算法之美 指导工作与生活的算法 (Brian Christian, Tom Griffiths)（Z-Library）

【One-Line Pitch】 A practical tour of the handful of algorithms that quietly govern everyday decisions—when to stop searching, when to explore versus exploit, how to sort and schedule, and how to stay flexible in an uncertain world. Best for readers who want computer-science thinking applied to work, love, and life without wading through proofs. 【Book Arc】 - **Opening (~0%–10%)**: Optimal stopping. The secretary problem, the 37% rule, and the "look-then-leap" strategy frame when to commit—dating, hiring, selling a house, parking—and why full information changes the threshold. - **Early (~10%–32%)**: Explore vs. exploit. Multi-armed bandits, the Gittins index, "win-stay, lose-shift," A/B testing, and clinical trials show how to balance novelty against proven payoff as remaining time shrinks. - **Middle (~32%–50%)**: Sorting and caching. Big-O, bubble/insertion sort, merge sort, bucket sort, and the "larger scale, greater pain" insight, plus memory hierarchies and why forgetting is a feature. - **Late (~50%–75%)**: Scheduling, Bayes, overfitting, relaxation. Prioritizing tasks, priority inversion, Bayesian prediction, the dangers of complexity, and how to loosen an over-constrained problem. - **Ending (~75%–100%)**: Randomness, networks, game theory. Simulated annealing, packet switching and exponential backoff, congestion control, equilibrium, mechanism design, and information cascades—closing on "computational kindness." 【Key Takeaways】 - **Optimal stopping has a counterintuitive answer** (Opening): survey roughly the first 37% without committing, then take the first option better than all prior ones—accepting that even the best strategy fails most of the time. - **Information changes the rule** (Early): when you can rank candidates fully, you skip the observation phase and set a threshold, raising success from 37% to about 58%. - **Explore early, exploit late** (Early): the value of trying new things falls as remaining time shrinks; the Gittins index shows an untried option can beat a proven one. - **Sorting is fundamentally superlinear** (Middle): comparison sorting cannot beat O(n log n), and naive methods hit O(n²)—so reduce what you sort and choose the right algorithm. - **Caching and forgetting are strategic** (Middle): memory hierarchies and archival decisions trade speed against completeness; forgetting curves mirror sensible cache eviction. - **Scheduling is about priorities and interruption** (Late): deadlines, priority inversion, and context switching explain why "doing more" often means finishing less. - **Complexity invites overfitting** (Late): more data and more parameters can hurt; cross-validation and penalizing complexity are the antidotes. - **Randomness and feedback tame hard problems** (Ending): simulated annealing escapes local maxima, while exponential backoff and flow control keep networks—and relationships—stable. 【Reading Tips】 - Deep-read the opening two chapters on optimal stopping and explore/exploit; they carry the book's most actionable advice. - Skim the sorting and caching chapters if you already know Big-O—focus on the everyday analogies (socks, libraries, Hollywood sequels). - Treat the Bayes, overfitting, and relaxation chapters as a unit on decision-making under uncertainty. - Pause at each "what would I do?" example and apply the rule to a real pending decision before moving on. - The ending on networks and game theory is more conceptual; read for vocabulary (congestion, equilibrium, cascades) rather than step-by-step tactics. 【Coverage Limits】 This guide is based on stratified excerpts covering the table of contents and roughly the first half of the book in detail; later chapters (randomness, networks, game theory) are summarized from headings and partial text, so specific examples there may be thinner than the excerpts suggest.