I’m going to start gathering prompts I’ve found useful onto pages. You can check out the Machine Learning or LLM Prompts categories for more.
Reading Assistance
(You’re reading a paragraph with long sentences, and it’s getting confusing.)
(ChatGPT 3.5) Break the text below into paragraphs, one per sentence, and number them. Print the sentences. Then, in a second section, summarize each sentence in 10 words or less, and list each with numbers.
(You’re reading history, and wondering about a couple people in the text. Maybe you’ve got them confused or conflated two people.)
(ChatGPT 3.5) Was there a relationship between ____ and ____? If so, what was the time frame, nature of the relationship, and what political or social effects did this relationship affect?
(ChatGPT 3.5) Please explain how I can learn how to read the following statistical analysis from a paper. (paste in the Statistical Analysis below)
(GPT responds with a detailed explanation of the paragraph.)
Please explain each of the statistical techniques, and give me guidance about how I can study each technique.
(Then you go study.)
Programming Assistance
Please reformat and explain this code.
Add comments for each of the functions.
create an sqlite3 statement to create a new table with fields: sku and description
Odds and Ends, and Weird Stuff
(ChatGPT 3.5) please compile a shopping list, with bullet points, organized by supermarket sections (produce, meat, dairy, snacks, canned, frozen, baking and spices, other), for the following foods which I’d like to eat this week: pigs in blankets, shabu shabu, tacos, greek salad, miso shiru, chicken in black bean sauce.
(Bing Image Creator DALL-E) Make a marvel universe character, The Mechanic, based on Booker T Washington. Then have him battle with The Illuminator, WEB DuBois, the man who can think lightning into existence.
(DALL-E doesn’t like to draw hamburgers without cheese. This prompt produced one for me, but it took a couple tries.) plain grilled beef patty on bun, no condiments, no toppings. bland, nothing. white background, value menu.
(Note that I avoided using the word “cheese”.)
(ChatGPT 3.5) I’m looking for topic or forms of prompts that are likely to produce divergent results across two LLMs.
(This writes a document explaining ways to fool the LLM. When you have divergent results when you use the same prompt on two LLMs, one of them is likely to be wrong. If you have only one LLM, each of these techniques might produce a wrong result.)