Zero-Shot

Zero-Shot — explained in more detail

Classic machine learning models had to be retrained for every new task. Modern LLMs, by contrast, solve many tasks directly from the prompt — without any input/output examples being shown. That is exactly zero-shot: just an instruction, no demonstrations. It works especially well on broadly common tasks (classification, translation, summarisation, simple extraction) because the model has seen such patterns during pretraining. On rare, narrow, or strictly formatted tasks the hit rate drops noticeably — that is when few-shot starts to pay off.

Example / Practical context

Typical zero-shot prompt: “Classify the following customer feedback as positive, neutral or negative and return only the label.” No examples, no training data — just task and input. In evaluation pipelines, zero-shot is the default baseline: first measure how well the model performs without examples, then add few-shot or chain-of-thought selectively where quality is insufficient.

Distinction from related concepts

Few-shot prompting includes a handful of examples in the prompt and often improves accuracy significantly, but costs tokens and context. One-shot is the special case with exactly one example. Fine-tuning is a step further up the ladder: the model is actually retrained on many examples — zero-shot stays a pure prompting strategy.

Entdecke mehr

Blog

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.

Glossar

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.

Lexikon

Prompt Security — Injection, Leaking, Guardrails

How prompt injection, prompt leaking, and jailbreaks work — and which defenses (guardrails, spotlighting, sanitization) actually help.

Themenuebersicht

KI-KonzeptePrompting