AI engineers make an average of \$247k/year¹, 1/5 of venture capital is going to AI startups², and companies are planning to invest \$200 billion into AI infrastructure by next year³.

These figures will continue growing as AI becomes the most impactful technology since the Internet. And much like the Internet, AI will have countless opportunities for people with the right skills and experience.

For complete beginners seeking a straightforward, non-technical overview of AI, the best course is AI For Everyone by Andrew Ng on Coursera. Specifically designed without a focus on mathematics or coding, this course gives you a holistic view of AI's capabilities and its future prospects.

For readers wanting technical skills and experience, the formula is simple: you need a strong foundation of the fundamentals, some specialized knowledge in a category or subtopic, and a few projects under your belt.

My goal with this article is to help you determine which of the thousands of courses fit your goals and are worth your time and effort—something I recently worked out for myself.

While working on an AI startup recently, I needed a better understanding of fundamental concepts, LLM architectures, and cutting-edge techniques. The last thing I wanted to do was search every course platform manually, so I spent exponentially more time building an over-engineered course scraper that found and filtered everything for me 🙃.

The scraper produced promising course options, but only a few of which I felt worthy of enrollment and completion. I'll share the courses I found to be the best for me, along with others suited to learners with different goals.

Here's a preview of the courses we'll discuss:

Who should take this course?

AI newcomers desiring a broad, non-technical overview of the field

What you'll get from this course

AI For Everyone is the best introductory course for developing AI literacy.

Enrolling in this course will give you a comprehensive overview of AI, what it can do, its misconceptions and benefits, and what to expect for the future. This is a non-technical introduction for complete beginners, so there are no prerequisites, mathematics, or coding.

The core focus of this course is to give you the ability to converse intelligently about AI technology and how you can start incorporating it into your career or business.

Another beginner-friendly course that specifically examines GenAI is Ng's new course, Generative AI for Everyone. Take this course if you want a similar non-technical approach but for large-language models.

Who's teaching?

If you spend more than five minutes searching for courses in AI and Machine Learning, you'll see Andrew Ng's name across many of the best.

Some notable areas of Ng's résumé:

• Co-founded Coursera, one of the world's most popular online learning platforms
• Co-founded and led Google Brain, an AI research team
• Built the Artificial Intelligence team at Baidu
• Professor at Stanford University, where he teaches machine learning and AI
• Creator of the world's best machine learning course

Needless to say, Ng is a prominent figure in the AI and online learning communities and you'll learn a lot from taking his courses.

The rest of this article will recommend the best technical courses which require some math and programming experience. Enroll in AI For Everyone

Who should take this course?

Anyone interested in learning a wide range of AI techniques from some of the top AI experts

What you'll get from this course

This course's curriculum follows a similar but condensed path to Norvig's textbook, forming a general overview of core AI techniques. The tools you learn are the parts of AI that don't involve machine learning or generative AI—it's more of a foundational course.

After purchasing Norvig's textbook for self-study, I found this course helpful in adding context to the book and providing meaningful feedback via quizzes and projects.

Each course module culminates in projects that reinforce what you've learned. These include building a sudoku solver, a forward-planning agent, an adversarial game-playing agent, and a part-of-speech tagging model. These projects will solidify your understanding and serve as valuable additions to your portfolio.

Who's teaching?

Peter Norvig is co-author of Artificial Intelligence: A Modern Approach, the most popular textbook for AI university programs.

His experience includes:

• Taught at several prestigious universities, including Stanford and UC Berkeley
• Directed Google's search algorithm group
• Headed NASA's Computational Sciences Division
• Co-taught a Stanford AI online course that enrolled over 160,000 students

Norvig is known for his combination of practical and theoretical teaching, giving students a deep and intuitive understanding of AI.

Overall, this course offers a strong foundation in Artificial Intelligence techniques. The content mirrors many intro AI courses offered at universities and is presented by two top minds in the industry. Enroll in Artificial Intelligence Nanodegree

Who should take this course?

