Reading Log

by Kurt Pan

Claude Code (CC) feels great to use, because it just simply works. CC has been crafted with a fundamental understanding of what the LLM is good at and what it is terrible at. Its prompts and tools cover for the model's stupidity and help it shine in its wheelhouse. The control loop is extremely simple to follow and trivial to debug.

Edit is the most frequent tool, followed by Read and ToDoWrite

Claude Code choses architectural simplicity at every juncture – one main loop, simple search, simple todolist, etc. Resist the urge to over-engineer, build good harness for the model let it cook!

The difference in Claude Code's performance with and without claude.md is night and day.

If your coworkers are hopping between multiple agents, chewing on ideas, and running in the background during meetings, and you're not in on that action, then naturally you're just going to fall behind.

A model is either high safety, low safety, an oracle, or agentic. It's never both or all. Oracles are on the polar opposite of agentic. Oracles are suitable for summarisation tasks or require a high level of thinking.

But you don’t measure that by the energy you put in, or the hours you’re sitting in the office, but the output you produce. Burning out on twelve-hour days, six days a week, has no prize at the end. It’s unsustainable, it shouldn’t be the standard and it sure as hell should not be seen as a positive sign of a company.

Calligraphr works by having you print templates, write out the letters, then scan them in. It does some magical processing to extract the glyphs, provides tools to tidy them up, align them, etc, and then produces a TTF file for you.

All the same privacy harms with online tracking are also present with AI, but worse. This richer personal information can be more thoroughly exploited for manipulation, both commercially and ideologically.

I think part of how we get pushed into things we collectively don't want is because we stay quiet about it.

Many of my contacts in the open source community have been talking about plans to move from GitHub to Codeberg or a self-hosted Forgejo (Forgejo is the software used by Codeberg) over the last month.

I refuse using AI in the same way I don't take drugs or steal things – for me it's a matter of principle. So if this continues and Microsoft does not provide a way to opt out of AI for my repositories soon, I will move my code to a self-hosted solution and won't ever return to GitHub.

Out of the Fourier transform grew an entire field of mathematics, called harmonic analysis, which studies the components of functions.

He spent the next decade conflicted about whether to dedicate his life to religion or to math, eventually abandoning his religious training and becoming a teacher.

Most mathematicians at the time believed that no number of smooth curves could ever add up to a sharp corner.

The process only fails for the most bizarre functions, like those that oscillate wildly no matter how much you zoom in on them.

This infinite set is called the Fourier series, and — despite mathematicians’ early hesitation to accept such a thing — it is now an essential tool in the analysis of functions.

In the 1960s, the mathematicians James Cooley and John Tukey came up with an algorithm that could perform a Fourier transform much more quickly — aptly called the fast Fourier transform.

Mathematicians have also found that harmonic analysis has deep and unexpected connections to number theory.

There are several transformations, such as the Fujisaki-Okamoto (FO) transform, that construct a (secure) KEM from a (secure) PKE scheme.

A testament both to Wikipedia’s success and to the total degradation of the rest of the internet as an information resource.

It’s basically the only place on the internet that doesn’t function as a confirmation bias machine. Wikipedia is one of the few platforms online where tremendous computing power isn’t being deployed in the service of telling you exactly what you want to hear.

As varieties of truth go, facts are fragile. Unlike axioms and mathematical proofs that can be derived by anyone at any time, there is nothing necessary about the fact, to use Arendt’s example, that German troops crossed the border with Belgium on the night of August 4th, 1914, and not some other border at some other time. Like all facts, this one is established through witnesses, testimony, documents, and collective agreement about what counts as evidence — it is political, and as the propaganda machines of the 20th century showed, political power is perfectly capable of destroying it. Furthermore, they will always be tempted to, because facts represent a sort of rival power, a constraint and limit “hated by tyrants who rightly fear the competition of a coercive force they cannot monopolize,” and at risk in democracies, where they are suspiciously impervious to public opinion.

Facts, in other words, don’t care about your feelings. “Unwelcome facts possess an infuriating stubbornness,” Arendt wrote.

Heart rate is one of the most basic and important indicators of health, providing a snapshot into a person’s physical activity, stress and anxiety, hydration level, and more.

Traditionally, measuring heart rate requires some sort of wearable device, whether that be a smart watch or hospital-grade machinery. But new research from engineers at the University of California, Santa Cruz, shows how the signal from a household WiFi device can be used for this crucial health monitoring with state-of-the-art accuracy—without the need for a wearable.

Slowing down makes reality vivid, strange, and hot.

Dopamine is often portrayed as a pleasure chemical, but it isn’t really about pleasure so much as the expectation that pleasure will occur soon.

A stress hormone like cortisol, on the other hand, has a half-life of 60–90 minutes and so can take up to 6 hours to fully clear out after the onset of an acute stressor.

