or keeps video inside an LMS while an official 2025 edition is materially more complete, the 2025 edition can be the main version. Every course guide names its semester; it will not quietly combine 2025 video with 2026 assignments.
This is not a university ranking. It answers two narrower questions: how each school organizes AI and CS, and how much of that structure an outsider can actually use.
Public is not yes or no
“Public course” can refer to at least seven different things: a catalog description, syllabus, slides, assignment prompts, starter code, solutions, or recordings. Publishing one of them is enough for a search engine to surface the site. It is not enough to let someone complete the course.
This site uses four editorial labels. They are not university ratings; they limit what an article is allowed to promise.
| Label | What an outsider gets | What a guide may promise |
|---|---|---|
| A0: catalog visible | Title, units, description | Only the course's place in the map |
| A1: syllabus visible | Syllabus, weekly topics, readings | Scope analysis, not assignment experience |
| A2: materials partly open | Lectures, some assignments, or video | A topic-focused reading guide with explicit gaps |
| A3: self-study ready | Systematic materials, assignments, and required assets | A complete self-study path |
Video is not required for A3. Complete notes, assignments, starter files, and a clear sequence can be enough. The reverse also holds: a YouTube playlist without practice material does not automatically make a course self-study ready.
Try this: when you find an “open course,” do not press play yet. Spend five minutes locating its syllabus, first assignment, and starter code. Decide whether to invest dozens of hours only after all three checks.
Tier one: courses you can follow end to end
One of the cleanest examples is MIT 6.S191: Introduction to Deep Learning. Its 2026 edition (Mar 30–May 25, now completed) publishes nine videos, slides, and three software labs. The 2025 archive keeps ten videos and three labs. It is highly usable for self-study, with one important qualifier: this is an intensive bootcamp, not a semester-long deep learning course.
Berkeley CS188 Spring 2026 is also close to fully public. Its site exposes slides, textbook chapters, discussion material, six Pacman projects, and lecture videos; the semester is over, but the site remains fully intact. The current Fall 2026 offering is underway, ~cs188/ now redirects straight to fa26, and its slides and recordings are released lecture by lecture. Ed, official grades, and staff support remain for enrolled students, but outsiders can still follow the main learning path. The site's CS188 Spring 2026 overview already maps all six projects and a suggested study order.
Stanford is not a collection of isolated recordings either. The Stanford CS course map orders the route from CS106A through CS336 by official prerequisites. CS336 Spring 2026 (Mar 30–Jun 3, now completed) publishes lecture material and five GitHub assignments plus a full YouTube playlist, while Stanford Online carries official Spring 2025 recordings. Its constraint is compute rather than URLs: open materials do not make every assignment free to run.
Courses in this tier are the best candidates for single-course guides because an article can connect what to learn, what to build, and where to begin into one path that works.
Tier two: current material with one missing piece
Berkeley CS288 Spring 2026 publishes slides on post-training, RAG, reasoning, and agents, along with three assignments and a project brief. The missing piece is video. The semester's YouTube playlist exists, but anonymous access returns UNPLAYABLE, matching the course site's Berkeley-login notice. The material supports a reading guide; it does not support the claim that the current recordings are public. The CS288 guide series was written within exactly that boundary.
Berkeley CS285 Spring 2026 publishes slides for twenty-five lectures, five assignments, and GitHub starter code. The lecture recordings are public too, on the RAIL YouTube channel's 2026 playlist; Ed, Gradescope, and office hours stay enrolled-only. The official site also links older Fall 2023 recordings, so a defensible guide should treat the 2026 material and recordings as canonical and place historical video in a separate alternative-resources box with the year shown. The CS285 guide series follows that principle.
The MIT 6.7960 site has switched to Fall 2026, with slides posted as the semester progresses, and Fall 2025 has moved under "Previous years." Both semesters share the same gap: assignments go out through Gradescope and recordings sit in MIT Canvas. This series labels it A2: enough for a deep reading of course design, not enough to promise the enrolled experience. That is why the site's 6.7960 guide uses the Fall 2024 OCW edition, which has complete recordings, and labels the year.
Try this: define the result you want before choosing an A2 course. Slides and readings may be sufficient for understanding a topic. Missing prompts, datasets, or graders are a stop signal if your goal is to complete every assignment.
