Skip to content
TheChatArchiveThe Chat Archive
GeminiGemini (model unspecified)· GoogleHugging Face ShareLM (browser-extension contributions)

Is this prompt clear? """Analyze the discussion below to extract topics and their…

  1. User

    Is this prompt clear? """Analyze the discussion below to extract topics and their novelty. ## Rules: 1. Turn Splitting: A single speaker turn MUST be split into multiple records if it addresses more than one topic. 2. Handling Continuity (Novelty=False): If a speaker acknowledges, responds to, or agrees/disagrees with a topic introduced earlier, create a row for that topic and set novelty to False. 3. Handling Introduction (Novelty=True): If a speaker introduces a brand-new subject, set novelty to True. 4. Various stances to the same perspectives are considered the same topic. For example, being against or for the argument, from a perspective of young parents. 5. Turn Numbers: Use the same turn_number for all records originating from the same speaker utterance. ## Output Format: Return ONLY a JSON list of objects with these keys: ["turn_number", "speaker", "topic", "novelty"]. ## Output: ## Examples of logic for splitting and novelty: - Utterance: "ANNA: I believe a targeted excise tax on high-sugar beverages and snacks is a necessary intervention to disincentivize consumption and improve long-term health outcomes." - Row 0: {{"turn_number": 2, "speaker": "ANNA", "topic": "Health Concerns", "novelty": true}} - Utterance: "Benny: While I acknowledge the health concerns, we must consider the economic reality; these taxes are inherently regressive, disproportionately penalizing low-income families who spend a larger percentage of their earnings on food" - Row 1: {{"turn_number": 2, "speaker": "Benny", "topic": "Health Concerns", "novelty": false}} - Row 2: {{"turn_number": 2, "speaker": "Benny", "topic": "Economic Reality", "novelty": true}} - Row 3: {{"turn_number": 2, "speaker": "Benny", "topic": "Low-Income Communities", "novelty": true}} - Utterance: "ANNA: That is a valid concern, but if we earmark the revenue generated from the sugar tax specifically to subsidize fresh produce and improve nutrition programs in those same underserved communities, we could effectively offset the financial burden while steering behavior toward healthier choices." - Row 4: {{"turn_number": 2, "speaker": "ANNA", "topic": "Low-Income Communities", "novelty": false}} - Row 5: {{"turn_number": 2, "speaker": "ANNA", "topic": "Expected Revenues: Fresh Produce Subsidization, Nutrition Programs", "novelty": true}} Discussion Content: {disc_content} """

  2. Gemini Assistant

    This is a strong prompt. You have clearly defined the input, the output format, and the logic for the tricky part (boolean novelty).However, LLMs struggle with the granularity of "Topics" (e.g., is the topic "Food" or "The specific tax on soda"?). Without stricter constraints, the model might give you high-level categories for one turn and detailed sentence summaries for another.Here is an analysis of the weak points and a refined version.Critical Improvements Needed1. The "Granularity" Problem (Rule 4) In your example, one topic is "Health Concerns" (very broad) and another is "Expected Revenues: Fresh Produce Subsidization..." (very specific/long).Risk: The LLM will be inconsistent.Fix: Add a constraint on the topic label length or style (e.g., "Use short, descriptive phrases (2-5 words)").2. The Turn Numbering in Examples In your examples, every row uses "turn_number": 2.Risk: This might confuse the model. It might think 2 is a hardcoded value or a category code.Fix: Increment the turn numbers in the examples (1, 2, 3) to demonstrate that they map to the flow of conversation.3. "Global" vs. "Local" Novelty Rule 3 says "brand-new subject."Risk: If Topic A appears in Turn 1, disappears for 50 turns, and reappears in Turn 51, is it novel? Usually, strictly speaking, it is False (not novel to the whole conversation), but LLMs sometimes have a recency bias.Fix: Explicitly state that "introduced earlier" refers to the entire history of the discussion provided.4. Filtering NoiseRisk: The model might create topics for "Greetings," "Agreement," or "Farewells" (e.g., Topic: "Saying Hello", Novelty: True).Fix: Add a rule to ignore phatic expressions (social niceties) unless they contain substantive arguments.Refined PromptHere is a tighter version of your prompt. I have bolded the changes."""Analyze the discussion below to extract topics and their novelty.Rules:Turn Splitting: A single speaker turn MUST be split into multiple records if it addresses more than one distinct topic.Handling Continuity (Novelty=False): If a speaker acknowledges, responds to, or agrees/disagrees with a topic introduced anywhere previously in the discussion history, create a row for that topic and set novelty to False.Handling Introduction (Novelty=True): If a speaker introduces a brand-new subject that has not been mentioned yet, set novelty to True.Topic Granularity: - Topics should be short, descriptive labels (2-6 words).Various stances on the same perspective are considered the same topic (e.g., "Tax Impact" is the topic, whether the speaker claims it is positive or negative).Noise Filtering: Do not create records for pure social pleasantries (e.g., "Hello," "Good point," "Bye") unless they are attached to a specific argument.Turn Numbers: Use the same turn_number for all records originating from the same speaker utterance.Output Format:Return ONLY a JSON list of objects with these keys: ["turn_number", "speaker", "topic", "novelty"].Examples:Utterance: "ANNA: I believe a targeted excise tax on high-sugar beverages is necessary for long-term health."Row 0: {{"turn_number": 1, "speaker": "ANNA", "topic": "Public Health Concerns", "novelty": true}}Utterance: "Benny: While I acknowledge the health issues, we must consider that these taxes are regressive and hurt low-income families."Row 1: {{"turn_number": 2, "speaker": "Benny", "topic": "Public Health Concerns", "novelty": false}}Row 2: {{"turn_number": 2, "speaker": "Benny", "topic": "Economic Regressivity", "novelty": true}}Row 3: {{"turn_number": 2, "speaker": "Benny", "topic": "Impact on Low-Income Families", "novelty": true}}Utterance: "ANNA: That is valid, but if we use the revenue to subsidize fresh produce, we offset that burden."Row 4: {{"turn_number": 3, "speaker": "ANNA", "topic": "Impact on Low-Income Families", "novelty": false}}Row 5: {{"turn_number": 3, "speaker": "ANNA", "topic": "Revenue Allocation Strategy", "novelty": true}}Discussion Content: {disc_content} """Technical Note on Python FormattingI noticed you used double braces {{ in your examples.If you are running this via Python string formatting (f-strings): This is correct (to escape the brackets).If you are pasting this directly into a Playground/Chat interface: You should remove the double braces and just use single braces {...}, otherwise the LLM might output double braces in the final JSON, making it invalid.

  3. User

    Is the grammar correct?

  4. Gemini Assistant

    Yes, the grammar is correct and professional overall.There is just one minor phrasing improvement for Rule 4 to make it sound more natural:Current: "Various stances to the same perspectives are considered the same topic."Correction: It is better to use "stances on" or "stances regarding."Suggested Polish:"Differing stances on the same issue are considered the same topic. For example, arguments both for and against a specific policy should fall under the same topic label."Everything else is grammatically perfect.