Stupid questions and mental models

Inspired by Dan Luu's willingness to look stupid.

I guess also by Young's the importance of 'stupid' questions.

And also to Ben Kuhn's staring into the abyss as a core life skill and impact, agency, and taste.

One of my recent resolutions is to be better at this. At the end of Dan Luu's essay, he says that after Ben Kuhn read the initial draft, he said:

... Then tried to think about the last time I did something that made someone think I was stupid and realized I didn't have a readily available example)

Naturally, I then tried to think of the last time I asked something that made someone think I was stupid too, and couldn't really think of anything.

I definitely made a stronger effort to do this at work recently. There must not have been any super standout examples of this, but I remarked on it to Noah throughout the week.

Even if someone doesn't think you are stupid after asking the question, the act of trying to ask more "stupid" questions lets you actually just ask better questions, since it removes the pressure that clouds your ability to realize the things you don't have clarity on.

Building a mental model

In order to get better at things faster (especially relative to other people), you need to find leverage (or seek arbitrage). This means taking a course of action that other people can't or won't either because you have a competitive advantage to do it, or other people don't see it or are unwilling to do it.

e.g. don't do what everyone else is doing. don't follow in everyone's footsteps. don't take the default action.

So how do you get better at finding leverage points? I think this is all about building a good mental model.

Good mental models are frameworks that help you think and reason critically about anything.

Analogy. Imagine you are climbing a cliff. Every now and then, you want to reset your carabiner on the next hook so that if you fall, you won't fall that far. Building a mental model is like checkpointing your progress when you think. You can build abstractions that free up the working memory and make it much easier to reason about things.

Once it becomes a lot easier to reason about things within a mental model, it should become a lot more clear what you need to do in the short term, e.g. how to work on the highest leverage thing, even as your mental model updates or changes.

-- insert some good examples --

Example

Just a few months ago, most startups, but Ramp especially, were shilling ai coding super hard. The mandate was to use ai on everything - to pretty much burn as many tokens as possible and sling as much slop as possible. What could go wrong LOL?

For a while, I was experiencing symptoms from the build up of doing this:

Anxiety. I was always trying to ship as many prs as possible, and I felt like I always needed to be coding and prompting. There was a leaderboard of PRs - on a good week, I was up in the top 5, but that kind of output is really hard to sustain, not to mention, it leads to cutting a ton of corners!

Eventually, I would feel a lot of dread at work. I would start sweating even while just sitting at my desk, feel some level of decision paralysis from starting to many threads at once, and then getting frustrated once my output wasn't as high as I wanted it to be. In hindsight, it's so obvious that this is really unsustainable, and just not the right way to go about things.

Leverage is all about finding alpha - e.g. the things that other people don't know about or can't do. Leverage is very rarely about banging your head against the wall in a race to the bottom (as I realized the hard way)!

Slinging a tremendous amount of slop was also unsurprisingly causing a good amount of incidents. In an ideal world, these incidents should be quick to resolve. Everything is still fresh in your mind since you just worked on it - but not if it's slop. Slop destroys clarity and makes it extremely hard to reason about anything. It's like trying to find a book in a library but the books are just scattered in piles on the floor.

So I guess the overarching takeaways are:

• In this case, incentives were pretty clearly misaligned with what was best (for the product + my team + the business). Id also argue that it was bad for me - it was a kinda long period of time where I wasn't practicing the actual skill of forming mental models and making well-informed decisions, and slopping out specs and code. Now I have to unlearn this and re-calibrate.
• How could I have avoided this situation? I think a lot of senior eng were less affected by this. I guess my working model of why that is the case is because they likely have a lot more taste, in knowing what good vs bad looks like re code & product. In retrospect, I probably could have learned more from them in terms of how they were using AI, what they thought was still important etc. I think the alarm also should've sounded a lot sooner when I had those chronic feelings of anxiety while working. That is a major fucking red flag that should've prompted reflection sooner. If I had done so, maybe I would've come to the conclusion that the leverage in this case is to go a bit slower while everyone else is going fast (e.g. don't be afraid to look stupid / go against the grain) to build better mental models to work off of.

Amongst software companies, ai coding capabilities have now been pretty democratized. Just because it has accelerated coding, nothing has fundamentally changed about how to get leverage - it still requires a robust mental model and a lot of clarity.

Conclusion

So tl;dr being objective about what's confusing to you is important to know what questions to ask. Asking those questions helps build a better mental model. Building mental models gives you mental clarity and frees up working memory so you can reason more effectively. Being able to reason more effectively leads to finding more leverage points therefore having much higher slope in the long term.