Sebastian Thrun wants to teach you how to program a self-driving robotic car(spectrum.ieee.org)
spectrum.ieee.org
Sebastian Thrun wants to teach you how to program a self-driving robotic car
http://spectrum.ieee.org/automaton/robotics/artificial-intelligence/sebastian-thrun-will-teach-you-how-to-build-your-own-self-driving-car-for-free
5 comments
I have a lot of respect for Thrun, but the online AI class was such an unmitigated disaster. I feel like he and Norvig really needed to work with someone experienced with organizing and producing that kind of project rather than just trying to wing it themselves. Maybe it was just a rough first attempt, but I would be extremely reluctant to take part in another one of his online teaching efforts, or to recommend them to others interested in learning.
Having successfully completed the AI class -- I have mixed feelings about, but definitely wouldn't call it a disaster. I enjoyed the experience and learned quite a bit.
For me, the main problem was the course was trying to attract audience without any common background. So lectures ended up very superficial unless the subject is very self contained.
For example, Novig gave a few very nice lectures introducing game theory and various search based solutions. But then they jump to a very superficial description of particle filters.
On the one hand, they dedicate time to explaining Bayes theorem (something that is taught in any basic probability class), and on the other hand, they describe advanced topics like particle filters and, perhaps, Markov Decision Processes very briefly.
Hopefully, Udacity will have, in the future, courses that build one on another. So students will have more uniform background and the instructors will have something to rely on.
Good luck, Professor Thurn!
For me, the main problem was the course was trying to attract audience without any common background. So lectures ended up very superficial unless the subject is very self contained.
For example, Novig gave a few very nice lectures introducing game theory and various search based solutions. But then they jump to a very superficial description of particle filters.
On the one hand, they dedicate time to explaining Bayes theorem (something that is taught in any basic probability class), and on the other hand, they describe advanced topics like particle filters and, perhaps, Markov Decision Processes very briefly.
Hopefully, Udacity will have, in the future, courses that build one on another. So students will have more uniform background and the instructors will have something to rely on.
Good luck, Professor Thurn!
I suspect that Bayes reasoning is much more complicated than particle filters. The particle filters seemed to be a simple but powerful concept.
What went wrong? I ask because I completed the first unit of Robotics AI on Udacity yesterday, and I had a big fat grin on my face for having so much fun programming. Aside from a few minor bugs (which were responded to promptly on the forums), no technical issues whatsoever.
Well, first off, there was the fact that the "whiteboard" consisted of them writing on notebook paper. Had this been the only shortcoming, it would have been fine, but seriously, there are so many better ways to present something on a computer screen than writing in a notebook under a camera.
But what really got under my skin and ultimately caused me to give up on the program were the errors in the videos, and how they were dealt with – namely, by putting some correction text underneath the video on the website. This was annoying and frustrating to try to learn with even for simple errors. But it was ridiculous when the error was a single character change within three lines packed with newly introduced notation. The minimal correct way to handle this would have been simply to overlay a YouTube annotation over the incorrect text. Those would have shown up in full screen mode, and they wouldn't have required students to simultaneously mentally amend what they were watching while juggling new concepts in their heads.
I was also not thrilled when newly introduced terminology got thrown around imprecisely in some of the videos, leaving the viewer wondering what these terms actually mean, and what the teacher was actually trying to say.
I have no doubt that Thrun and Norvig are great computer scientists. But a great scientist and a good teacher are not often found in the same person. For that person to also be a good video director would have to be a miracle.
But what really got under my skin and ultimately caused me to give up on the program were the errors in the videos, and how they were dealt with – namely, by putting some correction text underneath the video on the website. This was annoying and frustrating to try to learn with even for simple errors. But it was ridiculous when the error was a single character change within three lines packed with newly introduced notation. The minimal correct way to handle this would have been simply to overlay a YouTube annotation over the incorrect text. Those would have shown up in full screen mode, and they wouldn't have required students to simultaneously mentally amend what they were watching while juggling new concepts in their heads.
I was also not thrilled when newly introduced terminology got thrown around imprecisely in some of the videos, leaving the viewer wondering what these terms actually mean, and what the teacher was actually trying to say.