AI learners who want a solid computer science foundation

What you'll get from this course

This is a five-month professional certificate on edX that tracks Harvard's in-person CS50 program. It's one of the most popular beginner-friendly computer science courses you'll find online, and it stands out for several reasons.

The hallmark of this program is the heavy use of unique and challenging problem sets. In the first course, you'll build programs that challenge your understanding of coding, algorithms, and data structures, all crucial software engineering concepts for developing AI.

The next course introduces fundamental Artificial Intelligence concepts, with each module requiring you to build various AI programs.

The projects you build will have an AI do things like:

• Play tic-tac-toe and minesweeper
• Build crossword puzzles
• Identify traffic signs in photographs (i.e., for self-driving cars)
• Parse sentences and extract noun phrases (NLP)
• Predict a masked word in a sentence (language models)

Even though this course has sections on multiple programming languages, you may find it challenging to keep up if you've never seen code before. However, if you already have an adequate CS foundation, jumping to the second course might be a better fit and save time.

Finally, what really makes this program unique are the instructors and community. Backed by incredible video production quality, the teachers are some of the best I've ever seen in an online course. And due to this program's popularity, the instructors and teaching assistants have fostered a large, highly active online community of other CS and AI learners.

Who's teaching?

The first instructor you'll see is David J. Malan, a computer scientist and professor at Harvard's School of Engineering and Applied Sciences. Due to the popularity of CS50, he's most well-known because of this course. Outside of his experience at Harvard, he was also Chief Information Officer at Mindset Media, where he designed infrastructure to perform hundreds of millions of HTTP requests per day to create massive datasets.

In the second course, instructor Brian Yu will teach you about AI. Yu is a software developer and educator who, in addition to teaching at Harvard, produces educational computer science content on his popular YouTube channel, Spanning Tree.

I found the lessons to be an entertaining and insightful mix of on-stage presentations and code demonstrations. The lecturers are excellent teachers, but there's no hand-holding. You'll find this series challenging and demanding, something you'd expect from an actual college course. Enroll in Computer Science for Artificial Intelligence Professional Certificate

Who should take this course?

Comfortable Python programmers who want to build LLM projects with recent advancements

What you'll get from this course

Some neat LLM improvements, like Retrieval Augmented Generation (RAG), vector databases, and agentic workflows, let you make more complex systems that solve problems better than the raw chat APIs.

These improvements elevate your app from a simple wrapper around ChatGPT to a more advanced software product that allows you to outpace potential competitors.

This course teaches you the most popular Python libraries and techniques for taking advantage of these innovations. To demonstrate how they work in real projects, the instructor codes end-to-end examples, such as:

• An app that generates potential networking icebreakers by using LinkedIn profile data
• An assistant that reads code documentation and answers queries
• A slimmer version of ChatGPT's code interpreter, which writes and executes its own code

This course aims at developers who want to build advanced LLM applications quickly, and not so much for learning the mathematics and theory behind these things.

If you're like me and want to survey the current LLM tech stack, I can't recommend this course enough. The audio/video production quality is excellent, there's a great balance of breadth and depth, and the hands-on example projects let you easily see how everything fits together.

Who's teaching?

Eden Marco is an experienced software engineer and best-selling instructor at Udemy, with over 56k students and an average course rating of 4.5 stars. Marco currently works at Google Cloud as an LLM Specialist and Customer Engineer.

Eden is an excellent teacher dedicated to making good content. His courses are all well-structured and produced, updated frequently to reflect changes in technology, and students always have their questions answered quickly.

After completing this course, you'll have built several small projects with advanced LLM techniques. From here, you should feel comfortable creating new apps and solutions with these tools. Enroll in LangChain - Develop LLM powered applications with LangChain

Who should take this course?

Developers with machine learning experience who want to build more advanced LLM applications

What you'll get from this course

Over three weeks, this course will teach you how LLMs work, what they can do, and how to build projects with them.

