2025-04-14

• I’d like to publish my notes on the books I read as I make them. Ends up being a little hard to see the proper format. But I feel like there should be some sort of clean mapping between markdown and what gets produced as Anki flashcards. I currently take notes in a nested list format (topic is one level of nesting, questions are next level down, answers are one more level after that). This is nice for encoding the structure of information as a sort of tree. But it has downsides around writing notes on mobile in Obsidian (multiple levels of nesting takes up limited horizontal space on mobile, so the “answer” portion of a card ends up only having like 2 words per line, sad) and the natural display of this as markdown is similarly not very nice. I’d prefer a flatter hierarchy for showing this info in a website format, where indentation is not the primary marker of information structure, and instead you have a more “document native” system where a topic has a big heading, a question is a heading one level down, and answers are a regular md paragraph/block directly after the heading. There’s still some nuance around advanced things like “how do you have an image as part of the front/question of a card” where you’d maybe need extra structure, but overall this should work for basic cards. The ultimate dream is to have a system where you can take notes in a natural format (e.g. some basic structure on top of markdown) and then translate those notes into cards/published web pages/some format that can be consumed or used by LLMs/etc. But I’m still kind of far from that.
  
• some quick notes on what I think of as being “close to ideal” for a learning process these days. It won’t account for learning all types of information, but should be standard for a lot of knowledge-work-related things. First, you read a thing. Or maybe you watch a video of the thing. The output here is a set of highlighted sections with interesting/important information. Next, highlights are reviewed. You don’t know at this point how important the information will be, so you tolerate a low retention for things that don’t get past this point (e.g. fuzzy retention where you are aware the information exists and kind of how it fits together, but couldn’t reconstruct it from memory/first principles without re-engaging directly with the material). For sources/highlights that are identified as “sufficiently important or interesting” (usually bc they will come up again and again in important or useful situations) the highlights serve as a jumping off point for writing more detailed notes. The notes take the form of Anki flashcards, distilling key foundational facts from the source material, including things like defining key terms (ofc) and also relating ideas to other flashcards and similar “higher level” questions. There can also be room here for longer-form writing that is structured to explain the concepts to an unfamiliar person, with an emphasis/clear section focusing on what is important/interesting about the information (how is it used in real life, why would a interested novice benefit from paying attention to/learning this thing?). Those cards are added to long-term review, with a higher retention goal (80-90%). And then after that, if you are still interested in going deeper, you will need to actually get your hands dirty and physically/directly work with tools that involve applying the knowledge. So, in computer science terms, this would be something like actually implementing a data structure you have studied, or actually building a system that relies on exercising the knowledge (e.g. a basic web server after studying http). The real-world practice is the highest level of engagement. It won’t be something you do for all the things you pick up to read at the beginning. Probably it will be a handful of things (e.g. stuff related to your job or long-term hobbies). But I think this general tie ring structure is pretty good for balancing learning new things in various areas against the time constraints of doing things like creating/reviewing cards and also actually using the knowledge from the cards in the real world. These days, there is a lot of friction in going from reading to reviewing highlights and a decent amount of unnecessary friction in creating well-structured cards. The review experience is fairly optimized. And there is room for reducing friction around figuring out how to go from reviewing cards to engaging with the information in real world projects. LLMs have lots of potential to reduce the various friction points substantially.
  
• Had a lovely, relaxing day today in Bulgaria. We’ve been running around to tourist sites pretty aggressively over the previous week. Nice to have a day where we lounge around. I like this structure generally, and am reminded that I felt similarly during my trip with Rob to Indonesia, where we were all “go go go” for a couple weeks and then ended the trip by relaxing on a tropical island for a few days. Nice to have a “push” period and a “rest” period on a vacation.

---

---

---