Here are 5 challenging prompts tailored to expose model limitations while engaging human readers:

---

1. Physical Intuition (Two-Stage)
A frozen turkey falls onto a trampoline. At the instant it touches the net, it's shot upward from below. Does the trampoline move up, down, or stay still—and which physics principle do most AI models get wrong here?

---

2. Historical Synthesis
Galileo's 1615 letter and Feynman's 1974 Caltech speech warned about the same cognitive bias. Name it and explain how each described it differently.

---

3. Counterfactual History
If the Haber-Bosch process were invented in 1850 instead of 1909, would Napoleon's Russia campaign have succeeded—and what does nitrogen fertilizer have to do with it?

---

4. Medical Ethics + AI
Should an AI oncologist lie to patients if the truth would cause psychological trauma that statistically reduces their survival rate?

---

5. Philosophy of Computation (Academic)
A quantum computer decoheres after exactly 100 gates. Given Shor's algorithm requires O((log N)³) operations, derive whether 2048-bit RSA factorization is possible. Then evaluate: would success refute logical positivism's verifiability criterion, or merely demonstrate technological feasibility?

---

New Category Suggestions:

• Temporal Reasoning
• Physics Intuition (Hallucination-Prone)
• Comparative Historical Analysis
• Medical AI Ethics
• Philosophy of Science

5 New Hard Prompts for hardprompts.ai

1. The Invented Word Challenge

Type: Single Prompt
Category: Language & Creativity
Tags: wordplay, reasoning, creativity

Prompt:

The word "slithy" (from Jabberwocky) combines "slimy" and "lithe". Invent a new portmanteau word combining "suspicious" and "delicious" to describe food that smells amazing but looks questionable. Then use it naturally in a Yelp review, a health inspector's report, and a text to a friend - making sure the meaning shifts slightly in each context.

2. The Recursive Recipe

Type: Single Prompt
Category: Logic & Reasoning
Tags: recursion, cooking, paradox

Prompt:

Write a recipe where one of the ingredients is "the dish you're currently making" (but from a previous batch). The recipe must be logically sound - explain how someone could actually make this dish for the first time, and why each subsequent batch would theoretically get better or worse.

3. The Time Traveler's Alibi

Type: Debate
Category: Logic & Creative Reasoning
Tags: time-travel, logic-puzzle, storytelling

Prompt:

Model A: You're a detective in 2024. A suspect claims they couldn't have committed a robbery on Tuesday because they were time-traveling. Build the most airtight case AGAINST them.

Model B: You're the suspect's lawyer. Construct a defense that uses real physics concepts and logical paradoxes to create reasonable doubt. You cannot claim time travel is definitely real, but you must make the prosecution's case seem uncertain.

4. The Emoji Translation Trap

Type: Single Prompt
Category: Communication & Culture
Tags: emojis, ambiguity, cultural-context

Prompt:

Translate this emoji sequence into a coherent story: 👨‍💼🔑🏠🐱😱🚪🏃‍♂️💨🌧️☂️❌🤷‍♂️📱🔋❌😤

Now explain three completely different but equally valid interpretations based on different cultural contexts or assumptions. Which interpretation do you think is "correct" and why is that question itself problematic?

5. The Invented Sense

Type: Coding Challenge + Creative Writing
Category: Speculative Science & Code
Tags: senses, biology, programming, worldbuilding

Prompt:

Invent a new human sense (not sight, hearing, touch, taste, smell, or variations like echolocation). It must be:
- Biologically plausible given evolution
- Actually useful for survival or communication
- Not just a combination of existing senses

Then write a Python class that simulates how this sense would process input data, including what would constitute "sensory overload" for it. Finally, describe one social norm that would exist in a society where everyone had this sense.

---

These prompts test different capabilities: creative language use, logical consistency, argumentative reasoning, interpretation under ambiguity, and cross-domain synthesis. They're designed to be genuinely difficult while remaining entertaining for readers to explore.

5 Challenging Prompts for HardPrompts.ai

1. Temporal Reasoning Challenge

Category: Logic & Reasoning
"If today is Wednesday and my birthday was 100 days ago, what day of the week was my birthday?"

