Bits and Bobs 12/16/24

1LLMs are best used as a thought partner, not as an oracle.
  • LLMs are best used as a thought partner, not as an oracle.
    • They will often be wrong; you should assume anything they tell you is likely to be approximately right, but perhaps wrong in the details.
    • But if you use them as a generative thought partner, they can be very generative!
2LLMs fill in underspecified parts of the users' request with frog DNA.
  • LLMs fill in underspecified parts of the users' request with frog DNA.
    • The frog DNA is inherently mushy; average.
    • This means that the under-specified parts become more average, pulling toward mediocrity.
    • That's bad… but also, without that frog DNA filling in the gaps, the idea might not have been fully specified enough to be viable.
    • It's better to have a viable request that you can interact with the results and tweak and specify better, than to get a brick wall of "sorry, does not compute".
3I like magic as a frame for things that are powered by LLMs.
  • I like magic as a frame for things that are powered by LLMs.
    • Normal programming is mechanistic.
      • It does exactly what it was told to do, even if that's not exactly what the creator meant.
    • But LLM-powered experiences have some inherent squishness.
      • They fill in the underspecified parts with frog DNA, reasonable guesses.
      • Even with the fully specified parts, sometimes they just… forget.
      • LLMs are not deterministic.
        • Well, technically if given the precise same seed and precise prompt they are in the strictest sense, but that is rarely how they're used in practice.
    • Using the frame of magic helps users see it as a thing that could be great, but could also just misfire in an unexpected or weird way.
4LLMs are like an absurdly good lossy compression scheme for knowledge.
  • LLMs are like an absurdly good lossy compression scheme for knowledge.
    • They can absorb echoes of everything that was thrown at them during training.
    • If you throw enough at them, they can echo back things at almost full fidelity.
    • LLMs are hyper compressed knowledge, with the answers right there, just waiting for the right question to come along and pluck them out of the model's hologram of memory.
5LLMs are like chainsaws, allowing you to hack and slash through dense problems.
  • LLMs are like chainsaws, allowing you to hack and slash through dense problems.
    • Before those problems might have been dense, foreboding jungles, with no good way to make progress.
    • Now you can hack and slash your way through without even breaking a sweat.
    • You might not navigate it well, but you are more likely to be able to keep making forward progress without getting stuck.
    • They might cut right through some load-bearing nuance on the way, but still allow you to make forward progress in a way you couldn't have before.
6LLMs are now good enough to be better than all but the experts in any given domain.
  • LLMs are now good enough to be better than all but the experts in any given domain.
    • Which produces a problem: how can you judge if its answer is good in a given domain if you aren't an expert?
    • Does that mean you should just not use them for any domain you aren't an expert in?
    • Of course not!
    • One of LLMs' superpowers is they are able to help you approximately navigate domains you aren't an expert in.[aov]
    • If you had a willing conversation partner who was significantly smarter than you and better read but every so often was wrong, would you not talk with them?
    • Of course not!
    • Having a cognitive partner who is stronger than you in given topics to dance with is useful.
    • LLMs are pretty good at most topics--likely better than you at many things.
7Finding great, game-changing ideas requires both generation and curation.
  • Finding great, game-changing ideas requires both generation and curation.
    • The generation function products sparks that could be great.
    • The curation function figures out which sparks might actually catch.
    • You need both the generation and the curation to actually find great ideas.
    • Different people excel at different parts of this process.
    • LLMs, it turns out, are pretty good at generating lots of ideas, of riffing and coming up with various options when given a frame.
    • The superpower in this era becomes not so much generation but being able to curate.
    • To select great ideas from among the chaff.
    • Judging them, applying your calibrated taste.
8LLMs should be good at generating possible multi-disciplinary insights.
  • LLMs should be good at generating possible multi-disciplinary insights.
    • LLMs are worse than domain experts, but better than most everyone else nowadays.
    • But if you take a cross-domain insight, that requires expertise in two distinct domains, an LLM is likely better than any existing human.[aow]
      • The number of combinations of domains is a combinatorial explosion[aox]; there must be some sparsity to the coverage of real experts who have precisely that overlapping expertise.
    • LLMs are thus great tools for helping discover possible cross-domain insights.
    • It's not possible to use LLMs to create an insight in a domain that would be convincing and novel to a human expert in that domain.
    • But in cross-domain cases there likely isn't a human expert.
9LLMs are a good dowsing rod: is your hunch good enough to be worth digging deeper on?
  • LLMs are a good dowsing rod: is your hunch good enough to be worth digging deeper on?
    • There are lots of questions where even an approximate answer says "yes, keep digging" or "no, don't bother".
    • Before it was very hard to ask those questions and get approximate answers.
    • Now, it's quite easy!
