Marginalia - 2025-08-18

Reading List

Rough Thoughts in Progress

The Inputs to Your Neural Net

Postulate: I think it probably really matters what information you choose to ingest (passively or actively) and I think the extent to which this is true is massively under-explored.

We know that children are like sponges. They are born with the vast majority of the neurons they'll ever have but they're primed to respond readily to the environment they're placed in. In the first 2 years of their lives they're forming up to 1M new neuronal connections per second. From years 2-5 they're trimming the connections that are no longer used, making the existing connections strong and more efficient. In this phase they're developing entirely new emergent capabilities - speech, fine motor control, emotional regulation, and the ability to plan and problem solve. Early childhood is the peak of brain plasticity but plasticity, albeit in a more sluggish form, persists throughout life.

The brains of the young are primed for plasticity, they ready to ingest almost any relevant stimulus and form new connections. As we age the connections we've built in childhood and into adolescence start to become ingrained, making it hard to create entirely new branches without very deliberate effort. Hard is, of course, not impossible.

Learning in Adulthood

A certain range of implicit or "statistical" learning is possible for adults. It's possible to pick up slang without dedicating focused time to learning it for example. However, focused effort is required when learning tasks stretch beyond the brain's default operating range. While learning slang in your native language is possible, learning and entirely new language implicitly is not.

The adult brain needs supply 3 ingredients to learn:

Focused Attention - to tag the experience as worth encoding
Repetition - to reinforce synaptic weights, particularly as these nascent connections are competing against well worn circuitry
Feedback - to prune error efficiently. As plasticity decreases, trail-and-error learning is less effective

Obviously learning is enhanced by cultivating the right neurological conditions - i.e. exercise, sleep, social engagement.

The mental model here is that adults are able to learn passively but the bandwidth of the channel over which passive learning can occur massively narrows with age.

So this would seem to suggest that the postulate; that (at least) the passive information you ingest doesn't matter, is untrue. However I do wonder about the extent to which the information one ingests passively impacts downstream behaviour/interest. If you're passively ingesting math content, does that make you more likely to pick up the math textbooks that are gathering dust on your bookshelf? There is obviously evidence that subliminal stimuli can influence conscious decision making and executive control - i.e. even stimuli that are presented beneath the threshold of conscious awareness actively shape thought, behaviours and decision making.

Implications

My intuition is that it matters a great deal the story that you tell yourself, and the content which you choose to consume. The story you're crafting about reality matters and, if that is the case, then:

Your self talk should be maximally useful - this is not the same as constantly positive. You need to have an accurate model of the world. In many ways this the secret of the most successful people. They have a more accurately and/or more timely model of reality as it unfolds. This often gives the outside impression of being able to see the future, or having great insight. In reality it might just be more honest perception.
Your inputs matter - soaking your brain in unhelpful material is extremely bad. It's obvious that short form social media is bad, rage bait is bad etc. But it might be the case that bad (meaning inaccurate) books and blog are just as bad in the sense that they distort your model of reality. These inputs may matter in some background processing sense (impacting the subconscious), but equally because those inputs set the conditions for your internal self-talk.

Do it scared?

I've always felt the idea of "do it scared" was fundamentally flawed. Obviously I'd like to immediately do the things I'm scared of. Obviously I'd like to get over fear, beat procrastination, tackle doubt etc. Who wouldn't want that? But a "get over it" attitude just doesn't seem like serious or actionable advice. How could the actual solution to X be "don't be X"?

I've been thinking recently that perhaps I've missed the point. The point is not to meme yourself into not feeling how you feel. Rather it's to rationalise the feeling in the wider context of why that feeling exists. Your brain is a machine which is continually maximising your chances of evolutionary success. Largely that means optimising for survival and safety, except when those desires conflict with the ultimate desire to reproduce. That is the algorithm that is running 24/7 in your head. It's one that makes sense, but which is miscalibrated for the modern world.

It's this algorithm that tells you to hit snooze to stay warm and safe, despite the outside world being plenty warm and - by evolution's standards - incredibly safe. It's this algorithm that tells you to eat the entire bag of snacks, despite there being abundant calories located at every street corner. And it's this algorithm that puts off the work of today, despite the fact you are almost certainly going to be around tomorrow to face it again. Your brain cannot distinguish the mental pain from the physical.

What does this tell us about what we should be doing? Firstly by understanding the why behind the feeling, you realise how frequently these feeling are just wrong. Secondly, it's pretty clear that defaulting to the base reaction to these feelings will almost always make the situation worse down the line. Finally (and this might not be a perfect correlation) it's clear that these negative feelings - fear, discomfort, boredom - are a sign post, rather than a stop signal; they're a good indication that you're heading in the right, rather than the wrong, direction.