This course is perfect if you've heard concepts thrown around, like "Transformers," "zero-shot inference," "fine-tuning," and "RLHF," but haven't spent the time learning what they are. Each module introduces a concept, describes how it works at a high level and mathematically, and then walks through Python and PyTorch code to build an example model architecture.

To get the most out of this course, you'll need some machine learning experience. For example intro courses, see my best machine learning courses article.

Who's teaching?

This program is by Databricks, so you won't be surprised that your instructors are on the Databricks team.

In order of appearance, your instructors are:

• Sam Raymond - Sr. Data Scientist - received his Ph.D. in Computation Engineering and Machine Learning from MIT. Outside of Databricks, Sam was a postdoc researcher in deep learning and data science at Stanford and MIT
• Chengyin Eng - Sr. Data Scientist - received her Master's in Computer Science from UMass Amherst. Eng has spoken at many popular machine learning conferences, including Open Data Science and PyData.
• Joseph Bradley - Lead ML Product Specialist - received his Ph.D. in Machine Learning from Carnegie Mellon. Bradley was previously a software engineer at Databricks and a postdoc at UC Berkeley.

This course offers a broad survey of state-of-the-art LLM architectures, covering a wide array of recent advancements, optimizations, and challenges without demanding rigorous theoretical study. The labs reinforce these concepts through practical Python examples, ensuring a solid grounding in the principles. Enroll in Large Language Models Professional Certificate

Who should take this course?

Students with some experience who want to dive into the deep learning branch of AI

What you'll get from this course

Meant to follow up on Andrew Ng's Machine Learning Specialization, this course takes a deep dive into deep neural networks, the technology responsible for many AI innovations.

All AI learners should have hands-on experience with neural networks and deep learning to understand the most impactful AI innovations (e.g., self-driving cars and large language models). Over five months, you'll learn both how deep neural networks work mathematically and how to build them from the ground up via guided coding sessions.

There are too many concepts you learn and apply to mention here; check the syllabus for a detailed list. Just know that each module provides quizzes and challenging projects to ensure you burn in the information you've learned from the instructors.

If you already have the requisite knowledge—Python, Linear Algebra, Machine Learning—then this program will give you an intuitive understanding of the core architecture behind state-of-the-art AI.

Who's teaching?

Andrew Ng, Younes Bensouda Mourri, and Kian Katanforoosh teach this course. You've already read about Ng's background in the course above, so here are a few details about the other two instructors:

Younes Bensouda Mourri teaches AI on campus and online at Stanford. He co-created three graduate-level AI courses for Stanford and is also the instructor for the Natural Language Processing Specialization on Coursera. Additionally, he is the founder of LiveTech.ai, an application that builds AI tools for educators.

Kian Katanforoosh is a Stanford Computer Science lecturer and co-creator of Stanford's Deep Learning course. Kian is a founding member of DeepLearning.ai and founder of Workera, a platform for enterprises to evaluate and improve their employee's technical skills.

In the last two courses of this Specialization, you'll learn about computer vision and natural language processing. These two crucial AI subtopics will make for a well-rounded introduction to the field. Deep learning is only one of many AI techniques, so you may want a broader overview of AI before enrolling in this course series. Enroll in Deep Learning Specialization

Who should take this course?

Advanced learners who have a strong interest in autonomous driving and enjoy hands-on learning

What you'll get from this course

Many courses teach robotics and self-driving vehicles, but this one is unique in that it pairs lessons with a physical kit of road mats, traffic cones and signs, and a programmable self-driving robot, called Duckiebot.

With this scale model of the real world, you learn the essential functions of an autonomous vehicle on the road, like how to recognize and drive in lanes, stop at intersections, and detect and avoid crashing into objects. You build this capability through Python and machine learning frameworks like PyTorch and TensorFlow.

