An Algorithm For Success
- Starting point: a well-defined goal.
- Develop a theory of change. Work backwards from the goal, in concrete steps, to figure out what you can do to achieve it. There are always multiple paths.
- Use simulated annealing as meta strategy. Pick one of the possible paths at random. Identify the step that is most likely to fail. Start here.
- After each step, reevalute. Your options are either the next step on the current path or jumping to a different path you picked at random.
- Assign a likelihood to succeed to both paths given your current knowledge. Compare them but allow for jumps that seem worse with a certain probability. This makes sure you don’t get stuck in local maxima prematurely.
- Regardless of what path you picked, always start with whatever step is most likely to fail next.
- Initially do big jumps at random (almost completely disregarding which path seems better).
- Then gradually reduce the “temperature”. Only allow jumps to paths that are adjacent to your current path and reduce the probability to jumping to options that seem worse.
Written on January 12, 2024