Bits and Bobs 12/11/23

Think of the Bits and Bobs as my own weekly personal ritual of reflection and synthesis of my notes from the week before.

I do this primarily for me, but share them just in case they're interesting for others too.

They're best thought of as a kind of lightly distilled stream of consciousness.

These aren't meant to be read end-to-end; feel free to pick and choose, pull on the threads that tickle your fancy.

One final tip: they tend to go from concrete to more abstract and less obviously relevant to a day job.

Once you get to one that's too out there, it's safe to stop there--it will only get weirder from there!

Beauty is partially about quality but partially about the perception of it, the superficial characteristics.

If you're pushing a rock up a hill (pushing against the natural gradient of entropy) then you have to carefully plan every move and never slip up.

But if you're skiing down hill (going with the natural gradient of entropy), you still have to make decisions about your route and respond to conditions... but the default is unless you get stuck in a dead end you will get there eventually.

If you can choose between a push-a-rock-up-a-hill strategy and a ski-down-a-hill strategy, always pick the latter.

But either is sufficient to mean it is impactful.

Big numbers means it's already impactful.

Compounding curve (as long as the asymptote is still far away) means that it will be big numbers soon.

Humans tend to miss compounding curves: "This is a small number on an absolute basis, I don't need to pay attention to it".

Finding currently-small compounding curves that complement you and nurturing them is a consistent source of alpha.

Software is at the level of bits; a high pace layer.

If a given piece of software doesn't move quickly, it will be outmaneuvered by competitors.

Software also tends to be used in coordination with other software, which means it has to typically keep pace with development of other software.

The best examples of timeless software are things like Unix utilities.

The reason they can be timeless despite all of this is they focus on doing one contained thing, and use one consistent, flexible, well-designed abstraction to interact with everything else: a pipe / file.

Unix's "everything is a file" is like the lego dot: a timeless abstraction that allows pockets of timelessness in a constantly evolving larger system.

To be honest, this is a style of running a company I have a hard time grasping because it's outside my lived experience.

Apple has a very small number of senior, empowered, and accountable "chefs".

The vast majority of people in the organization are expert "line cooks".

Those line-cooks don't resent that they are implementing someone else's vision.

They are proud to work in the best restaurant in the world.

They are proud to use their expertise to create extremely high-quality output in their niche better than anyone else could.

They are proud to help realize, with the highest possible fidelity, a vision sketched by the chef.

Most of Silicon Valley relies on fractal layers of entrepreneurial people--which can create emergent great outcomes but also be a mess.

Of course, this doesn't have to be a binary.

When there's an inspiring, comprehensive, concrete product vision, it allows people across the organization to slot their individual pieces into something that allows them to have local autonomy on details, but still get global coherence.

When vision, mission, and strategy align, it creates the space for entrepreneurship within.

For example, at Pixar, the brain trust would give high-intensity feedback on a regular interval... but it was up to the director/producer to decide what to implement.

When you get mandatory feedback, you might feel coerced into following it even if you don't agree with it.

And if you're doing it from extrinsic pressure and not intrinsic alignment, you'll drag your feet and implement it poorly.

When the feedback is truly optional, you can treat it not as a threat to defend against, but a gift you have the option of accepting.

This allows you to go from "the thing they're missing is..." to "the thing they're seeing is...".

When you get to apply your own agency, you can commit to it much more deeply.

Paradoxically, having the choice to adopt the feedback often leads to it being adopted in a more powerful and resonant way.

Boring things here means, "unsurprising", which means "your mental model is sufficiently calibrated to correctly predict its output in practice"

For boring things, automate them. Get to good enough and don't overthink it.

Distilling useful Infrastructure makes it so the creative effort doesn't have to be wasted on routinized things.

Spend most of your time thinking about interesting things.

Otherwise you quickly get something like Brasilia: top down process focused on optics of cleanliness, not real-world effectiveness.

It takes watching carefully for an extended period of time, but when you do the size will become obvious to you.

But almost like knowhow, that obviousness to the observer is very difficult to transmit to others.

The sum total of indirect value often dominates the direct value caused by an action, but as humans we spend all of our time on the latter.

The more meta something gets, the more leverage it can have, but also the farther away from ground-truthing it gets.

It might sound genius, but be completely non-viable in the real world for some non-obvious reason.

If you don't have an edge, you have no natural advantage in that context.

Spend most of your effort playing in games you have an edge in.

Ask yourself: "What is my edge?" and then lean into it.

In an economy, an example is the competitive process between companies.

Inside of organizations, an example is the emergent internal political dynamics.

This expansion happens because social processes are recursive.

Each individual agent gets a little bit of edge from doing an extra level of recursion and response.

"They just did X, which means I should do Y in response."

And then later: "They just did X', which means I should now do Y' "

If the agent didn't do this extra recursion, they'd leave possible value on the table they could harvest... or maybe even get knocked out of the game.

This local individual process leads to the overall system emergently expanding until it fills all available space: the carrying cost of how much the system as a whole can get away without being ground-truthed into oblivion.

Organizations that are massively profitable and large scale can have a very high threshold of carrying cost; the internal political dynamics in those cases can expand to dwarf the actual value-creating activities.

When you're very far, it's very clear to everyone that you're not infallible, and they'll take actions to hedge.

This hedging creates resilience because it caps the downside.

But when you're very close, you start believing it, as does everyone else.

And then in the times you don't hit it, it's dangerous because it will surprise you and others in a big way.