I've slotted this course in the advanced category due to its prerequisites. Before starting, you should already be familiar with Python, Linux, and Git, as well as knowledge of probability, linear algebra, and calculus.

If you're interested in the robotics applications of artificial intelligence and want a real hands-on approach, this program is a fantastic introduction to everything from the physical to the digital.

Who's teaching?

Six instructors teach this course, which is too many to detail individually in this short section. Briefly, there is a professor and two senior researchers from ETH Zurich, a professor from the University of Montreal, and a professor from Toyota Technological Institute at Chicago.

This team has vast experience with robotics, aerospace engineering, computer science, and control and dynamical systems. I have yet to see another self-driving or robotics course with this level of expertise.

Duckiebot uses an NVIDIA Jetson Nano, a small computer built for AI IoT applications, for which you'll learn how to program. Once you've completed the course, you'll have knowledge of basic robotics, IoT, and reinforcement learning (e.g., Q Learning), after which you can go on and apply your new skills to all sorts of IoT and robotics applications. Enroll in Self-Driving Cars with Duckietown

Who should take this course?

Self-starters looking for a completely free, top-tier course

What you'll get from this course

MIT OpenCourseWare (OCW) hosts this completely free AI course, which includes lecture videos, problem sets and exams with solutions, and lecture notes.

This is an undergraduate course from 2010, and despite its age, it does teach neural networks. However, it's focused more on classic AI algorithms and applications, similar in scope to Udacity's AI course and Harvard's CS50 program. MIT's course is also lighter on programming than the latter; you'll still need some Python programming experience to solve problem sets, but not for quizzes and exams.

Unfortunately, MIT OCW isn't an interactive platform like Coursera, Udacity, or Edx, so you'll need to self-motivate, follow a schedule, and grade your own assignments. If you can manage that, you'll receive a similar AI educational experience to MIT students without the cost.

Who's teaching?

Patrick Winston was a computer scientist and renowned MIT professor. Winston received his Ph.D. from MIT in 1970, directed MIT's Artificial Intelligence lab for over 20 years, and authored multiple books on programming languages and AI.

Even if you don't commit to this course, you should still watch the first video just to see how well Winston teaches—he was truly a master educator.

The easiest way to watch the lectures for this course are through this YouTube playlist, but you'll still need to reference the OpenCourseWare page for notes, assignments, exams, and solutions. Enroll in Artificial Intelligence

Who should take this course?

Intermediate to advanced students with machine learning experience who want to lay the groundwork for language-based systems, like LLMs, speach recognition, and information retrieval.

What you'll get from this course

Stanford is following in MIT OpenCourseWare's footsteps by providing access to their world-class courses for free online. This particular course grants learners access to an analog of their on-campus course, CS224N: Natural Language Processing with Deep Learning, via a YouTube playlist.

If you've seen how you can query a model for language quirks, like "(King - Man) + Woman = Queen," you've probably wondered how the model can reason this way. This is an example of a core NLP innovation that helped spur AI's boom, and you learn this on day one of this 10-week course.

As an NLP enthusiast myself, I was impressed by the breadth and depth of NLP and Deep Learning topics covered in this course. A long time ago, I worked on a project that involved predicting stocks using social media (spoiler alert: it didn't work out), and this course would have been an invaluable, all-encompassing resource.

If you have the requisite knowledge—Python, Calculus, Linear Algebra, and Probability—then this is the absolute best NLP course you can find online and a perfect starting point for learning a crucial AI building block.

Who's teaching?

Professor Christopher Manning teaches Machine Learning, Linguistics, and Computer Science at Stanford, where he's also director of the Artificial Intelligence Laboratory.

Manning pioneered several NLP Deep Learning algorithms, authored multiple highly-rated textbooks in NLP and information retrieval, and has taught the NLP with Deep Learning course at Stanford every year since 2000.