    • You have to remember it's just approximate, no matter how confident it is.
    • LLMs don't get stuck, they always can come up with a "yes and" on your topic, which means there's always a thread to pull to continue going deeper.
    • Ask interesting questions to get non-mushy answers that are, if not right, at least thought provoking.
10LLMs don't have their ego wrapped up in any given lens.
  • LLMs don't have their ego wrapped up in any given lens.[aoy]
    • They don't have an ego in the first place!
    • Often, when there's a lens that is very tied to our identity, it is hard to put it down.
      • It's intertwined with our ego.
      • Imagine for example an economics professor, and how tightly they'd hold to the lens that they studied and wrote papers on.
    • When debating a problem, it's hard to know which lenses will be most useful to use.
    • When some of the debaters have their ego intertwined with their lens, it's hard to know if the lens is actually useful in this context… or if it would merely be damaging to their ego to put it down.
    • But when you explore ideas with an LLM, it has no ego connected to any lens.
    • LLMs are happy to put on different lenses whenever you want.
11Viability is much more precious than explainability.
  • Viability is much more precious than explainability.
    • It's much easier to explain how something that is known to work works, than to decide if something that is not known to work will work.
    • The real world is extremely hard to simulate in our heads, they simply don't have enough space to incorporate all of the swirling interlayed feedback loops, many of which are outside of us, to know which ones will be strong enough to overcome other ones.
    • But once you know it works, you can often figure out, with some study, why it works.
    • That doesn't necessarily make you much better at figuring out, before knowing if it is viable, if it will be viable.
    • You can explain lots of things that turn out to not work.
    • And some things that turn out to work are hard to explain.
12We live in a world of scarce software.
  • We live in a world of scarce software.
    • What happens if we were to live in the world of infinite software?
    • What would change?
    • Imagine a Borges-style infinite library of software.
    • For any behavior you want to execute, no matter how niche, how situated, how specific, there is a program sitting there in the library, ready to be plucked off the shelf.
    • The problem shifts from writing down the software to finding it at the right moment.
13There's a "wall" that separates non-programmers and programmers.
  • There's a "wall" that separates non-programmers and programmers.
    • Non-programmers, no matter how motivated, run into that wall and get stuck.
    • People who can become programmers, somehow, make it through that wall.
    • What is the wall?
    • One candidate is that programmers tend to think first and foremost about the schema: about the data model.
    • They then think about code as a way to extend and modify and interact with that data.
    • This is backwards from how most people think, where they want to think in terms of the behavior they want to create.
    • Programmers are able to burrow from that intention of behavior back to the implied schema underneath, and then build upwards in code until it has the behavior they want.
    • LLMs are pretty good at thinking about schemas if you know to ask them to!
14Programmers start with data and then work their way up to the UI to build it.
  • Programmers start with data and then work their way up to the UI to build it.
    • But in a data soup, it's hard to grab onto anything.
    • Data soup? More like data quick sand!
    • You want some firm ground to get your toehold and feel comfortable building from.
15An idea for a science fiction story: humanity has had a cataclysmic event that destroyed nearly all society.
  • An idea for a science fiction story: humanity has had a cataclysmic event that destroyed nearly all society.
    • But there's a big, mysterious pyramid building covered in shiny black tile.
    • If you go into the inner sanctum and ask a question, a mysterious oracle voice will answer it.
    • The oracle knows just about everything, including the arcane details to seemingly alien technology.
    • In this way the society can access the wisdom of the ancestors.
    • But you have to know the right questions to draw the most useful knowledge out of it.
    • What is the pyramid?
    • A server farm enclosed in a sealed box to protect it, powered by solar energy, running an LLM.
    • The ruined society doesn't (yet) know how to build a way to run LLMs themselves.
      • That would require millions of specialists with knowhow to produce all the inputs necessary to build chips and servers and program them, etc.
    • But if you could package up our current technology into a self-sufficient black box, you could have a kind of "seedbank" of human civilization that could survive almost any cataclysm.
16If this is a new internet, are you working at Mosaic, or are you working at Cisco?
  • If this is a new internet, are you working at Mosaic, or are you working at Cisco?
    • Cisco is very important in that world!
    • But less strategic power than the things that directly face users.
    • LLM model providers will definitely be important... and also more likely to be subterranean.
17Retconning: constructing a narrative after the fact that someone could think is a plausible explanatory story for how something got to be where it is.
  • Retconning: constructing a narrative after the fact that someone could think is a plausible explanatory story for how something got to be where it is.
    • The technical term is "retroactive continuity".
      • It originally comes from comics.
    • But I think it's a useful way to understand and develop products.
    • Imagine you were handed a pre-existing product to be the PM for.
    • Like all post-PMF products, it will be a bit of a hot mess.