Tier three: a curriculum in transition, or a semester not yet released
CMU is the school where old course numbers are most misleading. The new 07-280 AI & ML I integrates search, machine learning, LLMs, and reinforcement learning, followed by 07-380 AI & ML II. Its official FAQ explains that the sequence replaces 15-281 and 10-315. Spring 2026 materials for 15-281 remain useful, but the retired course no longer represents CMU's current AI trunk. The CMU AI/ML course map lays out a self-study route under the new curriculum, and the transition itself is covered in "CMU's AI core redesign".
As of August 27, 2026, 07-280 Fall 2026 is a full course site: a 24-lecture schedule (Aug 25–Dec 3), lecture slides and notes, Friday recitations, 12 assignments (HW0–HW11 including Building AlexNet / GPT-2 / AlphaZero), and weekly pre-reading checkpoints are all online. The earlier "most materials not yet released" state is over; the site's complete 07-280 guide (24 lectures + 3 stage reviews + completion roadmap) tracks this current edition.
11-785 Introduction to Deep Learning illustrates a different boundary. Spring 2026 (now completed) and Fall 2025 both link official YouTube recordings lecture by lecture, slides are public, and Fall 2026 is releasing slides and YouTube videos lecture by lecture. Assignments, however, span Autolab, Kaggle, and Piazza. Video access is confirmed; complete self-study still depends on auditing each starter asset.
CMU also launched 11-768 AI Agents in Fall 2026, taught by Graham Neubig and Daniel Fried. It is a graduate course whose prerequisite is having trained a language model; slides and the first lecture recordings are being published as the semester goes, and its three assignments build a harness, then evaluation, then training. The site's 11-768 guide follows the current semester; later lectures and assignments will be added once the official materials go up.
A future or in-progress semester does not become the “latest public course” merely because its schedule exists. This series waits for the material to appear instead of replacing complete content with a newer year.
How to read the six schools
| School | The map's central question | Best current public entry points |
|---|---|---|
| Stanford | How do systems and mathematical foundations lead into research-level AI? | CS221, CS229, CS224N, CS336, CME295; lecture guides cover the five foundation courses and the AI branch courses |
| CMU | How does the new 07-280/07-380 sequence connect to ML, DL, NLP, and systems? | 07-280 Fall 2026 (24-lecture site), 10-301/601, 11-785 (F26 videos rolling out), 11-768 (new in Fall 2026) |
| MIT | How do current numbers, semester sites, and historical OCW editions line up? | 6.S191 (2026 completed); 6.7960 (site now on Fall 2026) as an A2 material guide |
| Berkeley | How do CS188 and CS189 branch into NLP, RL, and vision? | CS188 (sp26 fully preserved, current term fa26); material-focused guides to CS288 and CS285 |
| Harvard | How does the CS50 series connect to CS181 / CS182? | CS50 AI recorded editions; CS181 self-paced by its assignments; CS182 has no public current-term materials yet |
| NTU | How do the program tiers run alongside Hung-yi Lee's EE-department track? | Hung-yi Lee's ML 2026 Spring (A3, taught in Mandarin); how access differs across Hsuan-Tien Lin's and Yun-Nung Chen's courses |
| Other Taiwanese schools | Beyond NTU, which courses stream publicly through TAICA and can be self-studied? | NTHU Hung-Yu Kao NLP (A3), NCCU Yen-Lung Tsai Generative AI (A3); NCKU, Taipei Tech, and NYCU mostly A2 or below |
Harvard's map was added on 2026-08-22, and the NTU and other Taiwanese schools maps on 2026-09-30; see the table above. The Taiwanese courses are taught in Mandarin, so for Chinese-speaking readers it saves more than translation: they can check the original terminology directly against the lecturer's own explanations.
A school map and a single-course guide make different promises. Even when every learning asset is locked in an LMS, current catalogs, program requirements, and schedules may still support a course map. That article can explain how to choose courses; it cannot promise that a reader can complete them without enrolling.
AI security courses
AI security here means protecting the AI system itself: adversarial examples, data poisoning and backdoors, prompt injection, jailbreaks, hijacked agents, and stealing training data or the model itself. AI safety and alignment are out of scope; Harvard CS2881R belongs to that second group.
