Stanford AI Professional Program Review


February 29, 2024

In 2023 I completed the Stanford AI Professional Program to deepen my understanding of Artificial Intelligence, especially with natural language. The courses I took were great and definitely broadened my perspective of AI, but I could have obtained a lot of without paying the course fee and I don’t think the digital credential means a lot. There were some aspects that were better as part of a course, in particular the assignments and course project for Natural Language Understanding, but if you are motivated you could get the same education without the course fee.

The program requires you to complete at least 3 courses, and I picked the three that most closely aligned with my current goals of deep learning for natural language:

It’s worth noting that pre-recorded videos for each of these courses are available free on the Stanford Online YouTube Channel, for example there is a CS224U playlist from Spring 2023. The course consists of these videos, but broken into smaller segments and collected over various years, maybe including more recent content than is on YouTube but the gap is small. The lectures are extremely high quality, taught by experts in the field, and are well worth watching. The main benefits you get from the program are interacting with other students, course facilitators, and getting access to assignments (though notably the CS224U exercises are available on Github) and access to the autograder.

Interacting with others was one of the main benefits of the course, but was made difficult by being in an unusual timezone and busy. Meeting other students is a benefit of the course, and is mainly self-driven, facilitated by course Slack channels and organised coffee-catchups (which were unfortunately in the small hours of the morning for me). The course facilitators were always quick to respond, especially for issues with the assignment and autograder, and for CS224U there was a facilitator that made the time to catch up with me to talk through the course project. The Slack channels allowed free discussion and I did benefit from other students questions about the material, but in my experience there wasn’t a lot of discussion outside of the assignments. I was also lucky to have live Q&A sessions with the lecturers for all the courses (which are not guaranteed to happen) at 3 AM local time, and while I got something out of them (asking about the field of AI was immersed in ChatGPT hype) they didn’t add a lot to my overall experience. There is a lot of potential to make new relationships out of the course but you have to put the work in to make it worthwhile.

The assignments were a huge benefit to my learning, but the types of assignment and quality varied a lot by course. Natural Language Understanding had the most open ended assignments and a self-driven project that were flexible enough to challenge a range of level of experiences. The project was particularly challenging, to actually write a paper, and although my project on applying multilingual distillation to retrieval models did not succeed I learned a lot in the process (and was marked fairly). Natural language processing had assignments that were straight-forward and focussed on the mechanics of models (you didn’t need to look at the data), and I did not learn a lot from them, aside how to implement multi-headed attention, because I already knew a lot of the fundamentals. Deep multi-task and meta learning had well-designed assignments and I really only understood meta learning through doing the assignments. All the courses had an autograder that gave you immediate feedback, and allowed resubmissions. The autograder for deep multi-task and meta learning assignments was very fragile and would fail code that used a different random function (that wasn’t specified in the assignment), or used variables in a different order, but the course facilitators were very responsive and helpful in getting code the autograder would accept.

The digital credential that you get with the course is nice, but I would not assign a lot of value to it. The courses are pass/fail based on achieving some score threshold, which is very easy if you put the effort in given the generosity of the autograder and helpful course facilitators. You get a digital credential that you can share, and does signal you have put that time and effort in, but I’m not sure whether it would be recognised by employers.

The Stanford AI Professional program was very helpful for to learn about natural language understanding and deep meta-learning. In particular my highlight was the Natural Language Understanding project and assignment “bake-offs”, which would be an in-class competition on a particular dataset. If you are in a US compatible-timezone and motivated then it could be a good way to make connections with other people working in machine learning and technology. However the program is relatively expensive and you can get a lot of the value by watching the very high quality lectures for free on the Stanford Online YouTube Channel.