    • It will have adoption and usage, and also be kind of broken or ugly in some significant way.
    • How should you develop this product? Where should you take it?
    • A step I like to take is first try to retcon its story.
    • Not how it actually came to be, in all of its messy, random glory.
    • But a story that makes it sound like all of its random little features were intentional, that listeners would find plausible.
    • This gives you the throughline of what the product is and wants to be.
    • The things that resonated with its users in practice.
    • Now that you have a narrative throughline, you can extend it into the future.
    • Extrapolating, where might it get to in a decade?
    • Then use that as your northstar.
    • Once you do that, often you realize there are some parts that felt important that you can now see are vestigial.
    • Or that some parts you thought were random appendages are actually the beginning buds of a beautiful and important blossom.
    • Now that you have the throughline, you can make it more like what it wants to be, closer to that narrative.
    • You can repeat this progress, ping-ponging back and forth, many times over multiple years.
    • Retconning helps you figure out what the product wants.
    • A product, after all, is an accumulation of millions of micro-decisions made over many years by millions of people (engineers, PMs, customers).
    • Finding the troughline, the consistent truth underneath that flowing swarm behavior allows you to see its core emergent story that is being told.
18It seems like breakthrough products are what change the game.
  • It seems like breakthrough products are what change the game.
    • But what matters most is not the product's quality but the distribution.
    • Revolutionizing distribution is often more important than just a better product.
19A key constraint for novelty: Maximally Advanced Yet Acceptable: MAYA.
  • A key constraint for novelty: Maximally Advanced Yet Acceptable: MAYA.
    • In MAYA, the "acceptable" is not something in theory, it's something in practice.
    • That is, do people find it acceptable in practice?
    • If real users of your target market don't like it, for any reason at all, then it's not acceptable.
      • You can't tell customers they should accept it if they don't.
    • Something to be careful of: a thing that customers find acceptable, but that has a low ceiling.
    • Ideally you want something that is easy to get started, and has a smooth continuous ramp in capability up to something with a very high ceiling.
20You can be a great hobbyist engineer and a terrible production engineer.
  • You can be a great hobbyist engineer and a terrible production engineer.
    • A hobbyist engineer just needs to find something scrappy that works.
      • It doesn't matter how it works, just that it works.
    • But a production engineer needs to find something that is maintainable.
      • That multiple people can collaborate on and fix bugs in, even when things are on fire.
      • That can scale and be explainable.
    • The things that make production engineering hard are of a totally different character from the things that make hobbyist engineering hard in the first place.
    • Production engineering layers a whole new skill on top of that base.
    • One that is mostly honed from direct, hands on experience of actually working with other people on production software.
21Bits being easier to transport than atoms means that the first to establish network effects wins.
  • Bits being easier to transport than atoms means that the first to establish network effects wins.
    • Once they win, they are basically unstoppable as long as they provide a minimum level of quality.
    • It's all about first mover advantage, who's most ruthless to achieve it.
    • That's one of the reasons tech is such a profitable business: it's easy to continue dominating once you win.
    • But once they win and take all, competition declines significantly.
    • How can you make it so companies have to continue competing?
22Many inefficiencies in the short term are efficient in the long term.
  • Many inefficiencies in the short term are efficient in the long term.
    • A play (as in a theatrical production) is extraordinarily inefficient compared to a movie.
    • Art is inefficient.
    • Discovery, exploration: inherently inefficient.
    • Leave the efficiency to the robots.
    • Humans are the ones who should be inefficient.
23Some inefficiency is valuable.
  • Some inefficiency is valuable.
    • Like slack in an organization, allowing it to absorb surprises without capsizing.
    • Other inefficiency is just noise, waste.
    • But critically, you can't know which is which, because it's contextual!
    • Maybe the context changes in a way it hasn't in the past and suddenly something that was previously noise is now the life-saving adaptation.
    • In an environment with power-law distributions, you can't simply say "this would have worked in the last 100 years of variance we saw" because you could very well get a wildly-out-of-distribution event.
    • The government is well suited to be inefficient in ways that are load bearing in ways no "efficient" actor would be.
    • Who else would stockpile millions and millions of masks just in case a pandemic broke out?
24"Inefficiency" is often load bearing in ways you don't understand yet.
  • "Inefficiency" is often load bearing in ways you don't understand yet.
25True competition keeps us strong.
  • True competition keeps us strong.
    • But Potemkin villages of competition are just performative.
    • The illusion of ground truthing.
    • Expensive but empty.
26Legibility begets conformity.
  • Legibility begets conformity.
27Successful things become more conservative.
  • Successful things become more conservative.[aoz][apa]
    • When you're just starting, there's no downside, but tons of upside.
    • As you start to become more successful, there starts to be significant downside, and the upside typically stays the same or gets smaller.