I have no doubt that Thrun and Norvig are great computer scientists. But a great scientist and a good teacher are not often found in the same person. For that person to also be a good video director would have to be a miracle.
Sebastian is using one of those electronic pens to write in this course. I'm not sure how exactly this works, but his hand is transparent in such a way that it never blocks the text (I didn't notice this immediately, it all seemed very natural). I watched the first few videos of the Fall AI class--this is far better than that approach.
I'm not sure how they'll deal with errors--too soon to tell.
So far Sebastian has been very good about not using jargon without introducing it--in fact he's been extremely empathetic in this regard. Moreover, so far the format of the course has changed to focusing on solving the specific problem (like robot localization) and then afterwards introducing the formal ideas, like Bayes' theorem. Its much better than the videos I saw from the first AI course.
I think you're right that sometimes doers are not great teachers, but I am convinced that you should give Thrun another shot.
I'm not sure how they'll deal with errors--too soon to tell.
So far Sebastian has been very good about not using jargon without introducing it--in fact he's been extremely empathetic in this regard. Moreover, so far the format of the course has changed to focusing on solving the specific problem (like robot localization) and then afterwards introducing the formal ideas, like Bayes' theorem. Its much better than the videos I saw from the first AI course.
I think you're right that sometimes doers are not great teachers, but I am convinced that you should give Thrun another shot.
I took ai-class as well and was actually really disappointed in Norvig's teaching style (considering I generally love what he writes). I thought Thrun on the other hand was great, so I have a lot of hope for the udacity courses. My suspicion is that each instructor had different levels of commitment to the project, and since both are extremely busy this affected the presentation.
My biggest issue with ai-class was that in almost all the exams and homework there were quite a few very poorly phrased questions. In a human-graded exam/hw this isn't so bad since the ability to "show your work" usually allows you to misunderstand a question and demonstrate that you do understand the material. In general I though very little effort was put into creating assignments that worked well for an online environment (ml-class did this perfectly).
My biggest issue with ai-class was that in almost all the exams and homework there were quite a few very poorly phrased questions. In a human-graded exam/hw this isn't so bad since the ability to "show your work" usually allows you to misunderstand a question and demonstrate that you do understand the material. In general I though very little effort was put into creating assignments that worked well for an online environment (ml-class did this perfectly).
> Well, first off, there was the fact that the "whiteboard" consisted of them writing on notebook paper.
What's wrong with this? Handwriting looks way better on notebook paper than on a tablet or on an actual whiteboard.
The only problem with the AI courses was lack of official programming exercises and, yes, some imprecisions here and there that were not hard to resolve by checking the forums, where they were inevitably brought up and fixed.
What's wrong with this? Handwriting looks way better on notebook paper than on a tablet or on an actual whiteboard.
The only problem with the AI courses was lack of official programming exercises and, yes, some imprecisions here and there that were not hard to resolve by checking the forums, where they were inevitably brought up and fixed.
I took the AI Class, and I have just finished first unit of Udacity's CS373. So far I'm impressed; they seem to have taken the lessons learnt from AI Class on board and made a very usable system.
The presentation of the content has been much improved. The "whiteboard" is now electronic, and it is still visible though the person's hand. Furthermore, I haven't noticed any errors yet - they seem to be much better at avoiding them.
The biggest improvement is that some of the quizzes now involve programming some of the concepts in Python, which IMO works really well as a teaching tool.
There is still a slight problem that people are starting at different levels, so some fairly basic concepts need to be revisited, but I imagine that will improve as Udacity offer a bigger range of courses.
The presentation of the content has been much improved. The "whiteboard" is now electronic, and it is still visible though the person's hand. Furthermore, I haven't noticed any errors yet - they seem to be much better at avoiding them.
The biggest improvement is that some of the quizzes now involve programming some of the concepts in Python, which IMO works really well as a teaching tool.
There is still a slight problem that people are starting at different levels, so some fairly basic concepts need to be revisited, but I imagine that will improve as Udacity offer a bigger range of courses.
Interesting. I'll have to check that out then. If they've really learned from the experience, perhaps we can write AI Class off as a prototype.
I really enjoyed it, although I hate learning through videos. It is way too slow. At least ML Class had the option to speed up the video.
