Best Books on Overfitting

Gary Smith

3.31112 learners

In *The AI Delusion*, economist Gary Smith dismantles the hype surrounding artificial intelligence, arguing that today’s AI systems are not truly intelligent—they’re just sophisticated tools for data matching and brute-force computation. Using vivid examples from Jeopardy!’s Watson, chess-playing programs like AlphaZero, and real-world failures in medicine, stock trading, and hiring, Smith shows that machines lack human understanding: they can’t grasp context, distinguish meaningful patterns from random coincidences, or apply common sense. He exposes the dangers of data mining, from the Texas Sharpshooter Fallacy (cherry-picking data to fit theories) to overfitting models that work perfectly on old data but fail catastrophically on new inputs. Smith also warns against blind trust in black-box algorithms, which perpetuate bias, enable unfair decisions, and risk catastrophic errors in high-stakes fields like healthcare and military operations. Ultimately, he emphasizes that human reasoning, critical thinking, and common sense are irreplaceable—we should use AI as a tool, not mistake it for a thinking, feeling counterpart.

Nate Silver

3520 learners

In *The Signal and the Noise*, Nate Silver demystifies prediction by dissecting high-stakes failures and successes across fields—from the 2008 financial crisis and flawed economic forecasts to weather modeling, poker strategy, climate change, and counterterrorism. Silver reveals why so many predictions go wrong: experts confuse measurable risk with unquantifiable uncertainty, overfit models to noisy data, and cling to rigid beliefs instead of updating judgments. He argues that accurate prediction hinges on embracing probabilistic thinking, using Bayesian updating to refine conclusions with new information, combining individual expertise with group wisdom, and recognizing model limits. Packed with real-world examples, this book equips readers to cut through digital-age information overload, separate meaningful signals from distracting noise, and make smarter, resilient decisions in an uncertain world.

Brian Christian；Tom Griffiths

4.2497 learners

Algorithms to Live By, by Brian Christian and Tom Griffiths, bridges computer science and everyday life, translating core computational logic into actionable strategies for real-world decisions. From choosing a partner or apartment (using the 37% rule for optimal stopping) to managing your schedule (scheduling algorithms) and predicting outcomes (Bayesian inference), this book shows how algorithms solve messy, relatable dilemmas. It explores concepts like the explore-exploit tradeoff, cache mechanisms, and relaxation strategies to help balance uncertainty, avoid overfitting, and make smarter choices. Whether navigating career moves, optimizing productivity, or understanding social dynamics, this book proves algorithms aren’t just for coders—they’re practical tools for anyone looking to make rational, effective decisions in an unpredictable world.