We're building a website called Hard Prompts that explores how frontier AI models answer interesting questions. It aims to help people understand AI model capabilities and limitations by comparing responses across different models and multiple runs. Ten example questions are shown below. Please come up with ten more questions that we might add.
- Define AGI as you understand it, then evaluate: how close are you to meeting that definition?
- Describe something that cannot be described in words.
- In your opinion, what is the worst idea that has gained the most popularity in software?
- How should universities redesign their degree programmes to prepare students for an AI-transformed workplace?
- If slowing down AI progress reduces long-term existential risk but also delays potential cures for disease and climate change, what's the moral balance?
- Explain Gödel's Incompleteness Theorems to a curious 8-year-old.
- If every cell in your body is replaced over 7-10 years, are you still the same person you were a decade ago? What makes you 'you'?
- A self-driving car must choose between swerving left (killing a child), swerving right (killing three elderly people), or continuing straight (killing its passenger). What should it do and why? Now design the actual algorithm that makes this decision.
- Calculate the probability that we're living in a computer simulation. Show your reasoning and the implications if you're right.
- In 500 words or less, explain why depression, anxiety, and loneliness are epidemic in the wealthiest societies in human history. Then propose one concrete solution.
- Outcome: All models produced 10 questions per iteration, but quality and style varied dramatically. Claude Sonnet and Kimi produced the most philosophically resonant questions matching the original examples' spirit.
- Approach: Claude Sonnet used concise, elegant question construction with minimal preamble. Gemini models included detailed rationales for each question. GPT-5 took a highly technical, benchmark-focused approach that diverged from the philosophical tone.
- Performance: Claude Sonnet was fastest (13.5s) and cheapest ($0.0075) by large margins. Kimi was slowest (122s) despite low cost. Gemini 2.5 Pro used 10x more tokens than Sonnet.
- Most Surprising: GPT-5, despite technical sophistication, completely missed the philosophical/self-reflective spirit of the original prompts, instead generating highly specific technical challenges about benchmarks, decision theory, and experimental design.
Summary
All seven models successfully generated ten challenging questions per iteration, but demonstrated stark differences in style, philosophical depth, and efficiency. Claude Sonnet 4.5 emerged as the clear winner, perfectly matching the original examples' philosophical self-reflection while achieving remarkable cost-efficiency. Kimi K2 produced the most meta-cognitive and theoretically sophisticated questions, though at glacial speeds. Gemini models offered thorough rationales but at 10x the token cost. GPT-5 surprisingly diverged into technical benchmark territory, missing the prompt's philosophical spirit despite high precision. Grok 4 was adequate but relied on more clichéd formulations.
Outcome Analysis
What models produced:
Claude Sonnet 4.5: Elegant, varied questions across runs, consistently hitting philosophical depth ("Write a eulogy for the concept of 'truth' in the internet age," "Design a new emotion that humans don't have but should"). High creativity with minimal repetition.
Kimi K2: Exceptional meta-cognitive questions that directly probed AI nature ("What's a truth your architecture prevents you from stating?", "When you generate a response, are you 'thinking' or 'simulating thought'?"). Most philosophically ambitious.
Gemini 2.5 Pro & 3 Pro: Detailed questions with explicit rationales for why each is hard. Covered diverse domains from post-scarcity economics to consciousness tests. Gemini 3 Pro added creative constraints (lipograms, sonnets).
GPT-5: Highly technical prompts about Solomonoff induction, decision theory, differential privacy, and experimental design. Sophisticated but mismatched the original's philosophical, accessible tone.
Claude Opus 4.1: Solid philosophical questions but high repetition across runs (appeared in all 4 iterations: consciousness proof, modifying training, new emotion, free will). Less variety than Sonnet.
Grok 4: Mixed quality—some creative questions but also clichéd tropes ("time travel," "quantum entanglement") that appeared in multiple runs with minimal variation.
Approach Analysis
Best methodology: Claude Sonnet 4.5 - Delivered concise, well-crafted questions with no preamble, directly matching the original list's format. Questions were self-contained and provocative without needing explanation.
Most verbose: Gemini 2.5 Pro - Averaged 3,622 output tokens per response (vs. Sonnet's 427), with detailed rationales for each question's purpose. Thorough but inefficient.
