I've been a university professor for over 20 years, helped found an entirely new school of biomedical informatics, directed academic programs as curriculum committee chair, and then as associate dean for academic affairs. I'm now starting a new program at a different university where I am trying, once again, to address major difficulties with the present university business model as it relates to teaching. Through this time I have found it very difficult to maintain course quality. There are numerous reasons for this: preparing quality course content and assessments is difficult and time-consuming, yet faculty are often penalized for spending much time on teaching. Students are often unprepared and unmotivated, so maintaining rigor is nearly impossible, because it means too much one-on-one time with students. Faculty are also being asked to increase class sizes and ensure that more and more students pass. Given all of these factors, the easiest approach is to spend as little time as possible on classes, while making them as easy as possible to ensure that students get through and don't complain too much. Good students will complain about this approach, but many will be happy with the credits. However, the approach is clearly a disservice to all.

In contrast, I feel that Udacity.com is a disruptive technology/service for higher education, similar to Kahn Academy for K-12. Yes, there are some problems with the platform and approach, but overall I feel that the Intro to AI, Ng's machine learning course, and the new Udacity.com courses are among the best courses I've ever taken or seen. In fact, I feel that for this material, they are far better than face-to-face classes.

There are several reasons for this:

(1) High quality content and assessments with excellent instructors.
(2) The emphasis on short videos, motivated by grounded examples, followed by meaningful assessments. Hour long talking head lectures are BORING and pedagogically poor.
(3) The use of mastery learning during the lectures and most of the assessments. I can't ask a question in class and wait for everyone to figure the answer out, but Udacity can do this online. Mastery learning is one of the most effective methods of learning, but it doesn't fit the industrial-age model of classroom learning.
(4) Courses and students live or die based on the quality of the course and quality and motivation of the students. The end result is that rigor can be high so that motivated, prepared students can actually learn something.
(5) Anyone can take a class, but no one has to. The result is that most students taking a class are taking it because they really want to learn. If you just want a job, there are plenty of brick and mortar (and traditional online) universities who are happy to take your money in return for a degree: very little learning required.
(6) High quality discussion lists. My experience is that in most face-to-face and small distance learning classes, discussion lists are used only if posting is part of the grade. In contrast, since many of the students at Udacity are self-motivated to take a class, the discussion lists are lively, informative, and helpful. The large number of students helps a lot here.

So overall, I'm impressed with this approach. So much so, that I have already adopted some of the style for my own classes. However, students have different goals and learning styles. Likewise, not all material will lend itself to this approach. As a result, students will continue to have a variety of educational choices, but my gut feeling is that those choices are going to have to improve in quality as a result of Udacity.com and other similar efforts. This is a good thing, because the current higher ed business model and costs are unsustainable.

In general I don't agree with your point of view... I've already spoken my mind elsewhere, and I won't repeat myself here. I'll just point out that the word "beta" should suggest you that udacity has been working and experimenting new things. Some things work (free learning) others maybe not yet (recruiting), and they'll surely learn and improve, but to me this does not diminish the value of the things I've learnt.

There is one thing I don't understand in your words: you say you've spent a significant amount of time on it, so it mustn't have been all too easy, still you say you learnt too little... there seems to be some contradiction. I think the workload was fair: most people here have a full time job and/or a family and can't committ to full-time study for months, but it's just my idea.

And no, I won't be downvoting your question, it's just your point of view. And congratulations for completing the class.

I think the news that scoring 100% may result in resume forwarding has created some sort of panic among students. I have a nice job and the course did require that I devote my free time to it - but otherwise how else I would learn? Learning from master has its own value and I would not compromise on that. I did rant about grading issues in the beginning but I am really past that. Sebastian did cover quite a few important topics and introduced them in a style that made it look easy. Just few years back these were research topics (they still are) and it needs patience and devotion to dig deeper. There are lot of unanswered questions but I would treat that as a positive outcome of this course. We must always be left with more questions. Even with the issues, it is so much better than the regular class. We have to admit that the staff is quite responsive and willing to work with the students - how many schools are like this?

So why would somebody want to attend Udacity courses? To get a better
job? Only at companies ready to hire people from a crowd who are not
really distinguishable in their skills.

I have found that anywhere worth working will not put too much emphasis on marks; they tend to emphasize what you can do, what you can demonstrate, instead. Udacity gave us lots of stuff to demonstrate, and enough knowledge to go out and start doing things/making projects of some sort (like the TORCS-based simulator programming some folks are doing).

To learn something new? Yes, the course does really make interesting
insights and teaches new stuff in a funny way. But that is the thing
you get for free in other places too, like iTunes University, Khan
Academy, etc.

The other places I have tried don't teach material like Programming Probabilistic Robotic Cars, and even if they did, they wouldn't present the material as well as Udacity has. I'm taking distance courses from MIT and Stanford as well, and the material is great, but their teaching methodology is is not as good. They do longer lectures with very few quizzes, and MIT never explains the answers to the quizzes. They do way less programming in the quizzes, etc. Last time I checked, Khan was just a bunch of video lectures (that was a while ago, maybe they have improved their interactivity since then....but I doubt they are offering in-browser programming quizzes).