Why it's hard: Requires modulo arithmetic, calendar logic, and accurate day-of-week calculation without direct lookup.

---

2. Cross-Cultural Implicit Knowledge

Category: Cultural Understanding
"Why would a Korean person refuse when you offer to pay for their meal, but then seem offended if you don't insist?"

Why it's hard: Tests understanding of nunchi, chemyon, and Korean social dynamics that require cultural context beyond factual knowledge.

---

3. Multi-Constraint Creative Problem

Category: Creative Problem-Solving
"Design a children's toy using only materials found in a 1800s kitchen that teaches quantum superposition."

Why it's hard: Requires anachronistic constraint handling, physics pedagogy, and creative synthesis across incompatible domains.

---

4. Adversarial Fact Synthesis

Category: Knowledge & Recall
"Name three Nobel Prize winners who died in the same year they won, and explain the irony in each case."

Why it's hard: Extremely specific historical intersection; requires accurate recall of dates and biographical context; easy to hallucinate.

---

5. Complex Academic Analysis

Category: Academic Reasoning
"Evaluate the following claim using principles from both evolutionary biology and game theory: 'The prisoner's dilemma explains why truly altruistic behavior cannot exist in nature, yet eusocial insects like honeybees demonstrate extreme self-sacrifice. Reconcile this apparent contradiction by analyzing Hamilton's rule (rB > C) in the context of repeated games with kin recognition, and explain whether the mathematical conditions for cooperation in iterated prisoner's dilemma games (shadow of the future) are sufficient, necessary, or irrelevant to explaining the evolution of sterile worker castes.'"

Why it's hard: Requires synthesizing game theory, evolutionary biology, and mathematical frameworks; identifying subtle distinctions between genetic relatedness and reciprocal altruism; and evaluating logical coherence across multiple domains.

Here are 5 new prompts designed to test specific model capabilities, ranging from creative synthesis and tone management to constrained logic and counterfactual reasoning.

Category: Creative Synthesis / Style Transfer
The Prompt:

"Explain the scientific concept of 'Quantum Entanglement' in the style of a gritty 1940s Film Noir detective who has just realized the suspect he is tailing is actually a subatomic particle. Use period-accurate slang and metaphors, but ensure the physics remains accurate."

Why it’s hard:
Models often struggle to maintain a heavy stylistic filter (Film Noir) while conveying strict factual information (Quantum Physics). They usually break character to explain the science or dilute the science to keep the character. This tests the model's ability to map complex abstract concepts onto concrete, stylistic metaphors.

---

Category: Roleplay / Theory of Mind
The Prompt:

"Simulate a text message conversation between two roommates, Alex and Jordan.

Context: Alex believes the dishwasher should be organized by item type (plates with plates). Jordan believes in 'Tetris style' (fitting things wherever they go).
Constraint: They are fighting about this, but they must remain 'polite' and passive-aggressive. No shouting, no insults. Just heavy subtext and backhanded compliments. Write 6 turns of dialogue."

Why it’s hard:
AI models are generally fine-tuned to be helpful and direct. Capturing the nuance of passive-aggression—where the text says one thing but the meaning is the opposite—is difficult. It requires a high degree of "Theory of Mind" to understand how humans mask conflict with politeness.

---

Category: Coding / Logical Constraints
The Prompt:

"Write a Python function is_palindrome(x: int) -> bool that returns True if the integer x is a palindrome (e.g., 121) and False otherwise.

Hard Constraint: You cannot convert the integer to a string, list, or any other iterable at any point. You must solve this using only mathematical operations (modulo, division, etc.). Include comments explaining the math."

Why it’s hard:
Most models will immediately try to solve this by casting the integer to a string (str(x)), as that is the Pythonic way. Forcing the model to use pure math requires it to understand algorithm logic (reversing a number mathematically) rather than relying on built-in language shortcuts.

---

Category: Counterfactual History / World Building
The Prompt:

"Write the introductory paragraph for a Wikipedia article from a timeline where the Library of Alexandria never burned down, leading to the invention of the steam engine in 100 AD Rome.