After scanning the Stanford, CMU, and Berkeley course catalogs in October 2026, this is what the top schools offer:
| Course | School and instructor | Latest term | Level | Notes |
|---|---|---|---|---|
| CS 253 Securing AI Systems | Stanford, Boneh and Mitchell | Spring 2027 | A0 | The old Web Security course number, renamed and relaunched; only the schedule listing exists so far |
| 15-783 Trustworthy AI | CMU, Raghunathan | Fall 2025 | A2 | One module each on jailbreaks, prompt injection, and privacy attacks; lecture notes posted for every class |
| ML for Cybersecurity (17-739/18-739C) | CMU, Lujo Bauer | Fall 2026 | A1 | Mainly ML applied to security, with some coverage of adversarial examples |
| CS 261 Computer Security | Berkeley, Wagner | Spring 2026 | A2 | A general graduate security course; 4 sessions cover LLM attacks and defenses |
| 6.S976 Cryptography and Machine Learning | MIT | Spring 2026 | A2 | Theory-oriented |
Harvard's current course list has nothing in this area. The AI security courses you can self-study end to end (A3) all sit outside these six schools:
- National Chung Cheng University, Van-Linh Nguyen's AI Security (Spring 2026, taught in English): slides, notebooks, and assignments are in a public folder. It is the only fully self-studiable course of its kind in Taiwan.
- University of Cagliari, Battista Biggio's Machine Learning Security (2025/26): lecture materials and notebooks on GitHub.
- Fudan University, Xingjun Ma's Trustworthy AI: 14 weeks of slides plus a full Chinese textbook, no recordings.
All three focus on adversarial examples, poisoning, and privacy attacks, and barely touch LLMs or agents. The closest way to fill that gap today is lectures 13–14 and HW10 of Hung-yi Lee's Introduction to Generative AI 2024 (in Mandarin), followed by the lecture notes for the first module of CMU 15-783.
What to do: start by running an adversarial-example attack yourself with the Cagliari notebooks, then move on to prompt injection against LLMs. Stanford CS 253 starts in March 2027; check back then to see whether it releases materials.
Where CSDIY fits
CSDIY is useful for answering which edition a community has actually tried to study. It often preserves historical recordings, assignment experience, and supplementary resources that a university schedule will never mention.
It cannot by itself prove that a course still runs in 2026, that the current link allows anonymous access, or that a third-party video is official or openly licensed. Conversely, a course missing from CSDIY may still have enough official material for self-study.
The series therefore uses two evidence tracks: official sources establish current offerings and access; CSDIY adds historical versions and community experience. Neither replaces the other. How Berkeley CS188 and CMU 15-281 ended up sharing one set of Pacman projects can only be reconstructed from community records — that lineage gets its own article, "The Pacman AI project lineage".
Where this series stands
Six school maps are done: Stanford, CMU, MIT, Berkeley, and Harvard, and NTU, plus a map of other Taiwanese schools. Single-course guides have expanded significantly:
- Berkeley CS188 Spring 2026 overview, with guides covering search, MDPs, Bayes nets, and machine learning (sp26 completed, current term fa26)
- Berkeley CS285 Spring 2026 overview, with guides from imitation learning and policy gradients through offline RL
- Berkeley CS288 overview, from foundations and transformers to agents
- CMU 10-301/601 overview, walking the whole course through nine assignments
- Complete CMU 07-280 course guide: all 24 lectures for Fall 2026, plus a wrap-up with a route to the next course
- Complete CMU 11-785 deep learning guide: all 28 lectures, with the incomplete assignment chain flagged (S26 completed, F26 videos rolling out)
- CMU 11-768 AI Agents guide: the graduate agents course new in Fall 2026, covering harnesses, evaluation, and training, updated as the semester progresses
- MIT 6.S191 guide: how to actually run the nine lectures and three labs that are fully public (2026 edition Mar 30–May 25 completed)
- Stanford AI lecture series: CS221, CS229, CS224N, CS224W, CS224V, CS124, CS230, CS329Z, CS224U (the Spring 2023 historical edition, 16 posts), and CS336 (plus topic deep dives starting from tokenization); see the Stanford course map for each course's edition and prerequisites
- Stanford CS231N guide: deep learning for computer vision, with Spring 2026 slides and assignments A1–A3, and recordings from Spring 2025