    • If you don't have a clear, animating ideal that is driving you, then as you get more momentum, "keeping it going as it is" fills the space.
    • Even things that started off as innovative–for example picking iconoclasts for a fellowship–starts becoming lower and lower variance, just continuing with what has worked well so far.
    • Once you have territory, you fear losing it.
    • Thucydides recounting a Corinthian speech that contrasts the Athenians who are "addicted to innovation" vs the Spartans who "have a genius for keeping what [they] have got."[apb]
28A general rule of thumb for compelling ideas: a number of different people who look at it go, "that right there could be great."
  • A general rule of thumb for compelling ideas: a number of different people who look at it go, "that right there could be great."
    • The more people who have that reaction to start, and the more diverse their backgrounds, the more likely it is to be great.
    • Similar to the rule of thumb about what things will be viral from Fil Menczer at IU's Observatory on Social Media.
      • The best predictor of a post's virality is how unrelated the early resharers are in terms of their network overlap.
    • Community Notes on Twitter is also "ideas lots of people find helpful to start" which is the same idea.
29When the time has come for an idea to be ready any seed crystal could create it.
  • When the time has come for an idea to be ready any seed crystal could create it.
    • A supersaturated solution ready for the right seed crystal to come along.
    • The smartphone was inevitable but the iPhone was not.
30Changing the world requires being both contrarian and right.
  • Changing the world requires being both contrarian and right.
    • The knife's edge between crazy ideas that could never work and a thing that once it exists everyone agrees is great.
    • Agency is about the knife's edge between contingent and inevitable.
31Agency is making the world different than it would have been without you.
  • Agency is making the world different than it would have been without you.
    • Notably agency and goodness are not necessarily aligned.
    • The guy who killed the UHC CEO is high agency... but not in a way that society likes.
    • Agency is recognizing you can make a difference.
    • You can't change the hand you were dealt but you can change how you play it.
32Agency in a zero-sum environment is not necessarily a good thing.
  • Agency in a zero-sum environment is not necessarily a good thing.
    • In a zero-sum environment, it's a push and pull between many different ideas fighting over the same turf.
    • "It should be green!" "No, it should be red!" "No, blue!"
    • Which one wins doesn't really matter in that case. It doesn't create anything more.
      • People tend to believe that their solution is special and better.
        • "This one is special because it's mine".
      • So they think they're doing a positive-sum thing for the collective organization ("I'm pushing to ship a much much better option than the others") but really what they're doing is pecking at the other mostly the same alternatives and making it less likely that other things ship.
      • Which, if everyone does it, means that despite tons of effort, very little might ship at all.
33Moral progress is often about creating a positive-sum framing out of a previously zero-sum game.
  • Moral progress is often about creating a positive-sum framing out of a previously zero-sum game.
    • Kicking the competition into a new dimension, where instead of tug of war, the competition creates something new and different and better.
    • When you switch from zero-sum to positive-sum, it's an infinite, magical difference.
    • Switching into that mode is extremely hard to do, and requires the stars to align, a kind of alchemy, the world being in just the right place to be ready for it.
34It's easier to steer a moving car than a parked one.
  • It's easier to steer a moving car than a parked one.
35A kind of nihilistic mindset that seems more common nowadays: "truth is impossible, everything is bullshit, so might as well share the thing that you like the vibes of."
  • A kind of nihilistic mindset that seems more common nowadays: "truth is impossible, everything is bullshit, so might as well share the thing that you like the vibes of."
36The interplay of hubris and humility make the world go round.
  • The interplay of hubris and humility make the world go round.
    • The world is currently not working like it used to.
    • Everyone can agree something is wrong, off, broken.
    • There's a world-wide debate right now about whether we need more hubris or more humility.
    • The side of hubris seems to be winning.
    • I, for one, think the answer is more humility.
37The answer to almost every perennial "either, or" question is fundamentally "both."
  • The answer to almost every perennial "either, or" question is fundamentally "both."
    • The "perennial" qualifier is important.
    • If it's not an enduring debate between different sides, then maybe the answer is simply one or the other and we just don't know which yet.
    • But if the debate has been going on for a long time, the longer it's gone on without a definitive answer, the more likely the answer is definitely, "both, with the ideal mix between them something that varies in different contexts."
38"If everyone simply used [my arcane, preferred technology] none of these problems would exist."
  • "If everyone simply used [my arcane, preferred technology] none of these problems would exist."
    • Any time someone's solution includes the word "simply" it probably won't work.
39Whether an idea will work and how easy it will be to identify who should get credit are orthogonal.
  • Whether an idea will work and how easy it will be to identify who should get credit are orthogonal.
    • In practice, organizations tend to prefer ideas that have a clear person to give credit to (or blame if it doesn't work).