The more time the different systems of the body have to synchronize with each other, and the deeper the experience gets.

When we let our attention linger on something, our bodily systems synchronize and feed each other stimuli in an escalatory loop that restructures our attentional field.

If you focus on your anxiety, the anxiety can begin to loop on itself until you hyperventilate and get tunnel vision and become filled with nightmarish thoughts and feelings—a panic attack.

If you learn to pay sustained attention to your happiness, the pleasant sensation will loop on itself until it explodes and pulls you into a series of almost hallucinogenic states, ending in cessation, where your consciousness lets go and you disappear for a while. This takes practice. The practice is called jhanas, and it is sometimes described as the inverse of a panic attack.

The fact that we can enter fundamentally different, and often exhilarating, states of mind by learning how to sustain our attention is fascinating.

I realized that good art—at least the art I am spontaneously drawn to—has little to do with communication. Instead, it is about crafting patterns of information that, if you feed them sustained attention, will begin to structure your attentional field in interesting ways. Art is guided meditation. The point isn’t the words, but what happens to your mind when you attend to those words (or images, or sounds). There is nothing there to understand; it is just something to experience, like sex. But the experiences can be very deep and, sometimes, transformative.

When you think from first principles about what generative AI can do, and what jobs it can replace, it’s the kind of things that young college grads have done in white-collar firms.

Young workers aged 22–25 in “highly AI-exposed” jobs, such as software developers and customer service agents, experienced a 13 percent decline in employment since the advent of ChatGPT.

Quantum mechanics deals in something more abstract: possibilities. It predicts the chances that we’ll observe an atom doing this or that, or being here or there. It gives the impression that particles can engage in multiple possible behaviors at once, that they have no fixed reality. So physicists have spent the last century grappling with questions like: What is real? And where does our reality come from?

“We’re privileged to live at a time when the great prize of making sense of quantum theory is still there for the taking,” Spekkens said, “and any one of us could take it.”

They thought AI was making them 20% faster, but it was actually making them 19% slower.

Tech leaders everywhere are buying into the FOMO, convinced their competitors are getting massive gains they're missing out on. This drives them to rebrand as AI-First companies, justify layoffs with newfound productivity narratives, and lowball developer salaries under the assumption that AI has fundamentally changed the value equation.

Nobody is shipping more than before.

You're not falling behind by sticking with what you know works.

We all know that the industry has taken a step back in terms of code quality by at least a decade. Hardly anyone tests anymore. The last time I heard the phrase “continuous improvement” or “test-driven development” was before COVID.

Solar powered blog! This summer I’ve been testing using a 100w solar panel I got from Harbor Freight Tools so I can learn more about how it all works before diving into larger projects. I have that panel connected to a Jackery 160w power station to keep it charged up and we use it to charge our mobile devices. I got the Jackery last year as a power bank I use while on jobsites.

Treating AI like a “junior developer who doesn't learn”.

Your job as an engineer is to find the best solution for the problem, not just write a bunch of code.

Pick one small, well-defined feature. Give AI three attempts at implementing it. Review the output like you're mentoring a junior developer.

Using ChatGPT to help write essays leads to long-term cognitive harm—measurable through EEG brain scans. Students who repeatedly relied on ChatGPT showed weakened neural connectivity, impaired memory recall, and diminished sense of ownership over their own writing. While the AI-generated content often scored well, the brains behind it were shutting down.

One thing is clear: if you currently use AI, take regular breaks—and give your own mind the chance to do the work. Otherwise, you may face severe cognitive harm and dependence.

They can care about work-life balance, because they’re not desperate enough to feel the need not to. And however successful your company has been so far, they have other options they like better.

Hiring is a negotiation, and you’re acting like you’re holding all the cards when you aren't. You’re looking for a highly competitive candidate pool, and you’re not being competitive: you’re just checking the same baselines as everybody else. You're acting like a replacement-level employer and expecting more than replacement-level candidates.

When you accept that you need a great engineer, and not the best engineer, you can deal with the trade-offs consciously. What traits are actually important? How much are you willing to give up to get them? What’s the dollar value of a hire this month versus next month? “What actually matters today?” is the most important question a startup can ask, and you haven't applied it to one of the most important aspects of running a company!

Trying to hire the best engineers is the enemy of actually hiring great ones. You’re going to have to give up something (possibly time, possibly comp, possibly workplace policy) to make the hire you want.

The longer you aren’t thinking about what to give up, the more you’re implicitly choosing to give up time, the thing startups treasure more than anything else.

The default outcome for a startup is always failure. You took a risk by even starting one. You ship things that might be broken all the time, because you know that speed is more important than perfection. You take moonshots, because you know that big wins matter more than small losses. And then you give up months of time because you refuse to apply the same philosophy to hiring!

Stop insisting on perfection, and move fast.

You already live in social credit. We just don't call it social credit.

Enter your email to subscribe to updates.