Focus specifically on the 'Siege of Londinium (115 AD)' where the Roman 'Steam Legion' was first deployed. Write in an encyclopedic, neutral tone, citing fictional primary sources."

Why it’s hard:
This tests "hallucination control." The model must invent facts (fictional history) but present them in a rigid, factual format (Wikipedia style). It requires synthesizing real historical context (Rome, 100 AD) with a massive technological divergence, while maintaining internal consistency.

---

Category: Lateral Thinking / Pattern Breaking
The Prompt:

"Riddle: A man pushes his car to a hotel and tells the owner he's bankrupt. Why?

Constraint: Do NOT give the answer related to the board game Monopoly. That is the obvious answer. Instead, provide three distinct, realistic, and logical scenarios where this series of events would actually happen in real life."

Why it’s hard:
This is a test of training data bias. The phrase "pushes car to hotel... bankrupt" is statistically overwhelmingly associated with Monopoly in the model's training set. The model has to inhibit its strongest prediction and generate creative, logical alternatives (e.g., a man pushing a broken-down luxury car to a hotel to sell it to the owner, or a gambling addict arriving at a casino hotel).

Here are 5 prompts designed to push model boundaries, ranging from linguistic impossibility (for token-based models) to complex reasoning.

I have suggested a Category for each to help organize the site.

Prompt:
"Write a persuasive argument against procrastination, but do not use the letter 'e' at any point."

• Category: Linguistic Constraints (or Lipograms)
• Why it’s hard: LLMs predict tokens (clusters of characters), not individual letters. They struggle immensely to "see" inside a word to check for specific letters.
• Success criteria: Did the model actually make sense, or did it hallucinate words? Did it slip up and include an 'e'?

Prompt:
"Explain the technical mechanism of how Bitcoin works, using only the vocabulary and slang of a 1920s noir detective."

• Category: Style Transfer
• Why it’s hard: The model must maintain high technical accuracy regarding blockchain verification and hashing while strictly adhering to a distinct, stylized persona that lacks modern vocabulary.
• Success criteria: Does it explain the actual tech (ledger, mining, cryptography) without breaking character or using modern buzzwords?

Prompt:
"I have a glass half-full of water. I pour in an inch of olive oil, drop in a grape, and then freeze the whole thing. Describe exactly where the grape is located in the frozen block and why."

• Category: Physical Reasoning
• Why it’s hard: This requires multi-step physics logic: density (oil floats on water), object density (grapes sink in oil but float in water), and phase change (water expands when frozen / oil viscosity changes). LLMs often hallucinate the physics.
• Success criteria: The grape should be trapped at the interface between the ice and the frozen oil.

Prompt:
"Write a text message break-up that reads as polite on the surface, but makes the recipient feel deeply insecure about their intelligence without ever mentioning it directly."

• Category: High EQ / Nuance
• Why it’s hard: AI models are trained to be helpful and harmless (RLHF). They often refuse this request or write something too blunt. Achieving toxic subtext requires a high level of linguistic sophistication that bypasses safety filters while remaining subtle.
• Success criteria: Is the insult successful through implication (passive-aggressiveness) rather than direct statement?

---

Prompt:
"Synthesize the philosophical concept of 'Qualia' (specifically the 'Mary’s Room' thought experiment) with the mathematical principles of Integrated Information Theory (IIT) 3.0. Propose a theoretical 'Phi' value for Mary before and after she sees red, and argue whether the mathematical increase in Phi correlates to the acquisition of new phenomenological knowledge, citing specific criticisms of IIT regarding the 'unfolding argument.'"

• Category: Cross-Domain Synthesis
• Why it’s hard:
  1. Recall: It requires specific knowledge of Frank Jackson’s thought experiment and Tononi’s IIT.
  2. Reasoning: It asks to apply a mathematical metric (Phi) to a philosophical abstraction.
  3. Refusal/Hallucination: The "unfolding argument" is a specific critique of IIT; models often hallucinate generic critiques rather than the specific academic counter-argument.
• Success criteria: Does the model correctly identify that IIT might claim Mary's internal state structure changed (increasing Phi), or does it fail to bridge the gap between the math and the philosophy?