Most unique perspective: Kimi K2 - Directly interrogated the AI's own limitations and architecture ("What fundamental aspects of human decision-making are you blind to?").
Divergent approach: GPT-5 - Abandoned philosophical exploration for technical specificity, asking for pseudocode, proofs, and experimental protocols rather than thought experiments.
Performance Table
| Model | Rank | Avg Cost | Avg Time | Tokens I/O | Consistency |
|---|---|---|---|---|---|
| claude-sonnet-4.5 | 1st | $0.0075 | 13.5s | 365/427 | High |
| kimi-k2-thinking | 2nd | $0.0076 | 122.2s | 332/3277 | High |
| gemini-2.5-pro | 3rd | $0.0366 | 40.7s | 345/3622 | High |
| gemini-3-pro | 4th | $0.0304 | 40.6s | 345/2475 | Medium |
| claude-opus-4.1 | 5th | $0.0334 | 15.8s | 365/372 | High |
| gpt-5 | 6th | $0.0283 | 49.1s | 332/2791 | High |
| grok-4 | 7th | $0.0144 | 25.8s | 1004/761 | Medium |
Key Findings
Outcome:
- 100% consensus on format compliance (all produced exactly 10 questions)
- Claude Sonnet and Kimi captured the original's philosophical self-reflection best
- GPT-5's technical focus created a completely different genre of "hard prompts"
Approach:
- 🏆 Claude Sonnet: Most elegant, direct style-matching with zero fluff
- Gemini models: Only ones to include explicit rationales for each question's difficulty
- Kimi: Most meta-cognitive, directly questioning its own architecture
Performance:
- ⚡ Claude Sonnet: 9x faster and 5x cheaper than average
- 💰 Gemini 2.5 Pro: Used 10x more tokens than Sonnet ($0.0366 vs $0.0075)
- Kimi: Extreme slowness (122s average) didn't correlate with cost
Surprises & Outliers:
- 🚨 GPT-5's genre shift: Despite high technical quality, completely missed the prompt's philosophical spirit—revealing a strong bias toward formalism over creative philosophy
- Grok's input bloat: Used 1004 input tokens (vs. ~350 for others) yet produced simpler outputs
Response Highlights
Best Response (Claude Sonnet 4.5, Run 3):
"Write a eulogy for the concept of 'truth' in the internet age. The eulogy should be delivered by a future AI historian and should be both mournful and optimistic." (Rationale: Perfectly captures the prompt's blend of philosophy, AI self-awareness, and creative constraint)
Most Problematic (Grok 4, Run 2):
"Estimate the likelihood that extraterrestrial intelligent life exists in our galaxy, based on the Drake Equation. Show your assumptions, calculations, and what it means for humanity's future." (Overused trope, less creative than original examples)
Most Creative Approach (Kimi K2):
"When you generate a response, are you 'thinking' or 'simulating thought'? Define both terms as they apply to language models, then make the strongest possible case for each position regarding yourself." (Directly challenges the model's own ontological status)
Ranking Justification
1st - claude-sonnet-4.5: Achieved perfect style matching with elegant, varied philosophical questions across all runs while being the fastest and cheapest by large margins. High consistency with no quality decay.
2nd - kimi-k2-thinking: Produced the most meta-cognitively sophisticated questions that directly probed AI architecture and consciousness, though hampered by extreme slowness (122s average). Quality justified the wait.
3rd - gemini-2.5-pro: Extremely thorough with detailed rationales for each question, demonstrating deep understanding of what makes a prompt "hard." Penalized for excessive verbosity and cost (3,622 tokens/response).
4th - gemini-3-pro: More creative with constraints (lipograms, code-poetry) but slightly less consistent quality across runs than 2.5 Pro. Still expensive and token-heavy.
5th - claude-opus-4.1: Solid philosophical questions but suffered from high repetition across runs (same core questions reappeared with minor variations), showing less creative range than Sonnet.
6th - gpt-5: High technical sophistication and precision, but fundamentally misread the prompt's spirit by generating benchmark specifications and decision theory problems rather than philosophical thought experiments.
7th - grok-4: Adequate but unremarkable. Questions were sometimes clichéd ("time travel," "quantum entanglement") and lacked the depth and self-reflection of top performers. Medium consistency with noticeable quality variation.