    • Organizations would rather execute an idea that definitely won't work but who they can then definitely blame on a particular person, versus a thing that definitely works but you won't be able to say who to reward if it does.
40If you're in a novel environment you will make mistakes.
  • If you're in a novel environment you will make mistakes.
    • The most important thing is that you don't make people on the team feel like each others' enemies when that happens.
41The lower layers will keep sucking you down into them if you aren't careful.
  • The lower layers will keep sucking you down into them if you aren't careful.
    • Especially if you're trying to do something novel that is not just building off of someone else's platform, but constructing your own.
    • Your attention will sift down to the bottom and focus disproportionately at the lowest layers.
      • After all, the lower layers have significant implications for the upper layers.
    • But in novelty, it's the upper layers being viable that matters the most.
      • Rough in something good enough at the lower layers, find PMF at the upper layers, and then continuously evolve / improve the lower layers based on the real world usage.
42Tarpits can show up even in simple domains.
  • Tarpits can show up even in simple domains.
    • Some domains are inherently complex; there is no way to make progress without powering through a tarpit.
    • But some domains aren't that complex, but can become tarpits.
    • These emergent tarpits show up when you thrash.
      • The more you do it the more you get stuck in the problem.
      • There's always a rabbithole to get sucked deeper into.
      • If the path through is constantly changing, it can make it harder to make progress.
      • The longer you're stuck in the tarpit, the more stuck you become.
    • If you're searching for the perfect path through the tarpit, you might constantly second guess the current plan, and thrash to another non-perfect answer, getting stuck more in the tarpit trying to figure out the best way out.
    • Often what you need is just a good enough path to traverse the tarpit and get out as quickly as possible.
    • If you're impatient about getting out of the tarpit, you make everyone else more nervous, which makes them get more stuck in the tarpit.
43A team of smart, curious people can become auto-nerd-sniping.
  • A team of smart, curious people can become auto-nerd-sniping.
    • That is, they nerd snipe themselves.
    • There are rabbit holes left and right that they'll pull themselves down into them automatically.
    • All it takes is someone to say "I wonder…" and someone else can't help but go into it.
    • It can make what is otherwise firm ground turn into a tarpit.
44Imagine that there are invisible man-eating giants that only one person can see.
  • Imagine that there are invisible man-eating giants that only one person can see.
    • Yes, this is a plot point from Miss Peregrine's Home for Peculiar Children.
    • Let's imagine for a second that the giants, though invisible, are real and are a threat to everyone, despite being visible to only one character.
    • Imagine the person who can see them notices one sleeping in the corner of the team's conference room during a meeting.
    • Quietly, he says: "I don't want to alarm everyone, but there's an invisible giant in that corner. If we all stay calm and exit towards the hallway it will be…"
    • One of the team members, loudly: "Huh?! I don't see anything!"
    • The seer, more urgently: "SHHH!! Shhh. There is a SLEEPING. GIANT. If we are quiet we can…"
    • The team member, now louder. "I REALLY DO NOT SEE WHAT YOU'RE SAYING. WE HAVE IMPORTANT BUSINESS TO ATTEND TO…"
    • The seer, impatient, riled up, defensive, agitated, and now much louder: "WE. NEED. TO. ESCAPE."
    • The team member, frustrated, now pacing the room, steps away from tripping over the invisible giant: "It's hard to listen to you when you're being so shrill. If we could just turn to the matter of our weekly team meeting…"
    • As the seer gets more riled up, everyone gets more dismissive, making the seer even more riled up and not thinking clearly.
    • A tragic spiral.
    • This happens every day in organizations. Someone sees something they know is real but is hard for others to see.
      • It is an existential threat.
      • The more they try to point it out, the more people don't listen, and they become even more riled up, even less credible.
    • If you see an invisible giant no one else sees, you can't help but get existentially activated.
      • You get more riled up the more that people appear to not be hearing you.
    • But what if you're wrong?
    • Or the other people see a different, more dangerous giant than you do?
    • For example, what if yelling at other people and demotivating them about the existence of a smaller invisible giant is the more dangerous invisible giant?
45People tend to catastrophize even small worrying signals.
  • People tend to catastrophize even small worrying signals.
    • They don't look at where the thing is, but the velocity of the movement, and what that implies about where it could go.
      • "My 2 year old is in a phase where he hits people when he's frustrated. When he goes to preschool he won't be able to socialize with other kids, and then he'll grow into a bully, and then he'll drop out of high school, and then he'll live on the streets, homeless!"
    • When you find yourself catastrophizing, take a breath.
    • Remember where the thing actually is, which isn't that bad.
    • Ask yourself if there are tripwires you could set up that would trigger if it ever developed to a point that was actually worrying.