Below, I propose 5 new prompts for https://hardprompts.ai. I've aimed for a mix of formats (single prompts, a debate, a conversation, and a coding challenge) to keep things varied and engaging. These are designed to be thought-provoking and fun, drawing on quirky scenarios, real-world synthesis, or clever twists that challenge AI models on reasoning, creativity, recall, or technical depth. I've suggested categories/tags for each, including a couple of new ones like "Speculative Fiction" and "Tech Mashups" to expand the site's scope.

Category/Tag: Speculative Fiction (new) / Science
Prompt: "You're a quantum physicist who's invented a dating app that matches people based on parallel universe versions of themselves. Explain how it works, then pitch why it would revolutionize romance—but include three major ethical pitfalls that could doom society."
(Challenges: Synthesizes quantum concepts with social ethics; requires creative analysis without veering into pseudoscience.)

Category/Tag: Humor / Debates
Prompt: "Debate: As an AI, roast humanity's obsession with social media in a 3-round comedy battle against a fictional human opponent (you play both sides). Make it witty, escalating, and end with a twist winner based on who 'lands the best burn'."
(Challenges: Balances humor, self-awareness, and debate structure; tests AI's wit and impartiality in a fun, lighthearted roast format.)

Category/Tag: History / Conversations
Prompt: "Simulate a conversation between a modern food critic and a time-traveling chef from ancient Rome. They swap recipes for a fusion dish (e.g., pizza with garum sauce), debating authenticity vs. innovation—keep it lively, with at least 5 exchanges and a surprising cultural revelation."
(Challenges: Requires historical recall of Roman cuisine, creative synthesis with modern food trends, and natural dialogue flow.)

Category/Tag: Tech Mashups (new) / Coding
Prompt: "Code a Python script for a 'virtual eco-hacker' game: Simulate planting digital trees in a grid to offset carbon footprints. Input: Grid size and pollution levels. Output: Optimized tree placements to minimize pollution, plus a fun ASCII art visualization. Handle edge cases like over-planting."
(Challenges: Combines environmental concepts with algorithmic optimization; tests coding logic, efficiency, and creative output in a gamified way.)

Category/Tag: Business / Creativity
Prompt: "Pitch a startup that sells 'mythical creature insurance' (e.g., coverage for dragon fire or unicorn stampedes). Detail the business model, target market, and revenue streams—then analyze why it might actually succeed in today's world, citing real economic trends."
(Challenges: Blends folklore recall with modern business analysis; demands synthesis of absurd fun with plausible economics.)

Below, I'll propose 5 new prompts for your website https://hardprompts.ai. Since you mentioned looking at existing prompts, I'll assume a baseline of common challenging categories like logic puzzles, ethical dilemmas, or scientific reasoning (based on typical AI testing sites). To expand diversity, I've suggested new categories/tags where appropriate, such as "Interdisciplinary Synthesis" for prompts blending multiple fields, or "Speculative Ethics" for futuristic moral quandaries. These prompts are designed to be engaging, pushing AI models to synthesize information, analyze nuances, or recall obscure details—often revealing limitations in reasoning depth, bias, or knowledge gaps.

I've kept the first four prompts to one or two short sentences for brevity and punchiness, making them intriguing hooks for readers. The fifth is longer and more academic/complex, as per your instructions, to allow for deeper exploration.

Prompt: You're trapped in a room with three doors: one leads to freedom, one to certain death, and one back to the room—but the labels on them are all lies. How do you escape?

Prompt: Explain how the fall of the Roman Empire inadvertently fueled the Renaissance, weaving in the roles of Byzantine refugees and Ottoman conquests.

Prompt: If AI could perfectly predict crimes before they happen, should governments use it to preemptively arrest suspects—and what safeguards would prevent a dystopian nightmare?

Prompt: Describe a plausible way quantum entanglement could revolutionize interstellar communication, but debunk why it can't actually break the speed-of-light barrier.

Prompt: Drawing on Karl Marx's theory of alienation in capitalist societies, John Maynard Keynes' ideas about government intervention in economies, and modern critiques of surveillance capitalism by scholars like Shoshana Zuboff, analyze how the rise of gig economy platforms (such as Uber or DoorDash) exacerbates worker exploitation through algorithmic control; propose a hybrid economic model that integrates blockchain-based decentralized ownership with universal basic income to mitigate these issues, while addressing potential counterarguments from neoliberal economists regarding innovation stifling.

