From Programming and Mathematics to AI: A Learning Path through Open Courses | Global AI/CS Course Map

🌏 中文版

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

First, a correction to an error that is easy to inherit from old guides: MIT has a formal AI degree. Course 6-4 is the Bachelor of Science in Artificial Intelligence and Decision Making, and students have been able to declare it since Fall 2022. It is not merely a few machine-learning electives hidden inside the computer science degree.

That does not mean MIT publishes a seamless online path from a first Python lesson to its latest work on large models. The resources split into three layers that often fail to align:

1. The current degree requirements show what MIT believes an AI student must know.
2. The 2025–2026 course sites show what was recently taught, but the material may sit behind Canvas, Piazza, Panopto, or Gradescope.
3. OCW, Open Learning Library, and archived course sites are often the most usable for self-study, but may use an old subject number or an earlier syllabus.

This guide keeps those layers separate. It does not pretend that an outside learner can “take an MIT degree online.” It builds a path whose links open and whose missing pieces are explicit. The site already has a detailed guide to MIT 6.S191; this article places that short course inside the larger MIT structure.

Two numbering systems: Course 6-4 is not subject 6.4

Course 6-4 identifies a degree program. 6.3900 and 6.4110 identify individual subjects. A second source of confusion is MIT EECS's Fall 2022 renumbering, which replaced most three-digit decimal subject numbers with four-digit ones.

These are the mappings that repeatedly appear in AI and ML reading lists:

Two traps matter. First, the classic OCW 6.034 is not an old edition of 6.4110. It became 6.4100; 6.4110 descends from 6.038. Second, 6.7960 Deep Learning is not the new number for 6.867. The latter became 6.7900, and both subjects now exist.

A YouTube title, a CSDIY page, or a decade-old reading list therefore cannot establish a course lineage merely because two subjects sound similar.

MIT's AI foundation is not “Python, then ML, then deep learning”

The 2025 Course 6-4 requirements begin with six foundation subjects covering programming, software construction, discrete mathematics, algorithms, linear algebra, and probability or inference. There are several approved substitutions; reduced to capabilities, the prerequisite structure looks like this:

Programming: 6.100A + 6.100B, or integrated 6.1000
        ↓
Software and algorithms: 6.1010 + 6.1200 → 6.1210
        ↓
AI mathematics: 18.06/18.C06 linear algebra + 6.3700/6.3800/18.05 probability and inference
        ↓
Core branches: 6.3900 machine learning | 6.4110 representation, inference, and planning

This is a route inferred from official prerequisites, not an MIT-issued semester plan. Its point is that 6.3900 is not the beginning. Without basic programming, algorithms, linear algebra, and probability, it is easy to confuse “I can run the notebook” with understanding the model.

Course 6-4 then requires at least one subject from each of five Centers:

• Data-centric: how data is generated, estimated, and learned from; 6.3900 is a common entry.
• Model-centric: how a system represents the world and reasons under uncertainty; 6.4110 is one current route.
• Decision-centric: how an agent plans, controls, and acts under uncertainty.
• Computation-centric: algorithms, computational resources, and system constraints.
• Human-centric: causality, feedback, and risk when AI enters social systems, including subjects such as 6.3950.

One subject cannot fill two required-subject slots. The rule deliberately prevents students from filling all five boxes with courses that merely look like ML. The degree also includes an Application CI-M, an advanced AI+D subject, an EECS or mathematics elective, at least two CI-M subjects overall, and a SERC-qualified subject. Social responsibility is part of the formal structure, not an appendix added after the models.

Students who prefer a broader systems and CS foundation before choosing AI as one area of depth can also take that route through Course 6-3, Computer Science and Engineering. MIT does not imply that Course 6-4 is the only legitimate way to study AI.

What an outside learner can actually access in 2025–2026

The ratings below use the editorial scale defined in the Global AI and CS Course Map: A0 means a catalog entry is visible, A1 adds a syllabus, A2 exposes some substantive material, and A3 is coherent enough to organize into a self-study course. These are not MIT ratings, and A3 does not include credit, instructors, grading, peers, or free compute.

