Optimal Stopping And Planning

Optimal Stopping

I wanted to get this idea down before I forgot it:

Optimal stopping is knowing when the best time to stop something, usually searching or researching is. I've noticed that there are other applications that could change how I approach my work.

For example, If you apply optimal stopping to using ChatGPT, then you know when it's better to keep prompting for the solution, rather than manually writing or creating it yourself. An arbitrary number would be like: after 5 prompts, if you can't get what you want, then you should do it yourself.

Optimal stopping also applies to the amount of time spent on an activity. Here is the idea:

Before working on a feature, set a predetermined amount of time that you are willing to invest in the feature. Then cut the scope of the feature down to 60% of this time. Try to complete the feature within the 60% mark. Once you hit the 60% point, re-evaluate how much longer the feature will take, and what you may need to cut or change in order to hit that point.

The general concept is to apply the optimal stopping rule to features. I find that when I'm working on a personal project, I will spend a lot of time on a feature to see if its possible but never re-evaluate or forfeit the feature. Instead I find myself sinking more and more time into trying to redeem the idea or feature.

If you also apply Parkinson's law to this idea, then you get the idea of reducing the scope and goal by 60%. Then if you also apply the idea that the last 10% of any feature is equivalent in effort to the previous 90%. You are much more likely to finish the feature within the maximum time your willing to invest in the feature.

The other side of this coin is that some features are fun to work on, or included becuase of the challenge. Which in general should take longer. The balance I'm trying to strike is being able to finish projects, while also be able to work on the features and ideas I have without falling to the time-sunk fallacy.

Footnotes:

  1. Brian Christian, Tom Griffiths: Algorithms to Live By