Archive

• Managers are paid to drive results with some support. They have experience in the function, can take responsibility, but are still learning the job and will have questions and need support.  They can execute the tactical plan for a project but typically can’t make it.
• Directors are paid to drive results with little or no supervision (“set and forget”). Directors know how to do the job.  They can make a project’s tactical plan in their sleep.  They can work across the organization to get it done.  I love strong directors.  They get shit done.
• VPs are paid to make the plan. Say you run marketing.  Your job is to understand the company’s business situation, make a plan to address it, build consensus to get approval of that plan, and then go execute it.

migra: Like diff but for Postgres schemas

• compare schemas
• autogenerate migration scripts
• autosync your development database from your application models
• make your schema changes testable, robust, and (mostly) automatic

migra is a schema diff tool for PostgreSQL, written in Python. Use it in your python scripts, or from the command line like this:

$ migra postgresql:///a postgresql:///b
alter table "public"."products" add column newcolumn text;
alter table "public"."products" add constraint "x" CHECK ((price > (0)::numeric));

migra magically figures out all the statements required to get from A to B.

how to recognize passive voice, where it is harmful, and how to remedy it. This page is a free chapter from Refactoring English by Michael Lynch, the creator of TinyPilot KVM.

1. rebuilding itself results in small improvements
2. Rebuild with clang and better flags
3. Rebuild with mimalloc, an allocator from [[microsoft]]. "me-malloc" is a general purpose allocator with excellent performance characteristics. Initially developed by Daan Leijen for the runtime systems of the Koka and Lean languages.

I discovered that putting a man-in-the-middle proxy between my Apple TV and the world lets me decrypt HTTPS traffic. From there, I can read the Protocol Buffer data Google uses to populate YouTube with ads. It is too CPU-intensive to decode Protobuf on the fly, so instead, I found a flaw in the Protobuf format which allows me to reliably change one byte to obliterate ads.

What follows is a reference guide for setting up a bare-metal network router to block malicious ads, obnoxious ads, tracking, clickbait, crypto-jackers, scam popups, Windows spying on you, etc. using blocklists to protect all networked devices.

My friend told me that (with respect to this phenomenon) people can be roughly divided into two categories that she called perfection-oriented and performance-oriented.

The people in the category perfection-oriented have a natural intellectual curiosity. They are constantly searching for better ways of doing things, new methods, new tools. They search for perfection, but they take pleasure in the search itself, knowing perfectly well that perfection can not be accomplished. To the people in this category, failure is a normal part of the strive for perfection. In fact, failure gives a deeper understanding of why a particular path was unsuccessful, making it possible to avoid similar paths in the future.

The people in the category performance-oriented on the contrary, do not at all strive for perfection. Instead they have a need to achieve performance immediately. Such performance leaves no time for intellectual curiosity. Instead, techniques already known to them must be applied to solve problems. To these people, failure is a disaster whose sole feature is to harm instant performance. Similarly, learning represents the possibility of failure and must thus be avoided if possible. To the people in this category, knowledge in other people also represents a threat. As long as everybody around them use tools, techniques, and methods that they themselves know, they can count on outperforming these other people. But when the people around them start learning different, perhaps better, ways, they must defend themselves. Other people having other knowledge might require learning to keep up with performance, and learning, as we pointed out, increases the risk of failure. One possibility for these people is to discredit other people's knowledge. If done well, it would eliminate the need for the extra effort to learn, which would fit very well with their objectives.

Book outline:

• The components of a RAG system
• How to create a RAG knowledge base
• The indexing and generation pipeline
• Evaluating a RAG system
• Advanced RAG strategies
• RAG tools, technologies, and frameworks

Bought this book from manning today. However, instead of paying $25.99 for this book alone, I decided to purchase a monthly subscription at $24.99 that would allow me to read all their books online. I also get to keep the PDF and epub of this book. I think this is a good deal.

To more writing and technical blogging!

