Reasoning Models: First Understand, Then Build : A Beginner-Friendly Guide to How Reasoning AI Works before You Learn to Build It

  • 予約

Reasoning Models: First Understand, Then Build : A Beginner-Friendly Guide to How Reasoning AI Works before You Learn to Build It

  • ただいまウェブストアではご注文を受け付けておりません。 ⇒古書を探す
  • 製本 Paperback:紙装版/ペーパーバック版
  • 言語 ENG
  • 商品コード 9781809311320

Full Description

Understand what makes reasoning models different from standard LLMs, how extra inference-time compute can improve complex problem-solving, and why reasoning traces and benchmarks alone cannot guarantee trustworthy answers.

Key Features

Understand how reasoning LLMs solve multistep problems without advanced mathematics or coding
Explore how chain-of-thought, test-time compute, and reinforcement learning shape AI reasoning
Learn to question reasoning traces, benchmarks, and verifiers before trusting AI-generated answers

Book DescriptionFluent large language model (LLM) responses can mask faulty logic. When a problem requires several steps, knowing the facts is not enough; an AI system must explore alternatives, check results, and recognize when to stop. Knowing how reasoning models meet these challenges matters for demanding tasks.

In Understanding AI Reasoning Models, T.N.K.TECH provides a concept-first introduction to what changes when LLMs devote more computation to solving problems. Starting with prediction versus reasoning and the limits of chain-of-thought prompting, you'll examine self-consistency, test-time compute, and the trade-offs of longer reasoning. You'll then explore supervised learning and reinforcement learning through examples from OpenAI o1, DeepSeek-R1, and R1-Zero, before examining reasoning search, outcome and process verifiers, and the role of external tools in AI agents.

Finally, you'll compare neural reasoning with automated reasoning and neuro-symbolic AI, learn what benchmarks can and cannot measure, and consider why a convincing trace is not proof of faithful reasoning or genuine understanding. Using diagrams and accessible examples rather than code-heavy exercises, you'll gain the grounding to read AI research critically, assess claims about reliability, and make better-informed decisions about reasoning-enabled applications.What you will learn

Distinguish next-token prediction from multistep reasoning in LLMs
Explain what chain-of-thought prompting improves and where it fails
Judge trade-offs between test-time compute and self-consistency
Explain how reinforcement learning develops reasoning behaviors
Analyze how reasoning search explores and ranks solution paths in LLMs
Compare outcome and process verifiers for checking reasoning steps
Explain how tool use connects reasoning LLMs with agentic AI systems
Interpret benchmark scores and the faithfulness of AI reasoning traces

Who this book is forThis book is for software developers, AI developers, ML engineers, and data scientists seeking to understand how reasoning LLMs solve complex problems and where they fail. Solution architects, technical leaders, and students exploring AI agents, model reliability, reasoning benchmarks, and verification will also benefit. Readers interested in reinforcement learning, automated reasoning, or neuro-symbolic AI will gain a foundation for further study and informed system design. Familiarity with LLMs or generative AI is helpful; advanced mathematics and programming expertise are not required.

Contents

Table of Contents

What Is a Reasoning Model?
Why Standard LLMs Aren't Enough
Is Reasoning Really Thinking?
Chain of Thought: The Shift from Answer to Reasoning
Test-Time Compute: Why "Thinking Longer" Matters
How Do Reasoning Models Actually Learn to Reason?
Reasoning as Search: What Is the Model Actually Looking For?
The Verifier: Who Checks an AI's Reasoning?
Reasoning Models and Tool Use
Reasoning Models vs. Automated Reasoning
Reasoning Models and Neuro-Symbolic AI
Reasoning Benchmarks
Is Reasoning the Same as Understanding?
Where Are Reasoning Models Headed?

最近チェックした商品