But the content was fine (the bug issue did not appear that often, did it?). The sad fact is that I've had books with that information sitting at home for months, but never got round to reading it. The lectures with their weekly deadlines at least made go through with it.
But the content was fine (the bug issue did not appear that often, did it?). The sad fact is that I've had books with that information sitting at home for months, but never got round to reading it. The lectures with their weekly deadlines at least made go through with it.
The video lectures for ai-class appear to have been awful and a waste of time. Luckily I have not wasted any time with that, and skipped straight to the questions instead. Which anyway I did mighty fine with a decent grade, thus highlighting the not-so-brilliant level of the material presented. All this despite the questions being confusing, badly formulated, and sometimes, plain wrong. I am following the new "university-level" (NOT) classes with the same attitude.
So you haven't looked for 5 minutes at the new free course, but are happy to waste your time belittling his efforts on this thread.
There's no way I would be able to form a reasonable opinion about an online course given 5 minutes. I had no intention of "belittling" anyone's efforts. A large part of my purpose in posting was to spark a discussion and possibly find out if those efforts have improved. From what people have said, it sounds like they have. So it was by no means a waste of my time to post my comment.
I don't know why you decided to be a jerk about it.
I don't know why you decided to be a jerk about it.
Instead of checking out the course and then commenting on your impressions you troll the thread so others can spend their time rebating your unsubstantiated, entitled whining. "Unmitigated disaster" is not belittling; it's plain badmouthing. Your messages reveal your attitude more than any qualities of the course, and thus are wasting everybody else's time.
That's how I felt too. ML-Class was an order of magnitude more "complete" feeling.
A clarification, the homework is NOT due later today. The FAQ said homework was given on Monday and due on Wednesday, but it meant next Wednesday[1]. The homework hasn't been released yet[2].
[1]http://www.udacity-forums.com/cs373/questions/276/homework-d... [2] http://www.udacity-forums.com/cs373/questions/129/has-the-ho...
[1]http://www.udacity-forums.com/cs373/questions/276/homework-d... [2] http://www.udacity-forums.com/cs373/questions/129/has-the-ho...
The strong statement from the mentioned robot.cc:
>I am against education that is only available to the top 1% of all students. I am against tens of thousands of dollars of tuition expenses. I am against the imbalance that the present system brings to the world. I want to empower the 99%. I want to democratize education. Education should be free. Accessible for all, everywhere, and any time.
When this becomes a bit bigger and mainstream, a lot of people will be interested in that space and will be ready (and happy) to pay for it, I guess. And this need not necessarily be the students: revenue could come from course sponsors (like course champions at OCW, http://ocw.mit.edu/donate/ocw-course-champions-program) or institutions who build own offers around and based on certain courses.
>I am against education that is only available to the top 1% of all students. I am against tens of thousands of dollars of tuition expenses. I am against the imbalance that the present system brings to the world. I want to empower the 99%. I want to democratize education. Education should be free. Accessible for all, everywhere, and any time.
When this becomes a bit bigger and mainstream, a lot of people will be interested in that space and will be ready (and happy) to pay for it, I guess. And this need not necessarily be the students: revenue could come from course sponsors (like course champions at OCW, http://ocw.mit.edu/donate/ocw-course-champions-program) or institutions who build own offers around and based on certain courses.
In an interview, David Evans, who teaches the search engine class, said Udacity's business model will be recruiters paying for access to the top students.
Can you find that interview? I've been wondering how Udacity and Coursera planned to stay in business.
Does anyone know how to pronounce the name? I see ooh-duh-CITY, but I presume that they were trying for you-DAH-city or maybe OOH-dah-city...? (like University + Audacity).
Regardless they picked a hard one to figure out.
Regardless they picked a hard one to figure out.
I think it is probably you-DAH-city
I love ASCII-IPA. But I'm hopeless at remembering the pronunciation.
It'd be great to have a simple thing for pronunciation; maybe a plug-in that lets me highlight and right click ascii ipa or even unicode ipa.
(http://alt-usage-english.org/ipa/ascii_ipa_combined.shtml)
It'd be great to have a simple thing for pronunciation; maybe a plug-in that lets me highlight and right click ascii ipa or even unicode ipa.
(http://alt-usage-english.org/ipa/ascii_ipa_combined.shtml)