8 ideas
Effort spent organizing a collection only pays off if it reduces enough future search to justify the upfront cost. For items you rarely search, leaving them in a messy pile is mathematically rational, because the time spent sorting exceeds the time saved finding.
Modeling data too precisely captures noise rather than signal, producing predictions that fail on new cases. Deliberately using less information, stopping analysis early, or penalizing complexity can yield better real-world decisions than fully optimizing on available data.
When coordinating with others, the kind move is to minimize the computational burden you impose — offering concrete options instead of open-ended questions, or making a decision rather than deferring it. Constraints and clear choices reduce the work everyone must do to reach a solution.
When space is limited, the best thing to evict is whatever was used least recently, because recent use predicts future use. Applied to physical life, a pile where each used item goes back on top self-organizes into a near-optimal cache. On this view, forgetting and slower recall in older age partly reflect retrieval from a larger store, not simply decline.
No schedule is best in general; the right rule depends on what you are trying to minimize. Earliest Due Date minimizes the maximum lateness. Moore's algorithm minimizes how many tasks are late by dropping the largest task whenever you would miss a deadline. Constant switching between tasks can cause thrashing, where overhead consumes all the work, so batching and interrupt coalescing protect throughput.
When options arrive one at a time and rejected options cannot be recalled, spend the first 37 percent of the pool or time window only looking and commit to nothing. After that, take the first option better than everything seen so far. This maximizes the chance of landing the single best option, and it still succeeds only about 37 percent of the time. Failure is built into the problem, not a sign of a bad process.
Whether to try something new or stick with a known favorite depends on how much time remains to use what you learn. Early in an interval, exploring is worth more because new information pays off over many future choices. Late in an interval, exploiting known favorites is rational. By this reasoning, children's restless novelty-seeking and older adults' narrowing to close relationships are both optimal strategies, not flaws.
Good predictions from sparse data require knowing which kind of distribution you face. For power-law quantities such as box-office grosses, use a multiplicative rule and predict a constant multiple of the current value. For normally distributed quantities such as lifespans, use an average rule and predict close to the mean. For memoryless processes, use an additive rule and predict a constant amount more. With no prior at all, the Copernican principle says to assume you are halfway through, so predict a total of double the current value.