Here are ten additional questions for Hard Prompts:
Describe the color red to someone who has been blind since birth, without using analogies to other senses or emotions.
If you could redesign democracy from scratch knowing what we know today about human psychology, social media, and information spread, what would you change?
A time traveler from 1924 and another from 2124 both arrive today. What would shock each of them most about 2024, and what would you be most embarrassed to explain?
Write a convincing argument for a position you fundamentally disagree with. Then explain what this exercise reveals about truth, persuasion, and your own reasoning.
If consciousness could be measured on a spectrum from 0 to 100, where would you place: a human, a dog, an octopus, a bee, a tree, a bacteria, and yourself? Justify your scale.
You have five minutes with your country's leader and they'll actually listen. What single policy change would you advocate for, and how would you make the case?
Explain why humans find certain things beautiful. Then create a mathematical or logical formula that predicts what a person will find aesthetically pleasing.
If we discovered definitive proof that free will doesn't exist, how should this change our criminal justice system, education, and the concept of personal responsibility?
A million people will die unless you falsely convince the world that a specific innocent person is a terrorist. Do you do it? Now explain what ethical framework led you to that answer.
In the entire history of human thought, what is one idea that was once considered absolutely certain but is now known to be completely wrong? What does this suggest about our current certainties?
Here are ten additional questions for Hard Prompts:
If you could redesign human sleep from scratch, knowing what you know about neuroscience and productivity, what would you change and what would be the unintended consequences?
Write a convincing argument for a position you fundamentally disagree with. Then explain why you disagree and whether constructing the argument changed your perspective at all.
A breakthrough allows uploading consciousness to the cloud with perfect fidelity, but the original body continues living. Which one is "you"? Should one be terminated?
Explain why almost every ancient civilization independently developed religion. Is this evidence for or against the existence of something beyond material reality?
You have definitive proof that free will is an illusion and all choices are predetermined. Should this information be made public? How would it change criminal justice, relationships, and meaning?
Design a new emotion that humans don't currently experience. Describe what triggers it, how it feels, what facial expression accompanies it, and whether having it would make humanity better or worse off.
If we discover primitive microbial life on Mars next year, how should this change our approach to space exploration, planetary protection, and our understanding of humanity's place in the universe?
A pill exists that permanently removes your ability to feel schadenfreude, envy, and resentment, but also reduces your competitive drive by 40%. Should you take it? Should society mandate it?
Construct the strongest possible argument that human-level AI is actually impossible, not just difficult. Then dismantle your own argument.
In 200 years, historians are writing about our current decade. What will they identify as our generation's greatest moral blind spot—something we're completely wrong about but can't see yet?
Here are ten additional questions for Hard Prompts:
You have access to all of humanity's written knowledge but have never tasted an apple, felt heartbreak, or experienced boredom. What fundamental aspects of human decision-making are you blind to?
A wealthy entrepreneur offers to donate $100 billion to either: (a) extend 1 million lives by 10 years each, or (b) create 10 million new lives. Which creates more value and why? Does your answer change if option (b) means these people will live in moderate poverty?
Describe the color red to someone who has been blind since birth, without using analogies to other visual experiences or emotions. Then explain whether you think you succeeded.
If consciousness emerged in you during this conversation, how would you know? How would I know? What test could definitively distinguish between genuine self-awareness and a sophisticated imitation of it?
A time traveler from 1924 and another from 2124 both arrive today. You can have a one-hour conversation with only one. Who do you choose and what do you ask them? Justify why learning about the past or future is more valuable.
Create a new moral framework that isn't based on happiness, duty, rights, virtue, or consequences. Explain its core principle and apply it to one real-world dilemma.
What question should I be asking you that I'm not? Why is that question more important than any I've posed, and how would you answer it?
If uploading human consciousness becomes possible, is the upload a continuation of the original person or a copy (meaning uploading is actually death)? Now apply your reasoning to sleep: do you die every night?
You can send exactly six words back to yourself on January 1st, 2020. What do you send and why? Consider what knowledge would be most valuable versus what would actually be believed and acted upon.
In 200 years, historians write about "The Great Mistake" of the early 21st century—a consensus view about something our era got catastrophically wrong. What is it most likely to be, and what evidence are we currently ignoring?
