few-shot vs. zero-shot prompting
Few-shot vs. zero-shot prompting in this material
What the terms mean hereZero-shot prompting means asking the model to perform a task with no worked examples — just the instruction itself. Few-shot prompting means including one or more example input–output pairs in the prompt to show the model what you want. In this channel's material, the contrast comes up as a deliberate choice about when examples are worth the space.The source-grounded usage: zero-shot firstNotion AI research engineer Linus Lee describes his default conversational habit as zero-shot:I actually almost never write few-shot prompts when I'm just talking to ChatGPT. I just write zero-shot and then if the output isn't exactly where I want, I'll usually ask a follow-up question.In his framing, mirroring software engineering vs. quick scripting, chatting with a model is prompt scripting: write zero-shot, then iterate — follow-up questions, resampling the same prompt, and rewriting freely — rather than front-loading examples into a single turn. [^OeKEXnNP2yA@2171] [^OeKEXnNP2yA@2714]Where the example-heavy style fitsFew-shot-style techniques still show up, but on the engineering side: Lee reserves example-based tricks like chain of density (draft, note what's missing, rewrite denser without growing longer) and putting words in the model's mouth via fake prior chat history for single-turn prompts in software and the OpenAI playground, where you can't just ask for a redo. [^OeKEXnNP2yA@4019] [^OeKEXnNP2yA@4362] The overall rule from the material: prefer rich context plus iteration over packing examples in, and remember that more context about your goals generally improves suggestions. [^OeKEXnNP2yA@2650]- Zero-shot = instruction only, no examples; few-shot = the prompt includes worked example pairs.- Lee's habit: almost never few-shot in chat — go zero-shot, then follow up if the output misses. [^OeKEXnNP2yA@2223]- Iteration (follow-ups, resampling, rewriting) replaces examples in conversational prompting.- Few-shot-style techniques live mostly in single-turn prompt engineering — playground tricks, chain of density, fabricated chat history. [^OeKEXnNP2yA@4019]