Unfortunately, as a free observer, you don't get graded assignments and exams, and the community that come with the real course. Regardless, give this course a shot if you want to experience the same NLP and Deep Learning lectures as Stanford students. Enroll in CS224N: Natural Language Processing with Deep Learning

I'm not sure why IBM programs have such high ratings on Coursera. Having enrolled in several, I found them disappointingly lazy and stale. The actual information isn't necessarily poor, but it's delivered using corporate slideshows and robotic voice-overs that kill the learning experience.

IBM courses appear to primarily serve as marketing for their cloud platform, which trails far behind industry leaders like Amazon Web Services, Google Cloud Platform, and Microsoft Azure. While these major providers also leverage courses for marketing, skills developed in AWS, GCP, and Azure result in better job opportunities and higher salaries. This isn't the case with IBM's platform. Searches on job sites like Indeed often return zero results for IBM-related skills and roles.

Given these issues, you're better off avoiding IBM courses entirely. If you go to "View more reviews" on most IBM courses, you'll see the most helpful results typically share my sentiment.

Perhaps IBM programs will improve in the future, but as of this writing, your time and money are better spent elsewhere. That said, you can decide for yourself by auditing their content for free. Go to their individual courses (not their programs), click the "Enroll" button, and click "Audit" at the bottom of the popup. Let me know what you think.

Unfortunately, not every course by DeepLearning.ai meets expectations. I don't really know who the target audience is for this one because it's simultaneously too superficial for developers and too technical for business leaders.

The audio and video quality is actually better than the Databricks LLM course I recommended above, but this course lacks advanced details on how the concepts work. It's a departure from the clear and intuitive teaching style for which Andrew Ng and DeepLearning.ai are known.

Moreover, the course lacks meaningful projects and the weekly lab assignments are prefilled and don't require any input or critical thinking. Maybe this course is simply meant to make you feel like you know how things work.

As with most Coursera courses, you can audit this one for free and see for yourself. I would avoid upgrading for the certificate and labs, unless you specifically want their AWS code examples.

The powerful AI you see in shows and movies—Westworld, Ex Machina, Ironman's Jarvis, etc.—represents Artificial General Intelligence (AGI), a sophisticated form of AI capable of performing any intellectual task a human can do. While AGI remains a theoretical concept, researchers and engineers progress daily towards its realization. 

All of the AI we use today is categorized as Artificial Narrow Intelligence (ANI), which refers to systems focused on specific tasks. Examples include autonomous driving, playing Starcraft, debating humans, and even systems like ChatGPT.

Online AI courses are primarily concerned with ANI, so if your interests lie in AGI, consider pursuing a Ph.D. from a top-tier university. Otherwise, we can use the theoretical components of AGI to unveil possible learning paths, guiding us on what to learn.

In Artificial Intelligence: A Modern Approach, the AI textbook preferred by 3000+ universities, the authors describe the various narrow intelligence subtopics that could produce an AGI system when combined:

• natural language processing to communicate successfully in a human language;
• knowledge representation to store what it knows or hears;
• automated reasoning to answer questions and to draw new conclusions;
• machine learning to adapt to new circumstances and to detect and extrapolate patterns.
• computer vision and speech recognition to perceive the world;
• robotics to manipulate objects and move about.

These categories serve as a general template for what to study online, with natural language processing (NLP), machine learning, and computer vision being the most widely taught.

Machine learning, in particular, is a critical component of many AI curricula, and there is a lot of crossover between AI and machine learning courses.

Nowadays, when people talk about AI they often mean machine learning. The Transformers that power Large Language Models are a deep learning innovation, fed with processed natural language data, and optimized using reinforcement learning (with human feedback).

To get to this level of technical understanding and execution, you need a strong foundation in a few prerequisites.

Most AI courses assume you have a working knowledge of probability, linear algebra, calculus, and programming. With this mathematics exposure, you'll find it easier to understand many AI concepts.

You don't need graduate-level knowledge, but AI is an advanced math and computer science subject, so comfort with these prerequisites is necessary.