It sounds like you just found the material too easy, that you didn't learn enough. That is fair, but I wouldn't blame Udacity for that. I am sure eventaully they will offer some 400-level courses :)

I am quite happy to move away from any system that equates learning with marks. I prefer systems that equate learning with learning, which Udacity seems to do better than any of their competitors. Little touches, like associating concepts with practical applications (their Intro to Programming course, cleverly presented as "Programming a Search Engine", is a perfect example), keying graphics over top of the drawing hand, short video interviews with industry folks or showing things like the Google car in action, super-frequent quizzes (many done as programming tasks!) that have the answer well explained in a walk-through afterwards...these things all add up to differentiate Udacity from their competitors. Everything is in beta, but they still seem to be far ahead of everyone else. Kudos to Udacity for getting it right!

I don't agree with this view.

My criteria for evaluating this class is whether I learned a lot. Personally I feel like I did learn alot.

Since the course didn't promise to land me a job or do anything that would help land me a job, I don't include that as a criteria for evaluation.

I don't agree with this view either. I don't really agree with anyone who posts about how 'easy' this course was, because I think that is an arrogant position. This is a relative thing depending on the person's previous experience and I'm sure there are lots of people who did this course and did find it challenging and did not get perfect scores all the time, even on the exam.

I have to admit that the tone of this post immediately turned me off. It sounded too much like ‘this course was a waste because the grading policy undermined the job benefit that I was expecting to get’ (even though there was no promise of a job benefit). Furthermore, I find it illogical to complain about things like being able to make multiple guesses on the final. The final was an un-proctored, untimed final. If multiple guesses weren’t allowed a student could either look up the answer or get the answer from someone else anyway.

However, upon further reflection, I think the tone is unfortunate because there is a substantive question buried in there which is: what grading policy would have the maximum career benefit for the participants. I suspect that Prof Thrun is considering this issue.

It may be that a different grading policy with proctored exams, such that participants could include Udacity CS373 on their resume as an official course, would produce a greater career benefit for participants. I think it’s fair to suggest/ask Udacity if they are/would consider a different grading policy such that the courses are official.
That’s a topic worthy of discussion. There are certainly pros / cons to each approach. Prof Thrun may be happy to teach 20,000 students, identify the top 200 by whatever means he has, and refer only those students, which is his prerogative. Or there may be a change to the grading policy that would increase the benefit to the overall group (and he could still find the top 200).

My answer for the Udacity business case:

Outsource the final verification of skill to employers by providing an additional verification service.

Example: I take CS373 and get 100% on the final (I'm AMAAAZING!). I advertise this on my CV. Several employers take an interest. Those employers subscribe to a Udacity-provided verification service. The service is essentially another version of the final, sans answer-checking feature. The employer provides 1/2 the security needed (proctor, no Google), and Udacity provides the second 1/2 - large enough pool of questions, sufficient randomization, etc, to maintain test integrity.

Udacity could even make money off this. Let's say the verification service is free...unless you hire the person verified. Then the employer owes Udacity $500/class verified for that person.

This way Udacity proper could focus on what it seems best at - providing a great environment for learning, at no cost. Those who only want to learn - no problem! Those who want to show they can separate themselves from others while in pursuit of a job can ask the prospective employer to test them with Udacity's DEFCON 1-grade verification of CS373, where like the first question is, "Write a Python program that accounts for the precession of the earth in the gyroscopes of a 9-DOF IMU used in a Kalman Filter" (or whatever).

It seems things are already headed this way, with the likes of Apple, Google, MS rigorously testing incoming job-seekers, and no longer just trusting a university-blessed B.S. degree (pun intended).

I'm afraid I have to agree. I almost peed my pants when I first heard about this course after finishing ai-class. After almost getting a perfect score in ai-class (missed only one question on the final -- went from tied for 1st to top 25%, wtf), I felt that I'd have a chance to redeem myself, and it was on the ai-class topic I was most interested in. However, after spending the first half of the class on filters, I slowly became disillusioned. This class is just way too easy. I really racked my brains trying to come up with the answers on some of the harder questions in the ai-class homeworks, but Udacity's homeworks are almost boring. After hearing about the change in the grading policy, I just stopped doing the homework altogether knowing I would get a 100% on the final. With two jobs and full-time graduate school, my time is stretched too thin as it is.

If everyone gets a 100% on the final (which shouldn't be too hard), then how can the excellent students be distinguished from the average? Isn't that Udacity's business model? To make the performance of Udacity's students available to companies? I'm all for increasing the availability of education, but, for Udacity to be successful, it needs to cater to the smartest people, those that companies would pay big bucks to recruit. Udacity should definitely be available to everyone, but it needs to be harder.

I agree yet disagree as well, I think the new grading policy completely makes the homework meaningless in terms of grading, they were great for learning and testing yourself. As for the final, the goal of the final was to be a learning experience and not just another test and I think they succeeded at that. With the influx of people wanting to register after the course has started (especially being the first 2 courses launched) they had little choice but to revise the policy.

As for the difficulty of the course I having no previous background experience in AI or Robotics found it pretty challenging at times yet incredibly interesting as well. If they made the course significantly harder to begin with then people like me would not be able to do it by any means. I think it's a great gateway to the field of robotics and AI, from this course they can provide more advanced courses on the subject matter knowing that this course is in place to get people to that level.

If you have previous experience in these topics then this course was probably just a refresher and I hope it was a good one at that. After all in the pre-req did they mention you must have previous experience in AI and robotics programming and theory, no I believe they did not. So if you found this course too easy Im sorry but please don't devalue the course for those of us who have just learned to this bike. :)