Maybe 5-or-so years ago I found Rob Miles’ YT channel and was immediately very interested, but for the longest time I didn’t have anyone in my social circle to talk about this and from Rob’s videos (esp. on computerphile) it seemed frighteningly technical and mathy.
Whilst I appreciate maths, I’ve not paid attention to classes prior to starting uni, which means I have enormous holes in my foundations. Perhaps I could catch up, but despite my appreciations for the subject, I have no strong drive towards it, which always made me very hesitant to engage w AIS as a career option since it seemed like I have no chance to meaningfully contribute. Or at least like I’d have to put years of work in even just to get to a level to where others are just starting their journey. I was probably also just scared it would be too technical and that I would beat myself up about not being able to follow.
Aside from the increasing urgency and shortening timelines, I’m now also socially more invested to the point where instead of getting pushed away from it1, I’m motivated to give this a try. To actually get into subject and see how far I get. To determine once-and-for-all whether this is smth I could meaningfully contribute to and if so in what way, or whether I should do smth else. Either way, I’ll get to learn many interesting new things and participate in the discussions my friends are having2.
My current plan
(under constant revision)
- EA Intro Program
- doesn’t focus on AIS but generally quite recommendable
- bluedot’s Intro to AI course
- unsure what it was called, seems to have been taken down
- was a 2h self-paced course
- ENAIS Collab course
- this was a fun 4 week course, we went through the AI Safety Atlas ch. 1-4 and had an extra session where we discussed more stuff and chat a bit more
- AI Safety Atlas
- I’m just missing ch. 8 atm
- the new audio narration is really good (e.g. it states that there’s a picture here and describes it (I assume read its alt text?) rather than saying “picture_5.jpeg”)
- bluedot’s AGI Strategy course
- got rejected in my first application, but since then finished most of the readings and some of the exercises by myself, which I can recommend. Curious as to what the cohorts add to this
- 3B1B Neural Networks playlist
- this was really fun and informative
- StatQuest: Neural Networks, Deep Learning, and AI
- I didn’t particularly enjoy it, but I could easily follow at 2x and it added some things I didn’t get from the 3B1B course, which overall made it worth-while
- Karpathy’s Neural Networks course (around 15h on building neural networks, from scratch, in code. Will see how this goes)
- ML4Good?
- Math Academy
Projects
- AI Safety Atlas quiz:
- whilst I like the Atlas, I’m unsure how much of it I really got, how much stuck w me, how much I glanced over and didn’t really take in
- having a quiz for each chapter would be great
- local prototype w basic quiz (multiple choice)
- public prototype w basic quiz
- public prototype w advanced quiz (freeform text, etc.)
- AI Safety Atlas podcast:
- whilst the audio version turned out well, it’s really hard to switch and I’d prefer not having to commit to either listen to or read the entire (sometimes 2.5h long) chapter
- having the option to go back and forth between text and audio would be great
- perhaps a podcast with each section (rather than chapter) being an episode, each w the original text attached (perhaps incl. time stamps, if not auto-scrolling to the right position)
- find out whether this is possible
- test w one shorter chapter
- showcase to CeSIA