$ duckdb -ui

can have multiple databases including in memory database attached. Can explore the databases using Table summary, the notebooks are useful in organizing work into notebooks. SQL is the language of the notebooks like [[python]] is for [[jupyter]] notebooks. The notebook cells support syntax highlighting and autocompletion.

The results can be visualized using the column explorer, which shows the summary of the results. It is "type aware", and thus can visulize things like percentiles, median etc.

Interestingly, the the UI extension is all written in [[cpp]] - https://github.com/duckdb/duckdb-ui

A native macOS app that allows users to chat with a local LLM that can respond with information from files, folders and websites on your Mac without installing any other software. Powered by llama.cpp.

lexer generator that supports C/C++, D, Go, Haskell, Java, JavaScript, OCaml, Python, Rust, V, Zig, and can be extended to other languages by implementing a single syntax file. The primary focus of re2c is on generating fast code: it compiles regular expressions to deterministic finite automata and translates them into direct-coded lexers in the target language (such lexers are generally faster and easier to debug than their table-driven analogues). Secondary re2c focus is on flexibility: it does not assume a fixed program template; instead, it allows the user to embed lexers anywhere in the source code and configure them to avoid unnecessary buffering and bounds checks. Internal algorithm used by re2c is based on a special kind of deterministic finite automata: lookahead TDFA. These automata are as fast as ordinary DFA, but they are also capable of performing submatch extraction with minimal overhead. re2c is used in other open-source projects, such as php, ninja, yasm, spamassassin, BRL-CAD, wake, etc.

Tl;dr:

• If you want a minimalist, elegant Lisp with a strong academic foundation: choose Scheme.
• If you want a powerful, batteries-included Lisp with a rich standard library and decades of history: choose Common Lisp.
• If you want a modern Lisp with functional programming, concurrency, and JVM interop: choose Clojure.
• If you want a beginner-friendly Lisp with great tooling and a focus on education and extensibility: choose Racket.

Cross-sectional age-skill profiles suggest that cognitive skills start declining by age 30 if not earlier. If accurate, such age-driven skill losses pose a major threat to the human capital of societies with rapidly aging populations. We estimate actual age-skill profiles from individual changes in literacy and numeracy skills at different ages. We use the unique German longitudinal component of the Programme of the International Assessment of Adult Competencies (PIAAC-L) that retested a large representative sample of adults after 3.5 years. Our empirical approach separates age from cohort effects and corrects for measurement error from reversion to the mean. Two main results emerge. First, average skills increase strongly into the forties before decreasing slightly in literacy and more strongly in numeracy. Second, skills decline at older ages only for those with below-average skill usage. White-collar and higher-educated workers with above-average usage show increasing skills even beyond their forties. Women have larger skill losses at older age, particularly in numeracy.

Text-to-SQL is dead 💀

The next generation of querying is agentic - and it’s already here.

This paper (https://arxiv.org/abs/2502.00032) introduces a type of agentic querying called 𝗙𝘂𝗻𝗰𝘁𝗶𝗼𝗻 𝗖𝗮𝗹𝗹𝗶𝗻𝗴 that uses an LLM to structure queries using predefined function calls in https://twitter.com/victorialslocum/status/1897612417237983580/video/1

Textbook, second edition

DeepSeek is pushing DuckDB beyond its single-node roots with smallpond, a new, simple approach to distributed compute. But does it solve the scalability challenge—or introduce new trade-offs?

self hosted alternative to pinboard.

plan to use this to migrate obsidian repo to a combo of public and private notes with a script to just pull out non-private files to publish on the website.

Diffusion and flow-based models have become the state of the art for generative AI across a wide range of data modalities, including images, videos, shapes, molecules, music, and more! This course aims to build up the mathematical framework underlying these models from first principles. At the end of the class, students will have built a toy image diffusion model from scratch, and along the way, will have gained hands-on experience with the mathematical toolbox of stochastic differential equations that is useful in many other fields. This course is ideal for students who want to develop a principled understanding of the theory and practice of generative AI.

good blog on LLM, Machine Learning etc

Keep your calendar in a plain text file.