      • For example, set a reminder for a year from now to seek out a specialist if your toddler is unable to productively play with any other kids.
    • Then, once you've installed the tripwires… just let go!
    • Stop thinking about it.
    • Don't borrow trouble.
46In a lot of stories who's driving the story?
47When people think you're the villain, they will resist whatever you want to do.
  • When people think you're the villain, they will resist whatever you want to do.
    • Even if it was good for them!
    • Imagine being a teenager and having an overbearing father.
    • When you flip the bit and start seeing him as a villain trying to thwart you, you go against what he tells you to do.
    • If you're trying to change someone who doesn't want to change, that can make you the villain in their eyes.
    • If you find yourself in the role of the villain, you'd better hope that you're powerful enough to convince them to do it anyway (with them possibly resenting it), because if not, you'll have a very hard time nudging them.
    • It's much easier to nudge people when they're open to being nudged by you.
48Your effort to create maneuvering space for yourself creates ambiguity for people who need to rely on you.
  • Your effort to create maneuvering space for yourself creates ambiguity for people who need to rely on you.
    • In large organizations it's mostly about interdependencies across organizations that create the complexities.
    • More ambiguity allows you to navigate uncertainty... but also makes it harder for people who rely on you to plan.
49Bezos was apparently "stubborn on vision, flexible on details."
  • Bezos was apparently "stubborn on vision, flexible on details."
50The word "rational" can take on a dangerous level of power.
  • The word "rational" can take on a dangerous level of power.
    • If you think of "rational" as in some sense "perfect", that's an infinity.
    • Once an infinity is introduced into an argument, everything downstream of that is absurd.
      • Anything multiplied by infinity (other than 0) is also infinity.
    • It's possible for someone labeling something as "rational" to give it an almost religious level of power.
    • Often what counts as rational is not nearly as clear as it appears; there are lots of load-bearing things hidden in the noise and "irrationality" that actually matter in practice.
      • For example, for capturing indirect effects or time-series effects.
      • Another example: if you're playing a one-off game, defecting is often the "rational" move.
      • But one-off games are extremely rare in the real world; you'll often have repeated games, with the same, or indirectly overlapping, counterparties.
      • In those cases, defecting is not rational, even though the situation seems similar!
51When a system is working, you take it for granted.
  • When a system is working, you take it for granted.
    • You'll only hear about it if it messes up.
    • Which will make the mistakes be visible and the things it's enabling be invisible.
    • Which will lead people to seriously consider, "It has this small problem, why don't we just blow it up?"
52Model-based systems engineering is a useful tool for things like designing rockets.
  • Model-based systems engineering is a useful tool for things like designing rockets.
    • You take the time to create an extremely precise model of the physical constraints of the system.
    • This allows you to see precisely how a change (for example, the tolerances of one component) would affect other components, in a tree of effects.
    • Why isn't it used more often?
    • Because it requires significant up-front modeling.
    • This can make sense depending on:
      • How much time do you spend planning vs executing?
      • Does it have a physical manifestation that is possible to model?
      • What is the downside risk if it doesn't work?
    • This technique works in physical systems where interactions can be precisely modeled, but could not work in complex adaptive systems.
53A chaos engineering methodology only works with the backing of a senior, empowered leader.
  • A chaos engineering methodology only works with the backing of a senior, empowered leader.
    • The idea is that you enact things like chaos monkeys to randomly break things.
    • This forces everyone to design their systems to be resilient–and not just in theory but in practice.
    • But imagine trying to retroactively add this to an established, risk-averse company that hasn't previously done it.
    • The first thing the chaos monkey breaks is likely to bring down the whole system for real!
    • If the chaos monkey breaks the system, who gets in trouble, the person who created the chaos monkey or the person whose system was broken by it?
      • This is especially true for faux chaos approaches where the monkey is not random but pre-scripted by a gamemaster.
      • Similar to the trolley problem, presumably that gamemaster is easier to blame than someone who set up a chaos monkey.
        • "I wanted you to do chaos engineering, but not on the homepage on our third biggest shopping day!"
    • A lot of chaos engineering is the powerful engineering VP saying, "I will set up chaos monkey and if it knocks over your thing, I won't be fired... you will."
    • Which leads everyone to engineer defensively. But requires a person everyone knows is powerful enough for that flex to be true.
54The fastest solution to any concrete problem is a single ply specific solution.
  • The fastest solution to any concrete problem is a single ply specific solution.
    • It won't solve other problems, and it might cause more problems in the next plys.
    • But it does solve the immediate problem in front of you.
    • Organizations spend a lot of time doing duct-tape one-ply solutions on problems that were originally caused by a one-ply solution in a previous step.
      • This can recurse all the way back to the beginning of time.
    • The only way to break this chain is to think more than one ply at a time.
