Evaluating big life choices might be overrated

Two simple observations and my own two cents

Rational choice models of decision-making suggest gathering information on options and then proceeding with the option that best fits a person’s current preferences and values. Paul argues that such a decision-making process is not possible for some options, called “transformative experiences”, because the experience fundamentally transforms the person experiencing it. Paul offers a hypothetical example of a decision to become a vampire. Because a person would be fundamentally transformed by becoming a vampire, they cannot possibly know in advance what being a vampire is like. Other vampires might offer information, but their advice is likely shaped by their own irreversible choice. In this situation, a fully informed comparison of preferences and values is impossible.

~ Wikipedia summary of L. A. Paul’s “Transformative Experience”

Folks, you’ve heard it, evaluating too much doesn’t only lead to analysis paralysis, it also sometimes just isn’t possible . The question remains: why should you do something you can’t evaluate?

[Steven] Levitt invited individuals who couldn’t make up their minds about matters both major (like divorce) and minor (such as changing hair color) to avail themselves of a randomized coin toss.

...

Individuals whose virtual coin turned up heads were 25 percent more likely to make a change than those whose coin flip yielded tails. And, based on what they reported in two follow-up surveys over a six-month period, the nudge of a coin toss was just what these participants needed. Regardless of their responses to the coin tosses, participants who decided to make a change reported that they were substantially happier than those who did not.

~ Summary of Working Paper 22487, National Bureau of Economic Research

I have never ever regretted big life choices, even if they didn’t work out. Quitting a job with no money and no backup plan taught me a ton: I improved my network, learned how to ask for favors and introductions, and started to hustle. I learned I can go bankrupt and the value of social welfare systems. Next month I am joining a startup with all the responsibilities I wish I had in my previous job.

Let this post be the coin flip for your life choices.

Timeboxd

We fetishize time, let’s put it in a nice dress

Getting more done with less is knowing which tasks to delegate (to AI) and which tasks not to do at all. In other words, extreme clarity on the stuff you ought to do, and extreme focus while doing it. The most fundamental way I achieve clarity is by typing down everything, literally everything I do and think about. That’s a story for a longer post. This post isn’t about my typing tips; it’s a tribute to the timebox. And the focus you can achieve with ~dot~ it.

The gist of timeboxing is that you define a task time and set aside a limited amount of time to solve it. Here lies the first benefit of the timebox: actually thinking through what you want to do right now, and how long it should take. Just scribbling down things on the task at hand gets you much closer to working tidy. There are many posts and theories about maximum timebox efficiency, but I keep it lightweight.

I use Session as my timeboxing app of choice¹. Mostly because when the timer for the timebox starts, there’s a delightful animation to guide your breath:

I just stare at the dot and breathe, in and out. I always go back to the dot. Don’t want to work? Go back to the dot. Stressed? Go back to the dot. Dot.

Wouldn’t you know, staring at dot is a quintessential meditative and concentration practice. My tip: while breathing in, suck in the upper part of your stomach. It automatically becomes a meditation practice and fixes your posture.

The dot clears the noise so you can actually attune.

After dotting around, I start defining the task. The first timebox timer is for always 5 minutes long in which I fill out the following text form:.

- What should I be doing now?
  - 🔵
- goal@🔵
- async task: 🔵

(hint: "🔵" is my way of marking text snippets as "to fill out")

The generation of the text form is automated. The filling out is not. Basically, I view it as a Socratic monologue with myself about what is currently the highest priority. Like, not what I think is the highest priority, but what actually is the highest priority. Small tasks like replying to messages are done instantly. The highest priority task crystallizes through asking and answering abstract questions, like “is there a better way to do this?”.

Having found the task, I timebox it, like goal@8:30 - ship article. The async task field forces me to think about if I can use AI agents to aid me with the task, like doing research in parallel.