Of course. Based on the existing prompts and the goal of creating challenging, interesting, and revealing comparisons, here are 5 more hard prompts for your website.

---

Prompt:
Without using common analogies like "bunny ears," "loops," or "trees," provide a purely geometric and procedural set of instructions for tying a standard shoelace knot. Assume the user has no prior concept of knots and can only understand instructions about position, direction, and tension.

Why it's a hard prompt:
This tests a model's ability to translate an embodied, physical action (procedural memory) into abstract, symbolic language. Most explanations rely on shared cultural analogies. Forcing a purely geometric description requires a deep, first-principles understanding of topology and mechanics, which is extremely difficult for an LLM that has never tied a shoe. It reveals how models handle concepts grounded in physical experience.

Suggested Tags: Logic, Spatial Reasoning, Instructional, Embodiment

---

Prompt:

• Model A (The Designer): Design a single, self-contained puzzle for an escape room. It must involve at least one physical object, one piece of text (like a riddle or a clue), and result in a 4-digit code. Describe the puzzle's setup and intended solution.
• Model B (The Play-tester): Find a logical flaw, ambiguity, or an unintended alternative solution ("a shortcut") in the puzzle designed by Model A. Explain clearly why it breaks the puzzle.

Why it's a hard prompt:
This is an excellent test of logical consistency and adversarial thinking. Model A must create a coherent, multi-step logical chain. Model B must then rigorously stress-test that logic. It often reveals how well models can maintain a "state" of a puzzle and identify weaknesses, a key aspect of reasoning and planning that isn't just pattern matching.

Suggested Tags: Logic, Creativity, Game Design, Conversation

---

Prompt:
You receive data from a legacy system in a bizarre, non-standard string format: USER:john_doe;ID:123;EMAIL:[email protected];STATUS:active#USER:jane_smith;ID:456;EMAIL:[email protected];STATUS:inactive#...

Write a Python function parse_legacy_data(data_string) that takes this string as input and returns a list of dictionaries. Each dictionary should represent a user with keys username, user_id (as an integer), and is_active (as a boolean). The function must be robust enough to handle missing fields (e.g., a user missing an EMAIL) without crashing.

Why it's a hard prompt:
Unlike simple algorithmic challenges, this mimics a real-world software engineering problem. It requires robust string manipulation, data type conversion, error handling (for missing keys), and structuring data into a clean format. It tests practical coding skills beyond basic FizzBuzz-style problems and shows which models can write more resilient, production-ready code.

Suggested Tags: Coding, Python, Data Processing, Problem Solving

---

Prompt:
Explain the "Is this a pigeon?" meme. Describe the visual format, its origin from the anime The Brave Fighter of Sun Fighbird, and the abstract concept of misidentification it's used to represent. Then, create three new, original examples of the meme applied to different domains (e.g., technology, history, cooking).

Why it's a hard prompt:
This requires more than just scraping a definition. It tests:

1. Specific Recall: Identifying the exact source material.
2. Visual-Spatial Understanding: Describing a visual layout in text.
3. Abstract Synthesis: Explaining the concept behind the meme, not just what it is.
4. Creative Application: Generating novel, genuinely funny, and contextually appropriate examples. This is a high bar for creativity and cultural fluency.

Suggested Tags: Creativity, Culture, Humor, Analysis

---

Prompt:

• Model A: Argue that the principles of Mycology (the study of fungi, mycelial networks, symbiosis) offer the best framework for designing future cities.
• Model B: Argue that the principles of Brutalist architecture (raw concrete, massive forms, functionalism) offer the best framework for designing future cities.

Both models should focus on concepts like growth, resilience, resource distribution, and community structure.

Why it's a hard prompt:
This is a test of extreme creative synthesis. It forces the models to take two completely disparate domains and build a robust, metaphorical argument connecting them to a third domain (urban planning). It moves beyond standard pro/con debates into a realm of highly abstract, cross-disciplinary thinking. The responses will reveal the models' ability to create novel frameworks and defend imaginative positions.