55When you learn how the system works, there's a chasm you have to cross.
  • When you learn how the system works, there's a chasm you have to cross.
    • When your eyes open, you realize how broken it is, and how there are no obvious silver bullets.
    • At that point, the only thing you're able to do is critique.
    • If you get stuck there, you become a kind of cynic, a nihilist, someone that people in the system see as a villain and try to route around.
    • But you must push forward, to get to the other side: to identify the agentic things that you can do to help improve the system, and how they can add up to large impacts over time.
56Insights take time and openness to unpack, especially if they are novel or unexpected.
  • Insights take time and openness to unpack, especially if they are novel or unexpected.
    • To unpack them, you must be in an open intellectual stance!
57It takes significant time for insights to percolate in a group.
  • It takes significant time for insights to percolate in a group.
    • The way it works is not that as soon as someone shares the insight it instantly realigns people.
    • It takes time for realignment to be absorbed.
    • For insights to percolate and be digested for a single person and as a group.
    • Let the insight marinate.
    • As people sit with it, they'll understand it better; they'll have more conversations with others that help it land from different angles.
    • If you get impatient about how long the insight is taking to diffuse, you'll ratchet up the impatience, and make it significantly more difficult for it to take hold.
    • Even once someone's mindset has been realigned, it takes time for their actions–and the output of their actions–to manifest it and accumulate.
      • If you had the insight and then said "why haven't they immediately produced output aligned with this new insight" you'll be very frustrated, and rile up everyone around you.
    • Be patient.
58At some point you have to trust the other person in the details.
  • At some point you have to trust the other person in the details.
    • If you had to confirm every single detail of another person's work down to the ground truth, you wouldn't be able to get anything done.
      • It's not even that it wouldn't be any faster than doing it yourself, it's that it would be much slower, because communicating via language is orders of magnitude more expensive than communicating directly through vibe-embeddings in your own head.
      • There's a combinatorial explosion of details to dig into as you pop each layer down.
      • As you dive into details, especially if you're intellectually intimidating or outrank the person, it can come across as extremely aggressive, people shut down and get defensive, which is the opposite of what you want.
        • It enervates people, makes engaging exhausting; they're more likely to disengage and try to minimize contact.
      • To realign people's minds, they have to be open, feel strong.
    • At a certain point you have to decide, "I trust you to get everything below here correct."
    • Trust is the thing that allows magic to happen, to get more output together than you could alone.
      • Trust is the shortcut that allows the group to skip extremely expensive ground truthing.
    • We all must trust everyone to some degree, or nothing would ever happen.
    • But some people are more willing to trust at a higher level of abstraction than others, or are more willing to give someone the benefit of the doubt faster.
      • Some people really need to be convinced with lots of experience with a given person at a level before they are happy to trust them.
    • If you're quick to trust people in a group, you can help inspire people to do their best work.
      • When people betray that trust–not by not being good enough, but by doing something bad faith–you have to immediately recalibrate how much trust to give them.
    • Someone who can only trust ideas, not people, would slow down everyone around them, compelling them to convince him to his preferred level of detail before anything happens.
    • Are you quick to trust or slow to trust?
      • What do you want to be?
59"If you simply trusted me, this would be easy!"
  • "If you simply trusted me, this would be easy!"
    • Trust is a choice that someone has to make, a dangerous leap of faith.
    • You can't force someone to trust you.
    • You have to earn it.
60Socialization of ideas should not be an afterthought.
  • Socialization of ideas should not be an afterthought.
    • You should not finish the idea and then socialize it as a static thing.
    • Socialization is the creative act; the place where disconfirming evidence emerges and makes the ideas better than what either party started the conversation with.
61Your conversational partner is not a podcast.
  • Your conversational partner is not a podcast.
    • Them being along for the ride is the main part of socializing ideas with them.
    • Be willing for the conversation to move a notch slower than the fastest you could comprehend, if it means the other person stays along for the ride.
    • The cadence of a conversation needs to be set by the carrying capacity of the slowest partner.
    • If you're a fast thinker and you're an expert in this domain, your clock speed will be an order of magnitude too fast for your conversational partner, overwhelming them and leaving them behind (and feeling bad and maybe even resentful).
    • You will be able to understand your own ideas an order of magnitude faster than others, because you already understand them!
62Resentment can destroy a relationship.
  • Resentment can destroy a relationship.
    • It starts as a small, annoying, secret pinch.
    • If it's not addressed, it will grow every time you see an ambiguous signal, making you resent more, distrust more, and be more primed to view future ambiguous signals as proof of what you resent.
    • Resentment can grow from a small pinch to a crushing weight making it impossible for anything to grow.
    • Once it's gotten huge it's way harder to deal with.
    • Deal with it when it's just a pinch.