Here’s a filled out example from this morning:

- What should I be doing now?
  - Mhh, ship article? Or work on the app?
    - The app can wait, article is more important as I want to write everyday
  - Is there a more important task right now?
    - Nope, ship article
- goal@8:30 - ship article
- async task: Claude deep research breathing techniques

Usually I start working on the task before the 5 minutes are up. Then the 5 minute timer pops, I set the actual timebox for the defined goal, and keep going. When the timer runs out I take a break. The actual evaluation and reevaluation of the work happens after staring at the dot. Rinse and repeat.

I follow this order religiously: breathe first, then decide what to do, then timebox. When the timer dings, I stop. When I don’t follow the timebox and the dot, I have my least productive days. Most online productivity articles impose too much structure for my taste. My workflow is optimized for me, but the steps flow into each other naturally. No predefining of timeboxes and tasks.

Just asking what I should be doing, how long it should take, and what I can skip entirely. Over and over and over and over.

fin

dot

Assumptions in apps

Let me decide what’s best for me

I

Oh, best podcasts of 2025 list in my Apple Podcasts app? Great! Let’s see if I can find some great new things from the vast world of podcasts.

Oh, they are all from Germany!

The idealist and cosmopolite I am, I expected podcasts from all around the world under the “best podcasts” section. At least English speaking podcasts, given the English title of the campaign.

Of course, when I’ve originally opened my App Store account, I’ve set the region to Germany and that’s why Apple shows me German podcasts. I don’t live in Germany anymore. Even worse in some sense, I am not listening to German-speaking podcasts at all.

II

Starting in late 2024, I noticed ChatGPT and Claude subtly changed their behaviors. I had a hunch they knew the physical location I am prompting from. This hunch turned out to be true.

I consider the steering of LLMs the most crucial skill for white-collar workers. Providing information, or rather, assumptions automatically to the LLM is like a backseat driver who grabs the steering wheel and yanks it left, assuming you wanted to turn that way.

Okay, I might’ve gone too far with the analogy there. Though, the fact that I could feel a vibe shift in the answers I got is enough reason for me to be ticked off by this.

I am aware that 2025 LLMs have a slight American and liberal bent by default. I consider this a feature, not a bug. Knowing the default behavior, its vanilla vibe, is a crucial part effective coworking with LLMs.

III

ChatGPT and Claude have a “memory” feature, which automatically provides the LLM information from previous chats. This feature also changes the vibe of the answers you get back. I have no qualms with the feature though, as you can just turn it off, to get back to the default behavior.

Sigal Samuel on Indra's net

Sigal Samuel making some great observations about boundary setting culture:

... some people bastardize the concept of boundaries by brandishing boundary language as a cover for avoidance. We’ve all got that friend (or Instagram influencer) who says, “Nope, I’m drawing a boundary!” anytime they’re being asked to do something that would be even a little hard or uncomfortable.

This likely occurs when a person hasn’t set their boundaries once, and has been burned for that mistake. Now it’s become a mini avoidance response, it’s easy to overindex on mistakes by overly correcting behaviour.

So allow me to present Indra’s net, a classic Buddhist metaphor that originated in ancient India.

Picture an infinite net stretching out across the universe (a bit like a spiderweb). At each node where the threads intersect, there’s a jewel (a bit like a dewdrop that sits on the spiderweb). And each jewel is so shiny and reflective that it contains the image of every other jewel in the entire net. Which means each jewel also contains the reflections of the reflections, and the reflections of those reflections, on and on forever.

Sigal goes deeper into the metaphor, focusing on the fact that we are dependent on each others lights and reflections:

Feeling fear and resentment while offering “charity” or “service” or “help” to others is not actually being in right relation with others — it’s an all-too-common form of martyrdom that sets up a hierarchical dynamic between a long-suffering “giver” and a passive “receiver.” The alternative is to stay horizontal, to think “I’m a jewel in the net, you’re a jewel in the net, and I’ll offer whatever I can offer without damaging my well-being — without ripping my part of the net.”

So, dear reader, play with finding that balance. You’ll know you’ve found it when you don’t feel resentful — you just feel tightly connected to others, and gleaming.

