Term
Chain-of-Thought
Chain-of-Thought (CoT) is a prompting technique that asks the model to spell out its reasoning in intermediate steps — boosting accuracy on multi-step tasks.
Chain-of-Thought — explained in more detail
The simplest trigger is the instruction “think step by step” or a handful of examples in which the solution is derived step by step. Instead of guessing an answer directly, the model produces a sequence of partial reasoning steps in between — much like working through a maths problem or a logic puzzle on paper. On larger models this noticeably improves accuracy on reasoning benchmarks, because the model no longer has to compress multi-step thinking into a single forward pass.
Example / Practical context
For a word problem like “Anna has 3 apples, gives 1 away, and buys 2 more — how many does she have?”, a zero-shot prompt sometimes produces the wrong number, while a CoT prompt with “step by step” yields the intermediate arithmetic and reliably arrives at 4. Reasoning-oriented models such as GPT-5 or Claude Opus internalise CoT mechanics (often called “extended thinking” or “reasoning tokens”) so that users rarely need to add the trigger explicitly.
Distinction from related concepts
Few-shot prompting shows input/output examples without exposing the reasoning path — CoT exposes the path on purpose. Self-Consistency is an extension: multiple CoT paths are generated and the most frequent final answer wins. Tree of Thoughts goes further and explores multiple solution trees in parallel.
Entdecke mehr
Effort level and deep thinking: two independent axes for AI tasks
Effort scales breadth, deep thinking scales depth. When each setting makes sense — with three clear examples and one rule of thumb.
GlossarConstitutional AI
Constitutional AI is a training method developed by Anthropic in which an AI model critiques and revises its own behaviour against an explicit list of principles (a "constitution") — rather than relying solely on human ratings.
LexikonPrompt Security — Injection, Leaking, Guardrails
How prompt injection, prompt leaking, and jailbreaks work — and which defenses (guardrails, spotlighting, sanitization) actually help.