# Office Hours Notes #### Prepare, Don't Predict > "*Prepare*, don't *predict*. Everyone is trying to predict. Everyone is trying to optimize for the next thing. It is way better over a long period of time to prepare. You are always going to look like a fool in the moment if you prepare, because there is always somebody who has optimized for this *particular moment*. But the minute things change-and we *know* things change-preparing, not optimizing, will have set us up for success. We want to prepare for a wide range of scenarios. That means you are never going to be *optimal* in the moment, but over a long period of time it is the *optimal solution*." A very interesting note on the above concept: it is highly associated with the idea of a *local* vs. *global* optimal solution in standard statistical learning. For example, consider the following * We have an overall "space" of possible events that can happen in the world. * A subset of the entire space would be the events where a pandemic (ex. covid) occurs. * Looking at this specific subset, there is a model that can be trained to have an optimal solution. The key idea is that **this model is only optimal in this subset of data**. * Now, let us take a step back. We do not know that a pandemic will occur for sure. We want to converge upon a solution that is optimal with respect to *all* possible paths that can branch from the present. From this *perspective* we can see that the model trained on the subset of events where a pandemic occurs *is only a local optimum* across the entire space of events. * We want to find a **global optimum**. By necessity that means we will never achieve the success of a *local optimum* for any given situation. * Yet, due to the nature of the world our lives are made up of a long sequence of events and decisions. It is naive to optimize for a certain subset that could branch off from the overall. * The way to achieve a global optimum is to discard optimizing for a single circumstance or subset of events, but rather to *prepare* for as many as you can. This gives you *optionality* to pivot once knew information (knowledge) becomes available. #### A Note on Interfaces and simplicity In *[The Sciences of the Artificial](The%20Sciences%20of%20the%20Artificial.md)*, Herbert Simon discuss the idea of an internal system and an external system, where these systems interact through and *interface*. They key idea to this interface is how it *simplifies* the analysis we must perform to get at the heart of the system. For example, in the case of an organization, we may not care what the environment is that they are functioning in (bull market, bear market, downturn, upturn, etc) and we may not care about it's internal structure (what is the exact hierarchical layout of their employees, who reports to who, etc). Instead, we are are concerned with the interface: how does the organization interact with the world, what it's incentives, etc? This is an *[Explanations](Explanations.md)*. By focusing on the interface we may be able to *drastically reduce* the overall level of complexity of our analysis. This is important because it allows us to spend more time thinking about the functions and forces that actually matter. This comes back to get a [360 degree view](360%20Framing%20and%20Multiple%20Lenses.md) of the problem. #### How does the person that you want to become behave? A great thought experiment to have is to think about who it is that you want to become. Envision that person and their low level (daily) and high level (yearly) behavior. What does it look like? This can be related to [Backcasting](Backcasting.md), but most importantly it requires deep thought about who it is you really want to become. A few things that come to mind for myself: * The ability to *turn off* my working brain each day so that I can be fully present each evening and weekend. * Have unyielding energy and curiosity until my final years. * Have a realistic view of the world that is matched with an optimism based on *always doing the "right" things.*. * To expand: I want to go through each day expecting that I will encounter struggles and pain (i.e. be realistic, this happens to everyone). Yet, I want to couple that with an inner belief that I will have *prepared* for these struggles as best I can and face them honorably when they come. This requires my decisions to (in general) be correct (for me). See bullet below. * Have my decisions converge to the "right" ones for me (my the right *subjective* decisions). * This is made up of two components: 1. I want to make the decisions that I believe are *right* each day. 2. I want to frequently be ruthless and thoughtful in my assessment of what shapes my beliefs of what is *right*. * For example, in the past I would have said that I believed it was *right* to hold off on purchasing new clothes because I was buried in debt and was making 58K a year. However, recently I opted to spend several hundred dollars on new clothes, even though I technically didn't have the money set aside. This is due mainly a change in my *objective* circumstances (I now make 110K), as well as my *subjective* beliefs (as I think about transitioning towards a consulting-esque role, I want to be a professional. Part of that involves looking the part). The key idea here is that I believe each was the *right* decision based on my circumstances and beliefs at each point in time. Yet they are entirely opposite. So clearly there is not objective "right"; rather it there is an object circumstance/environment and the subjective beliefs. Making the *right* decision at a given point in time amounts to matching those subjective beliefs to the objective environment and acting accordingly. So much of this process is dependent upon *consistent updating* and *adapting* of my internal beliefs, particularly when it becomes clear that I made a *mistake*. #### A series of tiny decisions... Over time a series of tiny decisions, each reversible and inconsequential in nature, can aggregate/accumulate an *irreversibility*. #### Velocity, not Speed We generally want to achieve [Velocity over