Prompts
Prompt bank
The goal for this doc is to provide inspiration about what is possible and specific language for how to do it. Multiple similar prompts are welcome!
Ancient Greek vocabulary
Title/Purpose | Greek NER |
Models used | Llama3 8B quantized ollama |
Author | David Mimno |
Prompt | Identify named entities in this passage: ξυνεβεβήκει δὲ ἀπὸ τοῦ πάνυ ἀρχαίου ἑτέρων μᾶλλον Ἀθηναίοις τοῦτο. ἐπὶ γὰρ Κέκροπος καὶ τῶν πρώτων βασιλέων ἡ Ἀττικὴ ἐς Θησέα αἰεὶ κατὰ πόλεις ᾠκεῖτο πρυτανεῖά τε ἐχούσας καὶ ἄρχοντας, καὶ ὁπότε μή τι δείσειαν, οὐ ξυνῇσαν βουλευσόμενοι ὡς τὸν βασιλέα, ἀλλ᾽ αὐτοὶ ἕκαστοι ἐπολίτευον καὶ ἐβουλεύοντο: καί τινες καὶ ἐπολέμησάν ποτε αὐτῶν, ὥσπερ καὶ Ἐλευσίνιοι μετ᾽ Εὐμόλπου πρὸς Ἐρεχθέα." |
Example output | The named entities in this passage are: * Athenaios (Αθηναίοις): a group of people from Athens * Kekrops (Κέκροπος): a legendary king of Attica, considered one of the first rulers of Athens * The first kings (τῶν πρώτων βασιλέων): a reference to the early rulers of Athens, including Kekrops and others * Theseus (Θησέα): a legendary hero-king of Athens * Eleusiniotai (Ἐλευσίνιοι): people from Eleusis, a city in ancient Greece * Eumolpus (Εὐμόλπου): a legendary king of Eleusis * Erechtheas (Ἐρεχθέα): a reference to Erechtheus, a legendary hero and king of Athens |
Create labels from topic keywords
Title/Purpose | Create labels for topic keywords |
Models used | llama3 8B quantized |
Author | David Mimno |
Notes | Adding the instruction "Be concise" removed some chattiness at the beginning and an explanation of the title afterwards. Specifying the format more specifically helped consistency. |
Prompt | Be concise. Provide just a single short text title in doublequotes for the topic associated with the following words: {LDA output} |
Example output | **"Safe Language Models: Mitigating Risks and Attacks in LLMs"** |
Classification and analysis with JSON output
Title/Purpose | Text classification and information extrusion in structured format |
Models used | llama3 8B quantized |
Author | Erik Fredner |
Notes | If using the API, set "format": "json" for a less verbose answer. (API docs.) |
System prompt | Determine whether the following Jeopardy! question references literature. Respond in JSON with the following fields: ["literature", "confidence", "authors", "works"]. "literature" should be boolean. "confidence" is an integer representing your confidence: 50 is guessing; 100 is certain. "authors" is a list containing any authors referenced in the question including their birth and death dates like so: "John Milton (1608-1674)". "works" is a list containing the full titles of any works of literature referenced, e.g., "Paradise Lost". |
User prompt | Category: LITERATURE Question: The British Library says of this 19th c. man, "One of his most famous poems... is a warning about the arrogance of great leaders" Answer: Shelley |
Example output | Here's my analysis: ``` { "literature": true, "confidence": 100, "authors": ["Percy Bysshe Shelley (1792-1822)"], "works": ["Ozymandias"] } ``` The question references a specific work of literature ("One of his most famous poems... is a warning about the arrogance of great leaders"), which is widely known to be Percy Bysshe Shelley's sonnet "Ozymandias". The British Library's quote confirms this connection. Therefore, I'm 100% confident that the question references literature. |
Classification of narrative
Title/Purpose | Narrative Prediction |
Models used | GPT-4, GPT-3.5 |
Author | Joel Mire, Maria Antoniak |
Associated paper | Where do people tell stories online? Story Detection Across Online Communities |
Prompt | Guidelines: Use the guidelines below to support your decisions, but ultimately, follow your best judgment as there are many edge cases. A story describes a sequence of events involving one or more people. Stories can be fictional or real, exciting or mundane. Stories describe the experiences of one or more specific people. “People” can include animals, aliens, etc. “People” can include groups as long as these are specific groups of people that exist at a specific time and place. “People” includes the first person narrator. Stories must include multiple, specific events. These events should be sequential: one event happens, then another event happens. It’s ok if the events are narrated out of order, but there should still be a clear sequence. These events should be connected: they might be about the same people, they might be causally connected, they might describe an overall change or transformation in the state of the world, they might describe a single experience. Jumbles of events that are unordered and/or unconnected (like lists of examples) are not stories. Events are “a singular occurrence at a particular place and time.” General, repeating, isolated, or hypothetical situations, states, and actions are usually not events. Most stories are told in the past tense. Present and future tense can also be used, but the bar is higher and the narrated events need to be strongly story-like. Most events are positively asserted as occurring, but depending on the context, negative verbs can also be events when occurring at a specific time and place. Events are usually verbs but can also be nouns and adjectives. Task: Does the following text contain a story? Answer YES or NO, and then briefly explain your reasoning. Text: <INSTANCE TEXT (EXCLUDING TITLE)> Answer: |
Example output | Yes, … [reason] |
Identifying metaphorical word usage
Title/Purpose | Metaphoric word usage identification |
Models used | GPT-4o, GPT-4, GPT-3.5 |
Author | Rebecca Hicke, Ross Deans Kristensen-McLachlan |
Associated paper | SCIENCE IS EXPOLORATION: Computational Frontiers for Conceptual Metaphor Theory |
System prompt | You are a helpful assistant. You have extensive linguistic knowledge. |
Prompt | For each word in the following text, determine if it has a *more* basic contemporary meaning in other contexts than the one in the given context. For our purposes, basic meanings tend to be:
Basic meanings are not necessarily the most frequent meanings of the word. Remember to only report YES for words whose use in this context is not their most basic meaning. Examples: "I demolished his argument."
"They write about their family."
<INSERT TEXT HERE> |
Example output |
|
Classifying focalization
Title/Purpose | Classifying focalization mode |
Models used | GPT-4o, GPT-4, GPT-3.5, llama3 8B quantized, llama3 70B quantized |
Author | Rebecca Hicke, Yuri Bizzoni |
Associated paper | |
System prompt | You are a helpful assistant. |
Prompt | Please state how the given novel excerpt is focalized, or the perspective through which narrative information is provided. There are three modes of focalization:
Speech in quotation marks counts as external focalization. Only respond with one word representing the mode of focalization. EXCERPT: <INSERT TEXT HERE> MODE: |
Example output | INTERNAL |