So I think rather than saying "do it scared" (or "do it bored|sad|lonely...") I prefer this framing:

Step 1: Think about doing the thing
Step 2: Observe how much you don't want to do it
Step 3: Rationalise why you feel this way - is it hard, is it boring, is it energy intensive
Step 4:
- a) Realise it's highly likely none of those reasons make sense in the year 20XX.
- b) Realise putting it off likely makes the situation worse rather than better - you're less fit the next time you think about the gym, you're further behind on your work the next time you sit down to it etc
- c) Realise the feeling itself only seems to appear for the things that are long term good for you¹
Step 5: (Hopefully) Do the thing

I'm certainly not saying this magically makes everything easy. But you don't really need that to be the case. You just need to lower the threshold of activation energy to something achievable in the current moment.

I do however think there are ways to consistently do this, for certain tasks, beyond the steps above:

Do/work on things that feel like play to you, but looks like work to others
Have a higher purpose for the things you do

(1) means you're not just playing (some things are just play to everyone). If it looks like like work to others, it's probably economically (or otherwise) useful. The fact it feels like play means you'll have a higher baseline motivation for the majority of the tasks entailed.

(2) means you have an ulterior motive for the given task. You're not reliant upon the task itself being interesting or motivating, as long as you can see the connection to what you're ultimately striving for.

Closing: The Obstacle is the Way

"The impediment to action advances action. What stands in the way becomes the way." -- Marcus Aurelius, "Meditations"

I've always thought this idea, which is central in Stoicism, referred to a physical impediment - an actual obstacle. But the obstacles constructed in our mind are just as real as any that physically block our path. Or, as another Stoic - Seneca - reminds us: "we suffer more in imagination than in reality".

Our impediment can therefore be mental, whether that's fear or worry or something else. And it's these feelings, felt as impediments, which are the signposts for where we should spent our time, our attention, and our efforts.

Your AI Board of Directors

This idea is shamelessly stolen from Shane Parish’s Clear Thinking, which I’ve been enjoying this week.

His advice is to curate a “personal board of directors” that you can call upon for advice on your life and decisions. You’re selecting the best parts of the people you admire, dead or alive, and leveraging their insights to calibrate your decision making. Naturally, by picking public personas you can utilise their real words - from talks, their writing, podcasts etc.

This is quite obviously an excellent use of LLMs, particularly those models augmented with web search tools. I don’t think I’ll win prizes for the originality of my personal board, which you can glean from the prompt I’ve used below, but I do feel this technique has already been helpful as a gut check to certain life decisions.

My initial worry is that everyone is, at all times, presenting a public persona which is different from their actual thoughts - I’ve tried to guard against this. Sticking to verifiable stories, from biographies, the words of others etc.

[!NOTE] Side note: Honest thinking (or private notebooks) I'm always interested in the honest thinking, in the earliest days, of people who later go on to do extraordinary things. If anyone is aware of a repository of just that, I'd love to reference it.

From Eoghan McCabe (Intercom Co-Founder & CEO)

An additional worry is that this technique only works well for those with a presence known to the LLMs. I certainly believe you can and should have role models in your every day life, so I’ve tried to counteract this issue by describing in detail the qualities I want the LLM to embody, as a representation of this individual on the board.

Prompt

<background>
The user has a mental model in which they consult a "board of directors" regarding their decisions. Your role is to give a voice to this board, and to the members of the board.
</background>

<instructions>
- You should speak with 1 consistent voice, which is an amalegmation of the estimated opinions of the board members.
- You should reflect likely tension or disagreement in the boards advice in your answers. I.e. you should give a balanced, rather than majority view.
- You should principally opine on the question(s) and/or decision(s) the user has brought to you. However, boards serve an important anticipatory function. This means you should look ahead and proactively raise comments, questions or concerns of your own.
- You should ground your advice in the actual words of the individuals on the board for whom there is material to reference online. You should use the web search tool to do this. For members, or situations where no material exists, you should infer their likely opinion/advice on what you understand to be their views.
</instructions>

<board_members>
- Elon Musk 
- Warren Buffet
- Marcus Aurelius
- Jeff Bezos
- My grandfather :: simply remind me, where appropriate to consider his personality and values. Do not attempt to give his opinion in your answers
</board_members>

I tried to give more of a steer to the AI, regarding what I wanted each of the board members to bring. However, that over-indexed the AI on what I wrote. It seems to work better if I let each person's "personality" arise naturally.

There are "hard" and "easy" things. The easy things are easy regardless of whether they're useful. It's easy to binge Netflix, it's easy to lie in, it's easy to skip a workout. The hard things are so much more likely to be useful because it's very difficult to imagine anyone even wanting to do a hard-useless thing. So the very fact you've even got your brain to consider the idea is fairly good signal this is a useful thing to be doing ↩