"Like many rites of passage, you’ll probably find it looks a little smaller, a little less daunting when it’s behind you than when it loomed ahead."

books im working through 1 intro to info retrieval: ideas on how to index information, common failures, and search 2 crafting interpreters: how do you build a scanner, parser, and runtime? 3 practical electronics for inventors: electronics is about reading information from the world and controlling a device with a microcontroller 4 programming massively parallel processors: ideas on how to do things like bfs in parallel using gpus 5 Nicomachean ethics: what's virtue, what's a good life mostly educational but you never know where your next idea will come from

The most frivolous product you can build is AI tools for onlyfans creators

With AI you can make the internet a time series

design should be rebranded to "aesthetics department"

Learning about interpreters and how to design a language from scratch. - software is translation layers all the way down - wonder if you could design a language that operates on llms natively. What python is to C, make some simplified english that targets python for the masses. I guess this is what vibe coding does today but skips directly onto the programming language. - "grammar" is the infinite set of possible strings that a language can express. by defining rules and making them self-referential, you can describe a grammar with only a finite set of rules. src: https://github.com/felgueres/cs/tree/main/interpreter

Make data traversable passage, treewalk

Failed hard at something today but on my bus ride I was reading Nicomachean Ethics and this beautiful line resonates: “And as in the Olympic Games it is not the most beautiful and the strongest that are crowned but hose who compete, so those who act win, and rightly win, the noble and good things in life” Reminds me of man in the arena

Semi analysis: - take satellite pictures of all data centers - search filings for every development - summarize all data - sell it and consult with it - profit

one cool thing about this experiment is that i could programmatically post on twitter from here. a write-only, one-way interface to prevent scrolling but still building an audience. maybe you could randomly pick from the feed or maybe add a button to post

coding resources: - https://github.com/FFmpeg/asm-lessons - https://engine-programming.github.io/ - https://github.com/practical-tutorials/project-based-learning - https://github.com/godotengine/godot - https://www.realtimerendering.com/ chess engine - https://www.youtube.com/watch?v=RFaFmkCEGEs

the trope to shit on software is that we're "optimizing a button color". software moves faster because regulation is lower and iteration loops shorter, this leads to many use cases that are often seen as frivolous. but it also advances protocols, control planes, and funds the next iteration of computing that eventually unlocks new leafs in the hardware tree. great insight: https://x.com/tszzl/status/1956926142998548580

Most startup advice is wrong. It should be framed on what do you want to do and figure out if there's a market for that. conviction is the thing that will allow you to spend all your attention and resources to outcompete others. Making something people want is neccessary but not sufficient. What do YOU care about, what do YOU think should exist, what do YOU see that others don't, etc, will be more predictive of your ability to grind it out and win.

Beauty is born out of purpose and performance. 1930-1950: Lockheed Constellation, Pioneer Zephyr. Aerodynamic, thinking of a metallic future. 1950-1970: Skunk Works SR-71 Blackbird and U-2. Frugal, austere, single purposed, unbeaten. 1970-1980: Concorde, Cray Supercomputers. Minimal, ruthlessly performant.

how python works 1. CPython is a C program that implements the interpreter 2. That program has 2 modalities: repl or in a loop with an exit 3. The interpreter turns source code into bytecode by lexing, parsing, and semantic analysis like compilers 4. The interpreter maps bytecode into their appropriate opcode handler. Those handlers are like a big if else statement with precompiled functions. 5. The handlers operate on python's stack, which is putting/poping data out it and then running instructions

reading about the history of web browsers and refreshing on how interpreters work, here's the steps: lexer or scanner: reads bytes, char by char, produces tokens. parser: check sequence of tokens against language grammar, produces an abstract syntax tree semantic analysis: ensure ops make sense, resolve variable names, scope rules interpretation: traverse AST, translate to bytecode, compile bytecode to native machine code on the fly runtime env: supporting functions to run program

why don't devs treat their development setup like a startup? instead of having many small repos scattered and each deployed separately just have a single backend folder with api stubs add webhooks to test and deploy, auth, grafana, etc, the entire stack that shared infra is like 80% of the work to launch a project, you just want to be swapping the application code to test different use cases

Reading: https://nlp.stanford.edu/IR-book/information-retrieval-book.html It hits different to read textbooks on stuff you've worked on. One obvious observation is that even very complex systems like google start with the most simple techniques. Understanding those building blocks gives you confidence to build in that space.

A training gym for keyboard shortcuts NVIM + window management

reasonablecomputing.com Plan: - Create ontology of an enterprise. - Create a hub with MCPs. - Continuously reason through incoming data. It’s like the moment electricity went from discrete events in experimentation to a continuous flow of electrons. Launch templated ways to map an org, or biolab, index the org. How is this different than enterprise search? What’s a good way to start?