From 6baf71fc070b35ff037370f19241325b42d97a62 Mon Sep 17 00:00:00 2001 From: Samuel Robinson Date: Fri, 29 Aug 2025 21:01:14 +1000 Subject: [PATCH] Expanded glossary, refined prompt tips --- source/_posts/chatgpt.md | 110 ++++++++++++++++----------------------- 1 file changed, 46 insertions(+), 64 deletions(-) diff --git a/source/_posts/chatgpt.md b/source/_posts/chatgpt.md index c6952b7..a975800 100644 --- a/source/_posts/chatgpt.md +++ b/source/_posts/chatgpt.md @@ -16,13 +16,13 @@ plugins: ## Glossary -### Glossary +### AI - Prompt: input or query that guides an AI model in generating a response.[[s]](https://time.com/collections/the-ai-dictionary-from-allbusiness-com/7273982/definition-of-prompt-ai-prompt) - Tokens: the smallest units of text processed by an AI model (e.g., words, subwords, characters, and punctuation marks) that varies depending on the model and the tokenization - method[[s]](https://time.com/collections/the-ai-dictionary-from-allbusiness-com/7273998/definition-of-token/) + method.[[s]](https://time.com/collections/the-ai-dictionary-from-allbusiness-com/7273998/definition-of-token/) - Token Limit: the maximum number of tokens an AI model can process in one interaction.[[s]](https://time.com/collections/the-ai-dictionary-from-allbusiness-com/7273998/definition-of-token/) - Artificial intelligence (AI): technology that enables machines to simulate human cognition (e.g., human learning, @@ -37,78 +37,60 @@ plugins: - Jailbreak: exploiting vulnerabilities in AI systems to bypass their ethical guidelines and perform restricted actions.[[s]](https://www.ibm.com/think/insights/ai-jailbreak) +### Prompt Engineering Techniques + +- Zero-shot: prompt with no examples.[[s]](https://www.ibm.com/think/topics/zero-shot-prompting) +- One-shot: prompt with one example.[[s]](https://www.ibm.com/think/topics/one-shot-prompting) +- Few-shot: prompt with multiple examples.[[s]](https://www.ibm.com/think/topics/few-shot-prompting) +- Chain of thought: prompt guiding the model to break down complex reasoning into intermediate + steps.[[s]](https://cloud.google.com/discover/what-is-prompt-engineering?hl=en) + ## Prompt Tips ### Efficiency -- Focus on achieving the desired output efficiently with minimal words in the prompt. -- Specify general instructions at the start of a conversation or when changing direction to avoid constantly restating - them in each new prompt. - -```{.wrap} -Example: specify the role once at the start of a conversation. -``` - -- Restructure text to eliminate filler words that don’t add meaning: - -```{.wrap} -Example: “provide a breakdown" -> "provide sales breakdown”. -``` - -- Do not explicitly refer to the item being processed unless addressing a specific part of it as what you provide will - typically speak for itself: - -```{.wrap} -Example: “Filter this piece of text" -> "Filter text” -``` - -- Do not explicitly ask the AI for help: - -```{.wrap} -Example: “can you help me x”. -``` - -- Do not explicitly reference yourself unless clarifying multiple entities: - -```{.wrap} -Example: "do X for me". -Example: ”I want to x”. -``` - -- Do not be polite to the AI: - -```{.wrap} -Example: "please do X". -Example: “thanks”. -``` +- Use the latest model as it's generally more capable and easier to + prompt.[[s]](https://help.openai.com/en/articles/6654000-best-practices-for-prompt-engineering-with-the-openai-api) +- Provide general instructions at the start of conversations or when changing direction to avoid repeating them. +- Don't say polite phrases to the AI (e.g., please, thank you) as it adds unecessary computational + costs.[[s]](https://www.techradar.com/computing/artificial-intelligence/chatgpt-spends-tens-of-millions-of-dollars-on-people-playing-please-and-thank-you-but-sam-altman-says-its-worth-it?utm_source=chatgpt.com) +- Don't explicitly ask the AI for help (e.g., can you help me do x?). +- Don't explicitly reference yourself unless clarifying multiple entities (e.g., I want to, do x for me). ### Clarity -- Consider breaking up complex ideas into granular bullet points to improve clarity, avoid searching for text, and - eliminate comma lists and conjunctions (e.g., and). -- Avoid vague or close ended questions. +- Be clear, specific, and provide enough context for the model to understand what you are + asking.[[s]](https://help.openai.com/en/articles/10032626-prompt-engineering-best-practices-for-chatgpt) +- Prompt engineering often requires iterative refinement by adjusting wording, adding context, and simplifying + requests.[[s]](https://help.openai.com/en/articles/10032626-prompt-engineering-best-practices-for-chatgpt) +- Break complex tasks into smaller focused + prompts.[[s]](https://help.openai.com/en/articles/4936848-how-do-i-create-a-good-prompt-for-an-ai-model) +- Specify focus areas, highlight important aspects, and ask for multiple + options.[[s]](https://help.openai.com/en/articles/4936848-how-do-i-create-a-good-prompt-for-an-ai-model) +- Provide examples of the desired output format to improve + accuracy.[[s]](https://help.openai.com/en/articles/6654000-best-practices-for-prompt-engineering-with-the-openai-api) +- Focus on specifying what to do and not just what to + avoid.[[s]](https://help.openai.com/en/articles/6654000-best-practices-for-prompt-engineering-with-the-openai-api) +- Before writing a prompt, identify the goal and expected + output.[[s]](https://www.techtarget.com/searchenterpriseai/definition/AI-prompt) +- Provide important keywords or phrases where possible to faciliate SEO and communicate term preferences to the AI + model.[[s]](https://www.techtarget.com/searchenterpriseai/definition/AI-prompt) +- Keep prompts precise and clean by omitting unnecessary or redundant + information.[[s]](https://www.techtarget.com/searchenterpriseai/definition/AI-prompt) +- Avoid conflicting terms that might confuse the + AI.[[s]](https://www.techtarget.com/searchenterpriseai/definition/AI-prompt) +- Open-ended questions usually provide more detailed responses than [closed-ended questions] and yes/no + questions.[[s]](https://www.techtarget.com/searchenterpriseai/definition/AI-prompt) +- When in doubt, ask AI to generate prompts for + you.[[s]](https://www.techtarget.com/searchenterpriseai/definition/AI-prompt) -```{.wrap} -Example: “Do you think our site is good" -> "Explain flaws in the designs of each webpage” -``` +### Security -- Ensure prompts are specific +- Don't share sensitive information with AI as you can't be certain how it's stored, if it will be shared, of if it's + vulnerable to data + breaches.[[s]](https://www.grantthornton.com/insights/articles/advisory/2023/anticipate-cybersecurity-and-privacy-risks-in-ai) -```{.wrap} -Example: “explain our sales" -> "explain a breakdown of our sales performance for the last quarter”. -``` - -### Response Refinement - -- If the AI model misunderstands your request then specify the intended action. -- Provide sufficient context so the AI model can understand the scenario. - -### Security And Privacy - -- Never share sensitive information with AI models as you can't be certain how it's stored, whether it will be shared, - of if a data breach may occur. - -## Prompt Engineering +## Prompt Structure ### Prompt Frameworks