Calendar.txt is versionable, supports all operating systems

another collection of small, semantic css frameworks. I found Open Props very interesting.

Build tools are a burden

something to keep in mind while I redesign this site for the umpteenth time

via this X thread

Rest of the issue

surprisingly good summary of the "flow state".

"Repression is not the way to virtue. When people restrain themselves out of fear, their lives are by necessity diminished. Only through freely chosen discipline can life be enjoyed and still kept within the bounds of reason." -- Csíkszentmihályi Mihály Róbert

The Evolution of APL, the HOPL I paper by Falkoff and Iverson on APL, recounted the fundamental design principles which shaped the implementation of the APL language in 1966, and the early uses and other influences which shaped its first decade of enhancements.

By Roger Hui. (108 pages). Should be a comprehensive historical review of the APL language.

Wuffs is a memory-safe programming language (and a standard library written in that language) for Wrangling Untrusted File Formats Safely. Wrangling includes parsing, decoding and encoding. Example file formats include images, audio, video, fonts and compressed archives.

shift a substantial portion of LLM workloads to consumer devices by having small on-device models collaborate with frontier models in the cloud. By only reading long contexts locally, we reduce cloud costs with minimal or no quality degradation. We imagine a future where an “intelligence layer” running persistently on-device interacts with frontier models in the cloud to deliver applications with cost-effective, “always-on” intelligence.

Reseach by Stanford's Hazy Research.

• Github: https://github.com/HazyResearch/minions
• arXiv Paper: https://arxiv.org/abs/2502.15964

ollama also has a related blog post

Cites this work: "LLM×MapReduce: Simplified Long-Sequence Processing using Large Language Models"

The proposed LLM×MapReduce framework splits the entire document into several chunks for LLMs to read and then aggregates the intermediate answers to produce the final output. The main challenge for divide-and-conquer long text processing frameworks lies in the risk of losing essential long-range information when splitting the document, which can lead the model to produce incomplete or incorrect answers based on the segmented texts. Disrupted long-range information can be classified into two categories: inter-chunk dependency and inter-chunk conflict. We design a structured information protocol to better cope with inter-chunk dependency and an in-context confidence calibration mechanism to resolve inter-chunk conflicts.

git butler, Git client for simultaneous branches on top of your existing workflow, talked to git core developers and asked them to help define What git config settings should be defaults by now. Here are some settings that even the core developers change.

# clearly makes git better

[column]
        ui = auto
[branch]
        sort = -committerdate
[tag]
        sort = version:refname
[init]
        defaultBranch = main
[diff]
        algorithm = histogram
        colorMoved = plain
        mnemonicPrefix = true
        renames = true
[push]
        default = simple
        autoSetupRemote = true
        followTags = true
[fetch]
        prune = true
        pruneTags = true
        all = true

# why the hell not?

[help]
        autocorrect = prompt
[commit]
        verbose = true
[rerere]
        enabled = true
        autoupdate = true
[core]
        excludesfile = ~/.gitignore
[rebase]
        autoSquash = true
        autoStash = true
        updateRefs = true

# a matter of taste (uncomment if you dare)

[core]
        # fsmonitor = true
        # untrackedCache = true
[merge]
        # (just 'diff3' if git version < 2.3)
        # conflictstyle = zdiff3 
[pull]
        # rebase = true

read rest of the article on why

consistent with an opensource mindset, deepseek are curating a list of "awesome" stuff built with/on deepseek.

This list not only serves as a reminder of what can be done with deepseek, but also a general overview of all things possible with rest of llms too.

I have some suggest windsurf as an alternative for Cursor, just days after I paid for annual subscription.. sigh..

In either case, there is nothing stopping me from really getting into Vibe Coding.

Moar coding!1!

Anthropic announced new version of Claude today, and it has been received very positively going by the reactions on X. Simon Willison ran it through his usual gauntlet.

The new thing is the claude CLI tool that allows you run commands against Claude api from the command line, which also would be a big aid for programming tasks.

I had unsubscribed from Claude for the last couple of months, and looks like it's time to get back on it.