Note: this serves as the first post in a sequence on how to efficiently become a great alignment researcher. I do not claim that I am a great researcher myself nor do I claim to have the best methods to become one, but I do hope this sequence at least marginally improves our ability to solve alignment.
Practice does not always make perfect. How you practice a skill matters, and you should consider reflecting to iteratively improve on whatever thing you are trying to get better at, e.g. being a better alignment researcher, getting better quality sleep, finding a long-term partner, maintaining great relationships, mental health, etc.
There are periods of my life where I wasted so much effort simply because I did not reflect on what I was doing in order to improve as efficiently as possible. And, in fact, by not reflecting and improving on mistakes, I simply reinforce bad habits!
There's a process I've begun to use in order to improve any skill (whether it's research, learning, getting better sleep, etc.): Kolb's Experiential Cycle. Very similar to deliberate practice (here’s a video about what people get wrong about deliberate practice).
Kolb's is something you can use every day to actively figure out what is working and what isn't. It allows you to reflect and improve on your 'mistakes.' It requires effort (an entire cycle could take from 15 minutes to 1 hour), but I believe it is worth it. In fact, I see Kolb's cycles as being a necessary component to becoming a great alignment researcher as efficiently as possible. Without something like it, you will potentially be stuck doing something that is above average in effectiveness, but not ideal.
The purpose is to use it for quickly iterating and getting better at a skill rather than letting things continue naturally with very little improvement. Without proper reflection, it could take you years to get to the same skill-level it should take you months. You could even plateau and never reach a higher skill-level you could have if you did some reflection.
Doing Kolb's cycles is a crucial component of the course I'm taking on learning efficiently and it's expected that every student in the course uses them regularly. The students who end up doing the best in school afterwards are the ones who actually make the effort to do Kolb's cycles rather than sticking to what is natural and comfortable.
Here's what Kolb's Experiential Cycle looks like:
The steps for Kolb's are as follows (note that you also become better at doing Kolb's cycles over time):
Step 1: Experience
To practice and try something new. This might be a new experience or situation.
Let's say you are trying to become good at writing high-quality posts on LW. So you try writing a high-quality post.
Step 2: Reflection
To think if something worked or not. You reflect on the new experience in light of what you already know. You're trying to reflect directly on the experience you just had.
This is where you collect as much information about the experience as possible. The insights from here will become fuel for the next step. If you don’t add enough detail here, you will find step 3 very difficult.
Reflect on questions like these
- How did you feel?
- When did you feel this way?
- If you struggled, at what points? Were there any triggers?
- How did you react?
- How did others react?
- Were there any factors that might have affected your experience before the experience? For example, perhaps you were tired, or it was during a stressful period.
- What did you do at each step? Try to draw out a step-by-step recount or chronology of events.
- How did you respond to difficulty or struggle?
- How did you feel before, during and after the event?
- I was able to write a post, though I'm not sure about the quality.
- It's a post I've been meaning to write for a long time but "never had the time."
- Besides taking months to get around to writing the post, it also took me about 10-20 minutes to write a paragraph. However, after that, I got into the flow of writing and wrote the post at a much more rapid pace.
- I heard that starting by writing an outline of the post and the questions I want it to answer is a great way to get into the flow of writing and make the post useful. Turns out it does feel like it got me into the flow of writing. And when I compare it to my previous posts, it feels like this post is better at communicating what's important.
- I also notice that I probably could have written up the post much sooner and it's possible my environment played a role. On top of that, I'm actually even busier than I was months ago, what gives?
- You feel like there was some level of anxiety about not being good enough that prevented you from starting in the first place.
Step 3: Abstraction
Taking your reflections to create ideas, questions, theories and hypotheses that you can apply further. This one is a bit harder to grasp.
In the abstraction step, we focus on generalized, transferrable and “abstracted” reasons as to why we experienced all those things we just wrote about in the reflection. So you might make the connection between a few points you reflected on and then abstract to something that is higher level that can be applied to a wider range of situations.
When looking to create abstractions, try to lean on observations you are making based on what you reflected on in step 2. You want the abstraction to be observational rather than theoretical because otherwise, you will have difficulties coming up with solid experiments in step 4 in order to initiate your next Kolb's cycle.
The following questions will be helpful
- How do I tend to act in these kinds of situations?
- What about my approach or perspective or overall strategy made me prone to making this error?
- Do I make similar mistakes in other areas as well?
- What habits do I seem to have in certain types of situations that make me behave, react or act in this way?
- Notice how the focus is on observations about yourself that transcend beyond just this single experience. There’s no point in having a great way to avoid this situation from happening again if you make the same mistake in a slightly different situation.
- "Based on my reflection, it looks like when I do x and y, then z tends to happen. After that, I start feeling anxious, which makes me respond by doing A and B."
- There are many tasks I've been meaning to get done, but they always take much longer than what I'd prefer. And then someday I just start doing the task and everything just flows.
- I find that some of the tasks tend to happen much sooner than others. Typically this seems to happen when I'm collaborating with someone else rather than doing it completely on my own.
- There's something about being curious about the task that just naturally gets me in a state where I move forward on the task. Like, having a set of questions that I'm trying to answer like, "why is this part important" or "I don't understand why everyone thinks this alignment concept is important?", I tend to be pulled forward.
- I notice that anytime I am about to put work out there (where I can be critiqued), I end up avoiding the task much more. I wonder if my anxiety of not being good enough plays a role, and then wonder about how to overcome it and what I can do to bypass those anxieties (like how starting to talk to someone might alleviate the initial anxiety of communicating with them).
Step 4: Experimentation
Taking your abstraction and applying it to something new. This experimentation will create a new experience (a new thing to try and then do a Kolb's cycle on) try and the cycle continues until you have perfected your practice. These new 'experiments' might be wrong, but you will have a better clue once you actually apply it in future Kolb's cycles.
- I think I would have likely finished this post a lot sooner if I had collaborated with someone and set a deadline for each other.
- If I want to get into the habit of writing more posts, it seems valuable to comment on other people's posts since it seems to act as a good prompt for getting started and getting me quickly into the flow of writing something that may end up as a post.
- As I'm trying to learn new things or write a post, I should try to take a few minutes to build up my curiosity by asking questions like, "why is this important?".
- I should come up with some experiments I can try to alleviate some anxieties that come with putting myself out there.
Kolb's allows you to discover what actually works
One important thing that the Kolb's Experiential Cycle does is to let you systematically explore what works for you. You are working within a framework that allows you to better integrate what you learn while also traversing your tree of optimality. While others might have some great advice, they might also have some bad advice they perpetuate that does not actually work in practice (either for you or everyone). The Kolb's cycles allow you to properly reflect on whether doing x is useful for you or if it's just something that everyone does without reflecting too much about whether it actually works.