Here are a few of my favorite (mostly free) online courses for each prereq:

• Probability: Fat Chance: Probability from the Ground Up from Harvard
• Linear Algebra: Linear Algebra 18.06 from MIT
• Calculus: Single Variable Calculus and Multivariable Calculus from MIT
• Programming: Learn Python from Codecademy or another top Python course

Solve many problems in each subject, and you'll have a good base from which to take advantage of the best AI courses.

If you are familiar with these subjects, you could streamline your learning by enrolling in an AI course and then referencing these prerequisite courses, textbooks, or YouTube when something doesn't make sense. 

Remember, you don't necessarily need a deep mathematical understanding of AI to start building something with it. The versatility of Python and the support of the open-source community provide you with access to numerous high-level AI libraries and comprehensive documentation. If you can code, you can leverage AI to create something innovative and exciting.

Each platform approaches online learning slightly differently regarding pricing and features, so I'll provide some context here in case you need help deciding.

Coursera and edX are the most similar in that they started off offering courses mainly from universities. For this reason, they are both excellent platforms for an academic approach. Both also have generous free tiers to almost all of their courses, allowing you to audit the video and written content. The catch is that you need to upgrade to a paid certificate to access to quizzes, exams, and graded assignments.

Comparing platforms themselves, Coursera has a more modern user interface than edX, so watching videos, moving through modules, and interacting in forums are more streamlined. That said, edX has outstanding educational content, and some of the best courses I've ever taken were on their platform.

Bottom line: if you want an academic approach to a subject, both Coursera and edX are good choices—just take the one with the best instructor and a syllabus that aligns with your goals.

Now, if you're looking for more practical guidance, Udemy has a massive collection of courses built by people who build things.

When I want to understand something quickly and use it in a project, Udemy is the first place I check. They rarely have free options, but the courses are usually marked down to $10, so they're relatively affordable compared to other paid courses.

Udemy's course UI has chaptered video modules with a community Q&A section under each video. Unlike other platforms, this makes it easy to ask the instructor a question directly under the content you're questioning.

The amount of effort the instructor puts into the course's content will dictate the quality of the experience, and this effort is usually rewarded with higher-than-average star ratings (>4.6 usually).

Bottom line: Udemy is a fantastic platform for quick insight on how to build things

Out of the top course platforms, Udacity is the most expensive. This is justified by additional benefits, like mentorship, career help, and more involved projects and assignments.

I see Udacity programs like mini bootcamps, where you pay extra for a full-featured online learning experience. Some programs are hit-and-miss, though, and they've had to recover from a stint of bad reviews a few years ago.

If you want a bootcamp-like experience and a financial incentive to complete a program, then Udacity is a great option. Just make sure the program you choose has positive reviews on third-party platforms.

Bottom line: Udacity is akin to a lite bootcamp in terms of pricing and features. Join their programs if you want a great education but need help staying committed.

These are the two core data-focused platforms. Both have a substantial catalog of interactive courses and programs and stay fairly up-to-date on trending topics, which is nice since they're subscription-based.

The primary difference is how they teach: DataCamp utilizes video and text, whereas Dataquest focuses solely on text. Personally, I find Dataquest more effective, but each person has their own learning style.

See our comprehensive review in our article: DataCamp vs. Dataquest

Bottom line: Choose Dataquest for highly effective text-based learning; choose DataCamp if you can't live without video content.

Learning AI from scratch might seem overwhelming, but it's important to remember that regardless of your background or level of education, you can learn anything with determination and persistence.

If you've taken one of the courses above, I'd love to hear about your experience. Alternatively, if there's a good AI course I missed, feel free to share your recommendation in the comments below.

References

1. https://www.levels.fyi/blog/ai-engineer-compensation.html ↩
2. https://news.crunchbase.com/venture/monthly-global-funding-recap-february-2024/ ↩
3. https://www.goldmansachs.com/intelligence/pages/ai-investment-forecast-to-approach-200-billion-globally-by-2025.html ↩