Archive

Github code search is awesome. I often find myself writing obscure and niche pieces of code that don’t fall under the bell curve of the internet. But with code search, I search for a specific or expected code pattern and voila.

So it’s indeed unique links associated with my Spotify account. That’s impressive strategic foresight right there.

I can’t figure out how Spotify kept track of songs I shared with people on other messaging apps. Is the URL unique and contain some kind of token? Fascinating.

Spotify has a chart now? Great. It should let me know if I have already shared a song before.

Using Claude to tailor my resume. Curious that Claude starts ending its response with “your resume is ready for submission” after 3-4 iterations. Wonder if this is baked into the global system prompt?

Throughout my PhD I despised books that attempted to describe and prepare you for what this experience would be like. Post PhD, unemployed and 2 papers still in the revise-review-resubmit limbo, and I can sympathize with the authors. You need to be in the trenches yourself.

Use the right tool for the job. I can be reasonably productive in the command line but there is a reason why GUIs dominate. It takes design, interactions and product thinking to make something worthwhile for the user.

Open source developers are philanthropists.

I saw drastic improvements in my table tennis technique in year 1. But it took another year to translate than into winning matches. I find that I need to shift my mind to a play state, focus on having fun rather than score. Sometimes the wins come naturally, most of the time I learn something new.

Found the perspective of considering AI wrt Economics refreshing. There are days when I cannot remember how I got anything done without AI. But on the other hand, I also distinctly remember swearing at the garbage Claude sometimes produces after hours of prompting.

Having the possibility to digitally capture thoughts and ideas actually backfired on me. It never gave me time to process what I was capturing. Intentionally keeping my TODO.txt only on my laptop lets ideas simmer in my subconscious. Often they turn out to be bogus, the good ones are hard to forget.

Learning a lot about gaming ATS today.

Modern LaTeX stack: tectonicand tex-fmt. No more 5GB texlive hanging about my on system.

Currently undergoing a digital detox. No more spiraling with note taking; back to the basics with a single markdown file to manage tasks & pen+paper for brainstorming. I was feeling anxious about all the note I was writing. Connections form in the brain, not on a screen.

Claude code is a game changer for me. It has brought an end to the text editor hopping syndrome I have been suffering from for the past few months. I am back in my natural habitat now, the terminal. I run (neo)vim in one tab and Claude code in another.

Actually, my TODO.txt is called 2025.md and its saved inside a directory called Life Admin. The name is inspired by the Johnny Decimal System.

Added some quick highlights for the keywords in my TODO.txt using mini.hipatterns.

I like my text in the terminal to look “chunky”. Helps me focus on the content rather than superfluous UI stuff.

The first principles of living life. Keep things close to the source. Do the work. Work with purpose. Accept the current circumstances, make the next best decision. Focus on what is in your control. Be proactive. Progress isn’t made in a day. Don’t ask how to write a book, ask how to write.

I need to exercise my body, to exercise my mind. I start my days early, (try) to work in 2-3 hours of blocks. Break the sessions with exercise (currently mix of long walks, HIIT, table tennis & weights). If I can’t focus, I stop, reset and try again. Discipline and consistency trump all else.

I have joined the TODO.txt movement. Single plaintext file (markdown for me, because I am not an animal), for each year. Divided into quarters, months, weeks and days using headers. A few keywords (TODO, NOTE, DONE, NOW) to facilitate search. Intentionally kept only on my laptop to reduce junk.

It’s called an AI assistant not an AI slave. Don’t tell AI to do the work for you. Do it yourself first, then identify steps that can be automated. Leverage the true power of LLMs (which is language, duh). Brainstorm search queries for Scholar, rewrite and rephrase text you authored. Do the work.

Using a tiling window manager has been the best productivity booster for 2025. All my frequently used apps are arranged into dedicated virtual workspaces (using pneumonic like “S” for social and “N” for notes). I have my text editor at 1 and terminal at 2. No more Cmd+Tab to …

Really enjoying Lilex as a monospaced font these days. It’s a serious contender to Source Code Pro, which has been my monospaced font of choice for nearly a decade.