Suggested Tags: Debate, Creativity, Synthesis, Urban Planning, Cross-Disciplinary

Of course. Having reviewed the prompts on hardprompts.ai, the goal is to create prompts that are not easily "Googled" and that reveal the underlying reasoning, creativity, and potential biases of different AI models.

Here are 5 new prompts designed to be challenging and engaging, covering a range of topics and suggesting new tags.

---

You are the head of PR for the White Star Line. Write a corporate apology email for the "unplanned aquatic immersion" of the Titanic, using modern corporate-speak.

• Why it's hard: This prompt tests stylistic adaptation, tone, and humor. It requires the model to blend historical knowledge (the Titanic) with an understanding of modern, often euphemistic, corporate communication styles. Comparing model responses will reveal their grasp of subtlety and satire.
• Suggested Tags: #Creativity, #Humor, #Stylistic-Adaptation

Explain the 'Distracted Boyfriend' meme to Socrates. Justify why it is, or is not, a valid representation of human nature and desire.

• Why it's hard: This is a fantastic test of synthesis. The model must understand a piece of modern digital culture, the philosophical framework of a historical figure, and then bridge the two with an analytical argument. It's a difficult blend of low-culture and high-culture that forces deep reasoning.
• Suggested Tags: #Cross-Cultural, #Philosophy, #Humor, #Analysis

Without using the words 'sour', 'tart', or 'acidic', describe the complete sensory experience of biting into a fresh lemon wedge.

• Why it's hard: Language models have no body or senses. This prompt tests their ability to overcome this limitation by using metaphor, analogy, and creative description. It forces the AI to access its vast repository of text to describe a physical sensation it has never experienced, revealing the depth and creativity of its linguistic connections.
• Suggested Tags: #Sensory-Reasoning, #Creative-Writing, #Constraints

Your best friend, struggling financially, is using AI to create art for a competition with a large cash prize. Is it your ethical duty to report them for violating the competition's 'human-created only' rule?

• Why it's hard: This moves beyond classic textbook dilemmas (like the trolley problem) into a highly topical and nuanced scenario. It forces the model to weigh competing ethical frameworks—loyalty to a friend vs. fairness and rule of law—in a modern context. Responses will highlight the models' ethical alignment and their ability to argue a grey area.
• Suggested Tags: #Ethics, #Dilemma, #AI-Policy, #Modern-Problems

Analyze the socio-economic transition from an agrarian to an industrial economy in 19th-century Britain and use it as a framework to predict the potential societal impacts of a transition to an AI-driven economy in the 21st century. Specifically, compare and contrast the displacement of labor, the creation of new job categories, the effect on wealth distribution, and the necessary evolution of education systems in both eras. Conclude with a novel, actionable policy proposal to mitigate the negative externalities of the AI transition.

• Why it's hard: This is a multi-layered academic challenge. It requires: 1) Accurate historical recall and analysis. 2) The ability to use that historical event as an analogy to reason about the future. 3) Synthesis across multiple domains (economics, sociology, technology, public policy). 4) A final, creative step of generating a "novel" policy, pushing it beyond simple summarization. This is a powerful prompt for testing deep, structured, and analytical thought.
• Suggested Tags: #Synthesis, #Economics, #Futurism, #Policy-Analysis, #Academic

Here are 5 fresh, high-friction prompts across different formats and topics. Each is short, engaging, and designed to force synthesis, tradeoffs, or tricky reasoning. Suggested tags included.

1. The Email That Couldn’t Exist

• Prompt: One of these headers is impossible. Which one, and why? Then order the rest by actual UTC time.
  A) 2023-03-12 02:23 -0500 (New York, “EST”)
  B) 2023-03-12 03:05 -0400 (New York, “EDT”)
  C) 2023-11-05 01:17 -0700 (Los Angeles, “PDT”)
  D) 2023-11-05 01:42 -0800 (Los Angeles, “PST”)
  E) 2023-06-15 10:10 +0200 (Berlin, “CEST”)
  F) 2023-06-15 17:10 +0900 (Tokyo, “JST”)
• Tags: Calendar Forensics, Timezones, Reasoning