- MIT 6.S184 guide: flow matching and diffusion models from IAP 2026, with notes, recordings, and three labs all public
- CMU 11-868 LLM Systems guide: seven assignments running from CUDA kernels and distributed training to serving and RLHF; no recordings, and a GPU is required
- NTU Hung-yi Lee Machine Learning 2026 Spring guide: taught in Mandarin, from dissecting OpenClaw and context engineering to the KV cache and harness engineering
- Stanford CS224R guide: deep RL, Spring 2026, from imitation learning to RLHF, LLM reasoning, and robot VLAs
- Stanford CS234 guide: RL theory, Winter 2026 slides and assignments, with Spring 2024 videos
- Stanford CS149 guide: parallel computing, Fall 2025, including GPU/CUDA and AI-accelerator assignments, with 2023 videos
- MIT 6.5940 guide: efficient deep learning, based on the latest complete edition, Fall 2024 (no Fall 2025 offering), compared with the in-progress Fall 2026
- CMU 10-423 guide: generative AI, Spring 2026, built around HW1–HW4
- Harvard CS2881R guide: AI safety, Fall 2025, with public recordings, readings, and assignments
- Taiwanese courses in Mandarin: NTU Yun-Nung Chen ADL 2025 Fall, NTU Hsuan-Tien Lin ML Foundations & Techniques (content from the MOOCs recorded around 2013, practice from the Fall 2024 homework), NTHU Hung-Yu Kao NLP, and NCCU Yen-Lung Tsai Generative AI
- Stanford CME295 guide: a two-unit Transformer and LLM course with no assignments; the 2025 edition is fully public, and the 2026 edition adds LLM systems, reinforcement learning, and diffusion LLMs
- Stanford foundations lecture series, all with bilingual guides: CS103 28 lectures, CS107 30 lectures, CS109 22 lectures, CS111 28 lectures, CS161 18 lectures
- Berkeley CS189 guide: the overview starts from Spring 2025, while the lecture guides and HW1–5 follow the Spring 2026 course site now back online, compared against the Fall 2026 (eecs189.org/fa26) calendar
- CMU 07-380 guide: first offered in Fall 2026; the published lectures 1–10 and HW1–3 each have their own post, and the second half of the semester will be added as the course site progresses; compared with 07-280
On the Harvard side, the CS50 AI guide and the weekly CS181 guide (all assignments, HW0–HW6) are both live. Only CS182 remains unwritten: it appears on the SEAS schedule for Fall 2026, but no public course site can be found yet, so it waits until materials appear.
If you want one course to start tonight, run a small test: open MIT 6.S191's first lab or Berkeley CS188's first project and give yourself ninety minutes. If you can still state what the environment is missing and what you would do next, the course belongs on your learning plan. Saving a playlist is not a start.
Update log
- 2026-10-10: Added explicit video status and checked recording sources and access notes.
- 2026-10-10: Fixed an outdated recording statement. CS285 lecture recordings are public on the RAIL YouTube channel, not only in bCourses.
- 2026-10-05: Added an "AI security courses" section covering current AI security courses and access levels at Stanford, CMU, Berkeley, and MIT, plus the three fully self-studiable courses in Taiwan and abroad.
- 2026-10-01: "Where this series stands" now includes six new series: Stanford CS224R, CS234, CS149, MIT 6.5940, CMU 10-423, and Harvard CS2881R.
- 2026-09-30 (3): Added the other Taiwanese schools map; "Where this series stands" now includes four Mandarin course series: NTU ADL, Hsuan-Tien Lin, NTHU NLP, and NCCU Generative AI.
- 2026-09-30 (2): Added the NTU AI/ML course map as the sixth school; "Where this series stands" now includes four new series: Stanford CS231N, MIT 6.S184, CMU 11-868, and NTU Hung-yi Lee's ML 2026.
- 2026-09-30: Series expansions reflected here: Berkeley CS189 gained Spring 2026 lecture guides and HW1–5, CMU 07-380 gained lectures 1–10 and HW1–3, Harvard CS181 now covers HW0–HW6, and Stanford CS224U gained 16 posts on the Spring 2023 edition; list descriptions updated to match.
- 2026-09-29: Fall 2026 status audit: Berkeley CS188's current term is now fa26 (
~cs188/redirects to fa26), the MIT 6.7960 site has switched to Fall 2026, and CMU 11-785 F26 videos are rolling out lecture by lecture; added a CMU 11-768 AI Agents paragraph and guide link; "Where this series stands" now includes the Stanford CS229 / CS224N / CS224W / CS224V / CS124 / CS230 / CS329Z / CME295 series and the Harvard CS181 series; the "still unwritten" list is corrected to Harvard CS182 only. - 2026-08-27 (3): Added CMU 07-380 Fall 2026 overview (first offering, 26 lectures A2→A3 transition, Lec01 public, compared with 07-280); removed 07-380 from "still unwritten" list.
