my feedback for Stanford AI course
~~~ The Good: 1) Video editing: This was a subtle but wonderful feature of the lectures. Specifically, editing out the lengthy seconds it takes the instructor to write out comments or equations made the viewing much more focused and engaging. I think this is an important feature because it really takes advantage of the medium and moves beyond one of the limitations of live, classroom teaching--pacing. 2) Short video clips: The short clips made it easy to consume the course in small nuggets completely up to the schedule of individual viewer. It also made it easy to review specific points…
What it does
In the maker’s words, at launch
~~~ The Good: 1) Video editing: This was a subtle but wonderful feature of the lectures. Specifically, editing out the lengthy seconds it takes the instructor to write out comments or equations made the viewing much more focused and engaging. I think this is an important feature because it really takes advantage of the medium and moves beyond one of the limitations of live, classroom teaching--pacing. 2) Short video clips: The short clips made it easy to consume the course in small nuggets completely up to the schedule of individual viewer. It also made it easy to review specific points later on. 3) Interactive quizzes: Having to stop and figure out the solution to a problem no matter how long it took was an extremely rewarding learning method. Contrast this with in-person classroom lectures in which if an instructor asks a question of the class, someone will answer and the lecture moves on even if everyone didn't understand it. If anything, I encourage there to be more quizzes of this type in future iterations of the course. 4) Office Hours: I thought the crowd-sourced office hours were a great addition to the regular curriculum. I wish the format had been in place from the beginning, but once it was implemented it was fun to vote on questions and see them addressed by the professors. ~ Areas for Future Development: 1) Include recommended readings: Either text book chapters or technical articles. A few students posted some interesting articles on reddit, including a great article by Thrun on Particle Filters, and Prof Norvig's piece "On Chomsky and the Two Cultures of Statistical Learning." Great stuff, would have loved to see more, especially articles considered to be "central" to the field. 2) Include programming assignments: To avoid the overhead of having to crunch student submitted assignments, have "Project Euler" style questions that have only one right answer, but have to be computed programmatically. The NLP assignment was a good example of this, but the questions were too small in scope and could be figured out by hand. 3) Restrict the forums to a single source: I felt as though there were three active forums in the class, (not counting some great discussions on Hacker News). By the end of the term it seemed like the three forums differentiated themselves by topic-- reddit for general discussion, ai-qus for hw/quiz questions, and the onsite forum for office hours questions. Nevertheless, I felt like I was constantly missing out on interesting threads/questions and wished there was a single forum to rule them all. 4) Send recruitment letters after the final: I congratulate all those who received emails from the Professors on being among the top 1000 and being invited for work opportunities. My only wish-- other than getting such a letter myself of course-- is that the letters be sent after the final is over. I'm guessing the Professors didn't expect the letters to become public right away, but for those who weren't at the very top it was a little disheartening to learn about that right before the final. ~~~
Does the same job
all alternatives →
- WSWatch Stanford CS Lectures Together2018 · classroomies.com · ▲360
- CLCurated List of Online Learning2020 · ▲43
Hi HN, The discussion pops up regularly about which MOOCs and courses to take so I figured I'd start a curated list of recommendations. It's by no means exhaustive (and it's just the ones I know) and is quite broad. If you had any suggestions, feel free to create a PR or Issue, or even respond to this post and I will add it. I'd love to be able to capture quality somehow, but not sure of the way to go about that just yet. https://github.com/tombasche/professional-development
Action Army: The good, the bad, the ugly of online courses2016 · ▲86And no, I'm not pitching a "6-figure" dream
- LBLearn beautifully, with content-specific AI lessons2023 · studywand.com · ▲5
Hi HN, For 4 years I've worked on the psychology of learning following a CompSci degree with an AI minor from back in 2019. I love the future of effective learning that is coming. This is a way for you to study for specific exams, lectures and videos. You upload your content, and get specific, citing flashcard exercises delivered in a fun way. Not only can you view the timestamp in the video that a question was generated from, for example, but your answers are used to inform a "Compilation stage". The Compilation stage uses your choices of answers (and wrong answers), as well as data on when…
- WMWe made a low-budget ‘how it works’ video. What do you think?2013 · ▲34
http://youtu.be/nT8xXIdwDGo
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, December 2011
the whole month →- IC
Life & fun · 2011 · itunes.apple.com
- MZ
Life & fun · 2011 · anonysize.me
- BA
Life & fun · 2011 · hacksandthoughts.posterous.com
- WJ
Life & fun · 2011 · subpug.com
- RR
Life & fun · 2011 · github.com
- THThe Honeybadger▲147
Life & fun · 2011 · gethoneybadger.com