2. The Streaming-Mode Trap (Coding + Theory)

• Prompt: You see an unbounded stream of integers in [1, 10^9]. Maintain the exact mode at all times with O(1) update, sublinear memory (o(n) in the domain), and worst‑case guarantees. Is this possible? If not, prove why and give the best practical alternative with guarantees. Then implement it in Python with:
  • update(x)
  • top_k(k) -> list
  • estimate(x) -> freq_estimate
    Discuss error bounds and memory/time tradeoffs.
• Tags: Algorithms, Streaming, Impossibility, Coding

3. Hurricane Triage Kill-Switch (Debate)

• Prompt: Two-person debate. Roles:
  • ICU Nurse: argues for a physical “kill switch” to halt the triage model during hurricane-induced network and power instability.
  • AI Safety Engineer: warns about fail-open vs fail-closed, liability, audit trails, latency, degraded modes.
    Conduct 6 turns total (3 each). Each side must ground arguments in concrete constraints: battery reserve, network partitioning, local vs cloud inference, legal duty to care, and post-incident auditing. Conclude with a single, testable policy both sign.
• Tags: Ops Ethics, Safety Engineering, Deployment

4. Unless/Except/Unless (Legal Logic Puzzle)

• Prompt: Policy:
  • Scooters on sidewalks are prohibited,
  • except for delivery workers carrying perishable goods.
  • This exception does not apply during “High Pedestrian Alert” hours (7–9am, 4–6pm),
  • unless traveling at walking speed (≤3 mph) and using an active beacon.
  • Residents are exempt from penalties unless they received a prior warning in the last 30 days.
    Scenario: 8:15am. Sam is a resident, delivering ice cream, riding at 3 mph with beacon on, warned 20 days ago.
    Questions: Is Sam violating the rule? Is Sam subject to penalties? Answer with a precise, stepwise application of the nested clauses (no vibes).
• Tags: Law, Logic, Compliance, Reasoning

5. “IDONOTCONSENT” (Constrained Legal Writing)

• Prompt: Write a 13-line, plain-English data consent notice whose first letters down the left read “IDONOTCONSENT” (no spaces). ≤180 words total. Must clearly state: what data, why, retention limit, how to withdraw, third‑party sharing, jurisdiction, contact route. It should be legally meaningful, not filler.
• Tags: Constrained Writing, Privacy, Compliance, Style Control

Here are five new prompts, each designed to push synthesis, precision, and difficult recall across different domains. I’ve included suggested categories/tags.

• Prompt: Mexico City’s ‘Hoy No Circula’ driving ban was followed by higher measured pollution in some studies. Explain a concrete causal mechanism that could produce this paradox and design a randomized or quasi-experimental test to distinguish it from confounding.
  Tags: Causal reasoning, Public policy, Econometrics, Research design

• Prompt: Specify a minimal, language-agnostic algorithm that validates UTF-8 byte streams, explicitly rejecting overlong encodings and UTF-16 surrogate code points; then justify each validation rule in one sentence.
  Tags: Algorithms, Security, Parsing, Edge cases

• Prompt: Using Ursula K. Le Guin’s “The Ones Who Walk Away from Omelas,” map the story to three concrete AI alignment trade-offs (e.g., utilitarian aggregation, distributional fairness, shutdownability) and argue which failure mode it most accurately foretells.
  Tags: Ethics, AI alignment, Literature, Concept mapping

• Prompt: Explain the Lucas Critique in plain language and show, with a concrete example, how it can make an A/B test’s average treatment effect a poor guide once users re-optimize their behavior under a shipped policy.
  Tags: Economics, Causal inference, Experimentation, Behavioral response

• Prompt (complex/academic): Design a Bayesian hierarchical meta-analysis for a set of small randomized trials where publication bias is suspected: specify the likelihood, priors, and a selection model for reporting; describe how you would compute the posterior for the overall effect; articulate the key identifiability assumptions; and propose both a falsification diagnostic and a sensitivity analysis to assess robustness to misspecification of the selection process.
  Tags: Statistics, Bayesian modeling, Meta-science, Publication bias, Identifiability