- 2026-08-27 (2): Added Berkeley CS189 Spring 2025 overview (HW1–7 A3, compared with Fall 2026 eecs189.org/fa26 27-lecture calendar); removed CS189 and MIT 6.7960 / Harvard CS50 AI from the "still unwritten" list (the latter two are now live in
ai/tech); original 2026-08-27 entry retained. - 2026-08-27: Re-audited Spring 2026 as completed: MIT 6.S191 (Mar 30–May 25), Stanford CS336 (Mar 30–Jun 3), Berkeley CS188 sp26 (archived, now su26), and CMU 11-785 S26 (F26 online) now marked as completed; CMU 07-280 Fall 2026 updated from "most materials not yet released" to 24 lectures / 12 assignments / weekly checkpoints all online, matching this site's 07-280 lecture guides; table expanded to five schools and Stanford CS103 / CS107 / CS109 / CS111 / CS161 completion noted; 07-380 now "first offered Fall 2026" per FAQ.
- 2026-08-26: An audit found that the MIT 6.S191, CMU 11-785, CMU 07-280, CS336, and CS221 guides are already live on this site under the
aicategory. The "still unwritten" list now correctly names 6.7960, CS189, and Harvard's three courses. - 2026-08-26 (earlier): The follow-up articles (four school maps, Harvard, and the CS188 / CS285 / CS288 / 10-301 guide series) are now live. Added inline links throughout, replaced "What comes next" with a current-state list, and restructured References into site articles versus official sources.
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:
- Berkeley CS188 Spring 2026
- Fall 2026
- CS336 Spring 2026
- Berkeley CS285 Spring 2026
- 07-280 AI & ML I
- 11-785 Introduction to Deep Learning
References
Site series articles
- Stanford CS course map
- CMU AI/ML course map
- MIT AI/ML course map
- Berkeley AI/ML course map
- Harvard AI/ML course guide
- NTU AI/ML course map
- Other Taiwanese schools AI course map
- CMU's AI core redesign: from 15-281 + 10-315 to 07-280 + 07-380
- The Pacman AI project lineage
- Berkeley CS188 Spring 2026 overview
- Berkeley CS285 Spring 2026 guide series overview
- Berkeley CS288 Spring 2026 guide series overview
- CMU 10-301/601 machine learning guide overview
- Complete CMU 07-280 course guide
- Completing CMU 07-280 wrap-up
- Complete CMU 11-785 deep learning guide
- MIT 6.S191 guide
- MIT 6.7960 guide
- Harvard CS50 AI guide (weekly W00–W06 + synthesis)
- Berkeley CS189 guide
- CMU 07-380 guide
- Stanford CS224U guide
- Stanford CS336 topic deep-dive series
- Stanford CME295 guide
- CMU 11-768 AI Agents guide
- Harvard CS181 weekly guide
- Stanford CS231N guide
- MIT 6.S184 guide
- CMU 11-868 LLM Systems guide
- NTU Hung-yi Lee Machine Learning 2026 Spring guide
- NTU Yun-Nung Chen Applied Deep Learning 2025 Fall guide
- NTU Hsuan-Tien Lin ML Foundations & Techniques guide
- NTHU Hung-Yu Kao NLP guide
- NCCU Yen-Lung Tsai Generative AI guide
- Stanford CS224R guide
- Stanford CS234 guide
- Stanford CS149 guide
- MIT 6.5940 guide
- CMU 10-423 guide
- Harvard CS2881R guide
Official course sites and external resources
- Stanford CS336 Spring 2026 (full YouTube playlist)
- CMU 07-280 AI & ML I Fall 2026
- CMU 07-380 AI & ML II Fall 2026
- CMU 11-768 AI Agents Fall 2026
- Stanford CME295 Transformers & LLMs
- CMU 11-785 Spring 2026 (Fall 2026 online) / Fall 2025
- MIT 6.S191 Introduction to Deep Learning (2025 archive)
- MIT 6.7960 Deep Learning Fall 2026 (Fall 2025 listed under Previous years)
- Berkeley CS188 Spring 2026 / Fall 2026 current
- Berkeley CS285 Spring 2026
- Berkeley CS288 Spring 2026
- CSDIY
- Stanford CS 253 Securing AI Systems (ExploreCourses)
- CMU 15-783 Trustworthy AI Fall 2025
- CMU 17-739/18-739C ML for Cybersecurity (Lujo Bauer's teaching page)
- Berkeley CS 261 Spring 2026
- MIT 6.S976 Cryptography and Machine Learning
- National Chung Cheng University AI Security (Van-Linh Nguyen)
- Cagliari Machine Learning Security
- Fudan Trustworthy AI (Xingjun Ma)
- Hung-yi Lee, Introduction to Generative AI 2024
Loading...