The important lesson is not which subject receives the highest score. It is that the same course can become less public over time. The Fall 2026 preview for 6.3900 says most materials will be open, yet the completed Spring 2026 home page and assignments are now staff-only. If a page used to open and no longer does, the old search result may not have been wrong; the archive policy changed.

The reverse also happens: the newest class is only partly public while the previous edition is complete. Fall 2025 6.7960 keeps recordings and solutions in Canvas. MIT OCW's Fall 2024 edition publishes the videos, notes, five assignments, and final project. For self-study, choosing 2024 OCW is more honest than forcing the 2025 site into a course it cannot provide. Check the newest edition first, but do not let the year outrank completeness. For how to combine the two editions and which route to take for which goal, this site's 6.7960 deep dive breaks both down edition by edition.

Three routes that an outside learner can finish

1. A modern AI engineering introduction

Build the Python, linear algebra, and probability prerequisites, then use the public 6.3900 notes to establish the language of supervised learning and modeling. From there:

6.3900 public notes
  ├─ 6.S191 Spring 2026: a nine-week introduction to modern deep learning
  ├─ 6.7960 Fall 2024 OCW: a semester of advanced deep learning
  └─ 6.S184 IAP 2026: a focused course on flow matching and diffusion models

6.S191 Spring 2026 quickly connects convolutional networks, Transformers, generative models, and AI for science. It met weekly from March 30 through May 25; it was not a one-week IAP bootcamp, nor is it the predecessor or equivalent of 6.7960. For a theory-oriented route with 24 lectures and longer assignments, use the 6.7960 OCW edition.

To go deep on generative models, add 6.S184. It is a short course from MIT's January IAP with only five lectures, yet its 84-page lecture notes carry you from ODEs and SDEs through flow matching, score matching, and guidance to a working latent diffusion model. The notes, slides, recordings, three labs, and official solutions are all public, which makes it one of the cleanest A3 courses of the year. The prerequisites are linear algebra, multivariable calculus, probability, and some PyTorch; rusty probability is where most learners stall. This site's 6.S184 guide series maps each lecture to its notes sections and labs.

2. AI beyond neural networks

Take the 6.4110 route through constraint satisfaction, logic, graphical models, MDPs and POMDPs, and planning. Its public assignments are more useful than its locked recordings, so an outside learner can read the slides and then work through the CAT-SOOP problems.

The classic 6.034 Fall 2010 OCW course still supplies video, assignments, and exams for search, symbolic AI, and traditional methods. It is valuable historical grounding, but it does not represent the 2025–2026 frontier and must not be relabeled as current 6.4110.

3. Vision, robotics, and efficient ML systems

The Spring 2026 computer vision offering ran under the special-subject number 6.S058. It publishes slides, a book, problem sets, Colabs, and project requirements, making it one of the few current branches close to A3. The catalog's 6.4300 is the related regular-subject number, but the two labels should not be silently interchanged.

For embodied AI, follow 6.3900, linear algebra, and probability with 6.4110, then use the public 6.4210 Robotic Manipulation notes. Its Gradescope and Deepnote assignments and Canvas video are unavailable, so an external version of the course needs to center on lecture notes, Drake examples, and a self-defined project.

For efficient ML systems, use Song Han's 6.5940. It was not offered in Fall 2025 because of his sabbatical; Fall 2026 is back and in session, and as of 2026-09-30 only L1–L6 and Lab 0–1 are out. For a complete pass now, use Fall 2024 as the backbone: 23 lecture decks, 23 recordings, and Lab 0–5 are all public, running from pruning, quantization, NAS, and distillation through LLM deployment and distributed training, with the course page promising that students deploy Llama2-7B on their own laptops. Prerequisites are 6.191 and 6.390. This site's 6.5940 guide series follows Fall 2024 and adds a Fall 2026 comparison to each part.

What CSDIY can and cannot confirm

CSDIY is useful evidence that a self-study community has organized a particular edition into a workable route. It is not sufficient evidence that MIT offered the subject this year or that its current site still opens anonymously.