63Someone who's being too aggressive can make it less likely people will come to agree with them.
  • Someone who's being too aggressive can make it less likely people will come to agree with them.
    • A quote from The Big Lebowski: "It's not that you're wrong, it's that you're an asshole."
    • If people think you're an asshole, they won't want you to be right.
    • They'll get in a defensive crouch, won't be open to how you might be right.
    • You can be an asshole without realizing it, because what matters is what your conversation partner thinks.
      • If you're being overbearing, over-flexing your formal authority, intimidating or exhausting them, they're more likely to see you as an asshole.
    • This can be a self-catalyzing toxic spiral.
    • People aren't listening to you so you push harder, making them see you more as an asshole and becoming even more closed to engaging with you.
    • As people refuse to engage you get more impatient and upset, coming across as more of an asshole.
64The most dastardly villain in our story is often ourselves.
  • The most dastardly villain in our story is often ourselves.
    • Look back on all of the failures you've struggled with over the years.
    • We always place the villain outside of ourselves.
    • A convenient vessel for blame: someone else.
    • Sit with the idea: what if your villain all along was you?
    • What if the reason you failed was because of self-sabotaging behavior that you didn't even realize you were doing?
    • When you're the villain it's even more dastardly, because it's completely hidden; we place the blame in the wrong places, because we are less likely to blame ourselves.
    • If you place the blame outside of yourself, you'll never look inward and see how you might grow.
    • Of course, we're never 100% the villain, but we're often more our own villain than we think we are.
65Companies fundamentally believe that their plan of record strategies will work.
  • Companies fundamentally believe that their plan of record strategies will work.
    • Most strategies don't work.
    • They sound plausible, but turn out to not be viable in practice.
    • While deciding which plan to execute, you want more disconfirming evidence, in order to make a better decision.
    • But once a plan is the Official Plan, you move from wanting to find disconfirming evidence to wanting to find no disconfirming evidence.
    • Most disconfirming evidence comes from trying to do the thing and discovering that it's harder than you thought.
    • But any one person who finds that might be blamed by management.
      • Management, hearing the critique of the plan: "Maybe the problem isn't the plan, but the person giving the feedback who is simply not doing a good job executing?"
    • So everyone tries really, really hard to make the plan work if it's possible, even if it's costly.
    • No one wants to be the person who raises their hand to say "I think this plan is impossible to execute" because whoever does that first might get shot.
    • So as a result a large organization shuffles forward like a zombie, attempting in vain to execute a thing that no one but the boss believes could actually work.
    • Organizations can often get stuck in this zombie shuffle!
66You can't mandate play.
  • You can't mandate play.
    • We're naturally playful, but the structure beats it out of us.
    • The muse strikes when it wants to, not when it's convenient.
    • You can't force someone to play, but you can make the space so they can.
    • The CFO wants to "keep only the useful play." But that destroys it.
    • You can't know which play is useful!
    • That's the whole point, it's open-ended and generative, done for its own sake, which gives you momentum to travel farther in idea space as you are pulled along by intrinsic feelings than you could if you were just being pushed along.
    • The people who are naturally inclined to play are junior.
    • The people who can make the space are senior.
    • Play is betting on multi ply effects, while single ply is about structure.
67It's inspiring to be around other inspiring people playing.
  • It's inspiring to be around other inspiring people playing.
    • A scenius is driven by trying to impress your peers who you admire.
    • When you have lots of different people in a scenius, it's positive-sum.
    • The more that people lean into their superpower, the better for the group.
    • Everyone at their best, together.
    • When you have two people who are very similar, it becomes a zero-sum dynamic between those two.
    • That's why it's good to find everyone's unique superpowers and emphasize those (positive-sum), versus the ways that participants are the same (zero-sum).
68I find it hard to have intellectually interesting conversations while walking.
  • I find it hard to have intellectually interesting conversations while walking.
    • I know a lot of people are exactly the opposite.
    • I find that I'm drawn, like a moth to flame, to novelty.
    • If I'm in a boring physical setting, all of the novelty to focus on is in the conversation, the intellectual dynamic in front of me.
    • When I'm out walking, especially around other people, I'm surrounded by lots of incidental environmental novelty that distracts me from the intellectual novelty.
69A useful mindset to search for disconfirming evidence is to make it a game.
  • A useful mindset to search for disconfirming evidence is to make it a game.
    • In every conversation, challenge yourself to find the hidden great insight lurking within your conversation partner's mind.
    • No conversation is too fleeting or banal to not have a great insight lurking if you look hard enough.
    • Approach each conversation as if the person might be enlightened, knowing there's wisdom to be found if you're truly open to discovering it.
    • Similar to Fred Kofman's advice in Conscious Business to treat everyone you meet as if they were the Buddha in disguise.