Predictability](Velocity%20over%20Predictability.md), not speed. This means we want to reduce changing direction too frequently, while also having enough speed. As it relates to startups: If you are a startup you want to make your competition (say, he more bureaucratic incumbent) run up stairs. Running up stairs is going to suck, but if you are 140 lbs and in great shape (the startup) it will be far easier than if you are 400 lbs and bloated (the large corporation). Velocity will be hard to match. #### How to tell if you are scared A sign that we are scared is when we start to use *general language*. This allows us to *hide*. If we use general enough language then we can never be *wrong*. If we are specific, and put something out into the world that conforms to the best of our knowledge and beliefs at the time, that is all we can hope to do. The next (crucial step), involves getting feedback and updating that belief system to allow for a better decision *next time*. #### Filter what you read Be sure to read from people who are as *close to the source* as possible; in other words practitioners. Note: If you are reading from two different sources and they appear contradictory, *look at the context*. What about the context is different? Can I reconcile those differences in my head? Why would one work in one case and not the other case? This allows you to then pick out *patterns*. Mind map approach for reading: * At the start of a book, pull out a blank sheet of paper. * Before reading any part of the book, create a mind map containing what you already know about the book. * Then, as you read a chapter, update the mind map with a new color of ink for that chapter and what you learned. * Then, before starting the next chapter, review your mind map. * Repeat this process until the book is done. This allows for *spaced repetition* and *reflection*. This is only done for the important books (which is really the bulk of what I should be reading...) #### Ruthlessly update you beliefs, seek disconfirming evidence One of the most healthy, effective practices that I feel that I have developed is to consistently (and to an extent, aggressively) seek a wide range of opinions, beliefs, and evidence. These all in turn help shape my beliefs. It is a very organic and iterative process, where the most crucial component stems from consistency in challenging your own beliefs. This is *hard*. It requires a great deal of mental effort to consistently ask: "How could I be wrong? How could this belief-which has served me well-actually be hurting me?". In a sense, I think that one of the most powerful things in this realm of my life is Dom. She forces me to consistently do all of the above, in the area I am least likely to. #### Time Boxing Decisions, making them quickly Consider the decision of [buying a new bed](ASAP%20and%20ALAP.md#2%201%201%20Decision%20Matrix). Some people take weeks to make this decision. Let yourself make the decision in 10 minutes; collect the most useful information (which bed has a great number of 5 star reviews, will it fit in your room, do they have a flexible return policy), and pull the trigger. Yes, you *will be wrong* at certain points with this decision making process. However, we need to think about this in the context of our entire life (see: the [law of large numbers](law%20of%20large%20numbers) and [Expected Value](Expected%20Value.md)). Let me be very clear here, you *must* do the following: * *Limit your downside* (i.e. ensure only take this approach with ALAP-consequential and irreversible-decisions) * View the fact that you are making the decision quickly as though you are *being paid with time dollars* To slightly formalize this, let us consider the situation as such: $\text{probability of making correct decision} = 0.8$ $\text{time saved making the correct decision} = 10 \text{hours}$ $\text{time lost if I make wrong decision} = 10 \text{hours}$ $\text{Number of these decisions made in my life} = \frac{100}{\text{year}} \times 60 \text{ years} = 600 \text{ decisions}$ $\text{Expected Value} = E[D] = 10 \times 0.8 + -10 \times 0.2 = 6$ So the expected value of time saved/lost for a decision (in the above context) is 6 hours. Note that we can consider that in the context of our entire life! $6 \frac{\text{hours}}{\text{decision}} \times 600 \text{ decisions} = 3600 \text{ hours} = 150 \text{ days}$ That is a *massive* amount of time saved under what are relatively conservative assumptions (likely the probability of making a correct decision is closer to 0.9). #### Consider the idea of disciplines An interesting thought: disciplines are (in a sense, artificially) sections of the world that we have broken down and separated out into unique entities. If we are trying to then be multidisciplinary we will try and reconnect them from the bottom up. The problem is that this *reductionist* approach, followed by trying to "glue" everything back together means that we have have lost *interconnections* that were present at the top level (in actual physical reality). By starting from a reductionist perspective, we actively *cut* interconnections and break things down into smaller unique entities, in order to understand them. We then try and place them back within the context of the whole, but is it any surprise that this may leave out crucial detail? Note the relation of this to [the science of the artificial, in chapter 1](The%20Sciences%20of%20the%20Artificial.md). Simon talks about the building up of ideas *from the top down*. This works because of different levels of complexity and abstraction between each layer (i.e. between particle physics, nuclear physics, chemistry, molecular biology, and biology). These different levels of complexity act as an *[Interface](Interface.md)*, insulating one level from the others. #### To be Antifragile Have a niche to fall back on-something you absolutely excel at-while also being a generalist. In other words, let your knowledge have the structure of a T (broad at the top, and going deep in a particular area). To not have a niche to fall back is extremely fragile since it makes you less employable.