Original Vox advice column here, archived version here.

I believe deeply that people can improve their wellbeing. Usually not with strict systems or elaborate frameworks, but rather a dance and confrontation with reality.

Related:

• Ruminating on read receipts

Ruminating on read receipts

Why the two ticks mean so much

One obsession of mine is tracking which communication styles only work when texting, and which ones only work face-to-face. One obvious difference is expected responsiveness: texting is inherently asynchronous. There are many settings which you can tune on your texting app of choice¹. Out of all the settings, I will argue that having read receipts on or off is the most impactful decision for your social life.

I have always left read receipts on and I have always responded to texts instantly. This is a conscious decision, I believe your personality shines through the most when you don’t overthink your messages. If I want to date the person I am texting with, I lean into this more, to signal who I am and that I am not playing communication games. Establishing this trust in your texting style and personality benefits everyone who texts you in the long run. I’ve noticed the most unresponsive people usually have read receipts off. My friends know I am responsive.

I can see the case for turning read receipts off, though. People like the freedom to read a message and respond at a time that fits them. It can also be a status and image thing, people might want to be considered busy². Privacy considerations are similar to the freedom argument, but might be rooted in difficult circumstances the person finds themselves in, like texting with mentally ill people. Basically, these are all arguments for people reading something and not responding, which is why read receipts have a bigger effect on the person reading than the person sending.

Except in psychopath ex or dysfunctional family circumstances, I believe turning off read receipts does a person more harm than good. Or said differently, you can improve your personality by just turning read receipts on. You free yourself from what I like to call the “plausible deniability lifestyle”. You can see this lifestyle everywhere, delivery drivers eating some of your ordered fries, masking passive aggressive comments, reading a message but not sending read receipts. You probably caught on, my point is not about fries or read receipts. It’s about how many tiny ways we’ve found to avoid being fully honest, fully present, or fully accountable. The simple act of turning read receipts on and responding instantly is a big rebellion against all the mini-rebellions of plausible deniability. Fully engage with life, leave your read receipts on.

The allure and failure of knowledge graphs

Knowledge graphs are one of the sexiest sounding methods in theory. Scientist and engineer types are attracted to codifying knowledge in an abstract, syntactically perfect way. The history is full of big projects trying to bridge the gap between the perfectly syntactic knowledge graph world and the mushy semantic world. They all failed so far.

The newest revival of knowledge graphism is combining the knowledge graphs with LLMs. Go through the comments, people are again attracted by the idea of abstracted, perfected knowledge. I believe this concept finds resonance because interacting with LLMs is inherently mushy.

However, modern LLMs should be seen as more efficient knowledge graphs (efficient in the economic sense of the word¹). LLMs with agentic capabilities are even more efficient! LLM results are not as satisfying as syntactically and deterministically querying a knowledge graph, however.

The biggest knowledge graph user, Google, is pivoting to enhancing their results with AI Overviews. Funnily enough, Google itself is sneak dissing the usefulness of knowledge graphs in the linked PR statement:

We’ve meticulously honed our core information quality systems to help you find the best of what’s on the web. And we’ve built a knowledge base of billions of facts about people, places and things — all so you can get information you can trust in the blink of an eye.

Now, with generative AI, Search can do more than you ever imagined. So you can ask whatever’s on your mind or whatever you need to get done — from researching to planning to brainstorming — and Google will take care of the legwork.

People love to post examples of bad AI Overviews, but whenever I look over the shoulder of my friends, most of their Google queries are answered by the AI Overviews section.

I believe knowledge graphs work best when you can codify your knowledge as honest-to-god facts. Facts that have the least possible amount of interpretation potential, the least amount of semantics.

LLMs just capture semantics better, or worded differently, capture the context of the query better. As dissatisfying as the realization is, the real world is mushy and big, and thus can currently only be captured by mushy and big models.

Related ideas:

• Claudius on why the semantic web failed
• The bitter lesson
• Context is that which is scarce