Even the most galaxy brained person, the most impressive organization, are imperfect earth-bound entities.

No one and no thing is infallible.

The intuition of a smuggled infinity: "If you slip an infinity into any argument you'll end up in an absurd place"

It's easy to accidentally slip in an infinity without realizing it.

Perfectionism: Not only perfect execution on a plan but also no surprises that invalidate the plan.

The former is unrealistic.

The latter is impossible.

Previous track records of perfect results become increasingly impossible to hit over time; the likelihood of at least one slip-up grows geometrically.

Everyone thinks they are special, because they have a special, situated perspective on the world.

"This one is special because it is mine."

But "this is special because it's mine" won't convince anyone else.

You think it's special because it's yours... but no one else thinks it's special for that reason!

Some things really are special and distinctive.

But it's very easy to accidentally fall into this trap of thinking the thing you have in your corner is globally special when it's actually not.

Companies fundamentally must believe they are good (or at the very least not actively bad)... otherwise why would employees choose to work there?

So companies will almost always moralize being disrupted. "It's a shame, the (disrupting tech) is terrible for people in (X forced logic way)"

No matter how galaxy-brained someone appears to be, they are situated people in the earth with biases and skin in the game.

And smart people can create plausible derivations of their behavior for things they morally don't like very quickly.

The test of if the argument works is: "would random people who don't work here find the argument convincing?"

"the users will like it, we're the good guys!" / "every company says they're the good guys!"

The gardener can plant another plant, or trim any given existing plant; they can change their bets and adapt.

But the plant is just that particular plant. It made its bet and just needs to do the best darn job at it and hope to survive.

The main idea is to embrace and redirect.

If a kid asks for skittles before dinner, say, "I love skittles! I wish I could eat skittles all day.. but unfortunately they aren't nutritious, so we can't eat them before our dinner."

You might think "that's great advice for kids, but not for me, because I'm an adult."

But as many layers as we have around ourselves, accumulated over years of maturation, there is still a child-like core.

Kids are in some ways just very pure, very un-adulterated adults.

Even as mature adults, we all have emotions.

Have some compassion for yourself.

The worst thing to do is to try to box away your emotions and pretend you don't have them.

You definitely have them, and if you try to pretend they don't exist they will squeeze out in weird self-destructive places.

Don't let it control you, but don't pretend it doesn't exist either!

Ride the elephant.

Surf the wave.

It is likely not what you think it is... or perhaps even directly at odds with your stated goal.

And when you are outside you are too removed from it to sense how it really works, to be able to touch it to sense it and change it.

To have great results you need both a broader perspective outside the system, and a situated perspective within the system.

You don't have to consciously understand a strategy you are intuitively following--especially if it's a survival strategy you've picked up over multiple rounds in a given context.

Of course, the "strategy" you are intuitively following might turn out to not be a strategy at all.

For example, you could get lucky with a thing once, and then just keep doing what you were doing.

The things you were doing might be totally unrelated to the outcome (and maybe even somewhat self-defeating!)

Sometimes the unimportant trappings of strategy that you learned by happenstance become ritual totems.

Similar to in professional sports how many top athletes have intense luck rituals.

(I would describe this kind of person as a Radagast).

An illuminator believes in you, even if you don't believe in yourself.

They inspire you to grow into your potential: a potential you didn't dare dream because it felt self-aggrandizing or even dangerous ("what if I don't live up to that expectation?").

Illuminators believe that all people are intrinsically valuable, and have a valuable perspective on the world.

They are willing to embrace even weird ideas and connect the dots to synthesize big insights out of those weird components.

Their goal is to find the seed of greatness in every person they talk to (a seed the person might not even realize themselves) and help them see it and grow it.

When someone tells them what is hard in their context, an illuminator never assumes they are whining.

They assume the other person is competent and there is a legitimate constraint they are running into.

By embracing that perspective, they are able to learn from it, instead of dismissing it.

The result is illuminators can discover the thicket of real-world constraints... and sometimes identify ideas that don't look that special, but unlike all of the other ideas, actually fit within all of the constraints.

There are two distinct uses for LLMs that pull in very different directions:

convergent mode ("spackle for toil")

divergent mode ("a muse that super-charges your creativity")

Convergent wants to pull whatever you're doing towards the centroid (professional, safe).

Divergent wants to push you away from the centroid (quirky, random).

In the former, hallucinations are a bug; in the latter, hallucinations are a feature.

In the former mode, the downside of hallucinations is not capped. In the latter mode, there is no downside of hallucinations.

These two different modes require very different orientations, tools, mindsets, etc.

They can get away for longer with not having to hire a given expertise, because they can do a "good enough" version of surprisingly high quality from people they already have working with AI.

AI-native organizations will need generalists more; people who can prompt the LLM and judge if the result is good enough, as opposed to people who are amazing at one particular expertise.

Because organizations can get away with fewer people, they will be able to be more nimble for a given magnitude and scope of output and impact,

Nimbleness is mainly determined by number of people that have to coordinate.

Inside-out things are alive. Outside-in things are engineered.

Humberto Maturana would describe these as:

Autopoietic: self building, inside-out.

Allopoietic: other-building, outside-in.

Autopoietic things tend to be self-maintaining They resist entropy on their own.

Allopoietic things cannot be self-maintaining; they need the effort of others to maintain them.

This means that autopoietic systems are the ones that build allopoietic systems.

The only local pockets of order in the universe, that successfully (and temporarily!) resist the ravages of entropy, are autopoietic systems.