The issue isn’t cognitive overhead, it’s not having rituals to review and refine your thoughts. Everyone has to jot down ideas from time to time, but if you never take time to stop, review, and organize your thoughts then sure it’ll feel like a lot of cognitive overhead.
He also managed to do quite a lot of other things: https://en.wikipedia.org/wiki/Andrej_Karpathy
It is called the append and review note, so I think the blog author engages with your point and agrees with it?
1. Editable description 2. Comments
By the way, here's my note-taking system: https://renormalize.substack.com/p/my-markdown-project-manag...
All joking aside, append-and-review does seem like a nice pattern for maintaining attention on a big heap of odds and ends, which is probably useful for a researcher like Andrej Karpathy.
If I happen to indeed forget, I’m one grep away from finding what I wrote about the topic based on some vague keyword.
But what people seem to find is, if a system requires a lot of work and doesn't show any benefits, they give it up pretty fast. Which is why a super simple system like TFA's is probably the only sustainable thing if you just want to remember "stuff" you hope will be useful later.
The nice side effect is that other than chatting with agents it solves the issue of getting sucked into feeds as everything external is a single feed curated by my cluster of ai agents.
Its basically an OS for creators.
I've developed a pretty unqiue approach to naturalistic non-arbitrary universally binding morality that has fixed my Ai alignment issues (without being able to retrain their model off their weird utilitarianism), but I'm not highly motivated to share it, it'll get around eventually if humanity doesn't implode.
Checkout the video: https://youtu.be/CpcsOiETgxA
[1] https://github.com/PratikDeoghare/brashtag
Apologies for low quality of video and code. :)
-----------
Example file:
```
[x*x for x in range(10)]
```
#out{}
Now if notebook program is watching the file then it will send the code block to jupyter server and write results to `#out{}` "bag". And file will look like this. ```
[x*x for x in range(10)]
```
#out{
````````````````````````````````````````````
[0, 1, 4, 9, 16, 25, 36, 49, 64, 81]
````````````````````````````````````````````}
I may keep separate append-and-review topics per major area (work, personal, cooking) but that’s about it.
Usually in form of an outline / list, append in the front, and with deeply nested sub-points, as I “discuss with myself in writing”.
(I append, the author's really prepending. Anyway...)
When the note gets too long, I cut and paste it to what I call the big note: a 127000-line, 4.9 MB text file I've been maintaining for 14 years.
Trivially searchable, can get context from neighboring notes (What else was happening around this time?), and easily parsable when necessary.
my dude you’re literally on Hacker News right now
It’s also much more similar to how you would take notes on paper or in a notebook.
I just wish some note taking app would have a setting that allows me to open a note at the bottom instead of the top!
redhale•9h ago
[0] https://heynote.com/