Among the MIT subjects above, CSDIY currently has dedicated pages for only:

• MIT 6.7960: linked to MIT OCW Fall 2024, estimated at about 90 hours, and positioned after an ML foundation as advanced broad-coverage deep learning.
• MIT 6.5940: organized around the 2023 and 2024 sites and labs as advanced ML systems study.

It does not list 6.S191, 6.3900, 6.4100, or 6.7900. That absence does not mean those subjects do not exist or cannot be studied. Conversely, the presence of a 6.5940 page does not mean MIT offered it in 2025–2026. A reliable check runs in this order:

1. Use the MIT Registrar archive or term site to confirm that the class actually ran.
2. Open the syllabus, slides, video, assignments, and code anonymously instead of stopping at the home page.
3. Use CSDIY to understand how a community sequences an archived edition, estimates its workload, and handles prerequisites.

If you do one thing today

Do not bookmark ten courses. Spend half an hour on three steps:

1. Open the Course 6-4 requirements and copy its six foundation subjects and five Centers into a checklist.
2. Choose one of the 6.S191, 6.4110, or 6.S058 A3/A2 routes and open the first lab or homework—not merely the first video.
3. Write down the missing prerequisite: Python, algorithms, linear algebra, or probability. Go back and fill that one first.

The most valuable thing to borrow from MIT is not a list of famous subjects. It is the way Course 6-4 divides AI capability into data, models, decisions, computation, and people. Public material lets an outside learner borrow much of that structure. Credit, feedback, peers, gated platforms, and the research environment remain outside the package.

Changelog

• 2026-10-10: Added explicit video status and checked recording sources and access notes.

• 2026-10-01: Moved 6.5940 Fall 2026 from A0 to A2 in the access table (term in progress, L1–L6 and Lab 0–1 released) and spelled out the Fall 2024 scope (23 lecture decks, 23 recordings, Lab 0–5); rewrote the 6.5940 paragraph in the vision, robotics, and efficient systems route with a link to this site's guide; added the Fall 2026 course page to References.

• 2026-09-30: Added 6.S184 Flow Matching & Diffusion (IAP 2026; notes, slides, recordings, and three labs with solutions all public, A3) to the access table and the modern AI engineering route, with a link to this site's guide.

• 2026-08-26: Backfilled an internal link — the 6.7960 deep dive is now live; linked it from the two-edition comparison passage.

• 2026-08-22: Corrected 6.S191 2026 to a nine-week spring course rather than a one-week IAP bootcamp.

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:

• MIT has a formal AI degree.
• 6.S184
• MIT EECS — 6-4_2025 Degree Requirements
• MIT Catalog — Course 6-4 Degree Chart
• MIT EECS — New Subject Numbering
• MIT EECS — Old and New Subject Number Crosswalk

References

• MIT EECS — Course 6-4: Artificial Intelligence and Decision Making
• MIT EECS — 6-4_2025 Degree Requirements
• MIT Catalog — Course 6-4 Degree Chart
• MIT EECS — New Subject Numbering
• MIT EECS — Old and New Subject Number Crosswalk
• MIT 6.S191 — Introduction to Deep Learning 2026
• MIT 6.3900 — Introduction to Machine Learning, Spring 2026
• MIT 6.4110 — Representation, Inference, and Reasoning in AI, Spring 2026
• MIT 6.7960 — Deep Learning, Fall 2025
• MIT OpenCourseWare — 6.7960 Deep Learning, Fall 2024
• MIT 6.S184 — Generative AI with Stochastic Differential Equations, IAP 2026
• MIT 6.S058 — Introduction to Computer Vision, Spring 2026
• MIT 6.4210 — Robotic Manipulation, Fall 2025
• MIT Han Lab — 6.5940 TinyML and Efficient Deep Learning Computing, Fall 2024
• MIT Han Lab — 6.5940 TinyML and Efficient AI Computing, Fall 2026
• MIT 6.7900 — Machine Learning, Fall 2025
• MIT 6.7920 — Reinforcement Learning: Foundations and Methods, Fall 2025
• MIT 6.8610 — Quantitative Methods for Natural Language Processing, Spring 2026
• CSDIY — MIT 6.7960
• CSDIY — MIT 6.5940