What matters is willpower: What I am increasingly recognizing is that having a sophisticated system to manage your tasks does not make you productive. What matters is the willingness to execute on the tasks that are piling up, regardless of how I “feel” about working in the current state. You see, the last year I wasted …

The secret to finding great content on the internet: Is to remember the people behind the content. Often, I find something interesting, I go to the author’s website and find a treasure trove of information. Remember the people. Ultimately its the human you need to reach out to for anything.

The more I detach myself from the use of AI, the more grounded and creative I feel. I remember when I produced entire research papers with just pen, paper and a simple text editor. The more tools and gizmos and gadgets I incorporate into my workflow, the more my brain atrophies.

I wonder what the arguments against connected note taking are? The fundamental idea behind Logseq, Roam and Obsidian is essentially regex patterns and search. I think what would be more interesting is to identify potential entities automatically. Maybe ML can do this already?

I do like the declarative style of configuration though, it definitely fixed my system config itch. I have stopped tweaking my config every 5 minutes since building the entire system using sudo nix-darwin switch is just the right amount of resistance to make me think for a minute.

I am unsure about using Nix for system and package management. I think the benefits did not outweigh the cost of setting it up in the end. Maybe if the Nix language itself was not so obscure? Also it feels that things are not as transparent as with homebrew and dotfiles.

Wondering what’s the best way to group related entities together. I am debating between using tags and folders. With tags, I get to keep a flat directory structure inside obsidian. I don’t see a clear winner here because with the search features, I can narrow down the notes I want either way.

AI cannot innovate. It may be able to write code, but it can’t figure out what to write code for i.e., it cannot identify problems to solve. A deeper reason for that is because it cannot talk to humans, accumulate knowledge and be proactive.

DHH just released Omarchy 2.0 as a Linux ISO. Boy does be know how to market his products. People have been ricing Linux for decades and along comes this guy, puts his dotfiles in GitHub, spreads the word amongst his followers and lift off. Amazing.

Obsidian follows standardized format of storing metadata related to notes using the YAML frontmatter. I find this design decision better compared to Logseq which introduced its own format.

Vivaldi had a horrible user interface. It feels like a bunch of developers got in a room together and built it. Nothing against doing that, just did not appeal to me as a polished product. Back on Arc.

My mind feels unlocked. Its only been a day since I started using Obsidian and I feel like the movie Limitless. When I let my mind ponder, jump and hop without restrictions I am discovering so many interesting and wonderful things.

What distinguishes the Knowledge Graph I am creating within obsidian, from say wikipedia is that the entities here represent my interpretation and are related to how I perceive and use the information.

I think a limitation of obsidian’s alias plugin is that it does exact word searches. I think a more powerful alternative would be allow cycling between a regular search and a regex search. This is akin to emac’s DWIM functionality.

What is the point of explicitly creating links between notes when the unlinked mentions section automatically does a good job of showing connections between notes? It seems to be that the 8020 approach here would be to thoughtfully create these entity notes.

Wondering if I can view notes that were created/modified in a given date range. This would allow me to get an overview of what I have been thinking about in a given week/month (e.g., during my monthly reviews). I can do this using the bases feature which allows you to filter based on modified time.

A nice strategy for expanding my knowledge has been to find episodes of people I discover on the internet, on the Fridman podcast. There is something engaging and intimate about listening to the human being. Often, I find that the persona they portray on the internet is not the same on the show.

The backlinking (unlinked mentions) is just a grep for the filename of the current file in all other files.

I just figured out how I would implement the linking functionality of Obsidian using unix tools. The forward linking (linked mentions) is easy enough, look for the regex pattern \[\[\(.*\)\]\] which captures the text inside wiki links and search for those file names using find.

Some notes in obsidian actually represent high level topics or entities that I frequently write about. I find that this is similar to Entity Linking which I learned about in one of my masters courses.

Writing on paper feels more creative: I don’t know if this is because of familiarity? Or because it feels more intimate to write on paper knowing others can’t reach my handwriting? Another thing is this realization that “oh I am looking at a screen again”. And that puts a mental block when writing on my phone or laptop. It’s much easier …