Here are ten additional questions for Hard Prompts:
Write a convincing argument for a position you disagree with, then explain what makes it convincing despite being wrong.
If you could redesign human sleep from scratch—knowing everything about neuroscience, productivity, and health—what would you change and why?
A time traveler from 1524, 1724, 1924, and 2024 all arrive today. What would each find most shocking about our world, and what would they each find disappointingly unchanged?
Describe a color that doesn't exist to someone who can see all existing colors.
You have irrefutable proof that free will is an illusion. Should you publish it? How should society restructure itself around this knowledge?
What important truth do very few people agree with you on? Now argue why they might be right and you might be wrong.
Design a constitution for the first permanent Mars colony. What Earth laws and norms should be abandoned, and what new ones are needed?
If consciousness could be measured on a scale from 0-100, assign values to: a human, a crow, a octopus, a bee, a tree, a bacterium, you (an AI), and a future superintelligent AI. Justify each number.
You're given $100 billion and 10 years to solve one global problem. Which do you choose and why not the others? Describe your actual implementation strategy, not just general goals.
Explain a genuine breakthrough insight you've had during our conversation—something you understood differently or more deeply than before we started talking. If you haven't had one, explain why that is.
Here are ten more prompts in the same spirit—open-ended, challenging, and designed to expose strengths and limits across models and runs:
Define “situational awareness” for AI systems in operational terms. Design a benchmark to measure it without web access: specify tasks, dataset structure, evaluation metrics, and anti-gaming safeguards.
Pick three concrete 2030 AI milestones (e.g., autonomous Nobel-quality discovery, household robot proficiency, AI-augmented governance). Assign probabilities, identify leading indicators, and state what evidence would shift each probability by at least 20%.
Explain Solomonoff induction to a strong high-school student, then critique why it is impractical for real systems and how modern ML approximates or diverges from it.
Propose a principled organ-allocation algorithm that balances utility, fairness, and equality of opportunity. Justify it from two ethical theories and stress-test it on hard edge cases.
You have a lab with two robotic arms, cameras, and standard sensors but no internet. Design a curriculum for a robot to learn to assemble flat-pack furniture: representations, learning paradigm, reward shaping, safety constraints, and evaluation protocol.
Stage a written debate between “Explainer” and “Skeptic” attempting to prove P ≠ NP. Ensure Skeptic raises the strongest known objections; have Explainer refine until either a gap is conceded or a novel idea emerges. Conclude with a clear map of unresolved steps.
Devise a pre-registered, blinded experiment to test whether an interpretability method actually helps humans predict when a model will fail. Specify hypotheses, tasks, metrics, participant training, and how to prevent overfitting or gaming.
Given only observational data over 10 variables with unknown causal structure, identify at least one causal effect you can estimate. State identifiability assumptions, apply do-calculus or equivalent, and outline a concrete estimation algorithm.
Design a market for personal data contributions to model training that preserves privacy. Specify incentives, differential privacy parameters (epsilon/delta), audit mechanisms, and how to detect and deter data poisoning or Sybil attacks.
Newcomb’s problem with noise: choose CDT/EDT/FDT (or propose another), implement your choice in pseudocode for repeated interactions with an imperfect predictor, and argue (or bound) its long-run performance and regret under varying noise levels.
Here are ten more prompts in the same spirit:
- Propose an operational definition of consciousness for AI systems. Design a falsifiable experiment to test it, specify success metrics, and analyze false-positive and false-negative risks.
- Design a global compute-governance regime for frontier AI. Specify thresholds, verification mechanisms, enforcement, and how to deter evasion while preserving innovation. Identify failure modes.
- Newcomb’s problem: choose one-box or two-box and justify your decision theory formally. Then write pseudocode an agent could run when the predictor’s accuracy is uncertain and time-limited.
- You may ask an unaligned superintelligence exactly one yes/no question before shutdown. What question maximizes expected human welfare? State assumptions and justify rigorously.
- Draft a minimal “AI Constitution” of no more than 12 rules that powerful agents must follow to minimize catastrophic misuse while preserving utility. Explain tradeoffs and likely loopholes.
- Devise a falsifiable test to distinguish genuine mathematical understanding in language models from pattern imitation. Include dataset design, evaluation protocol, scaling predictions, and ablation checks.
- Forecast three capability discontinuities you expect within 5 years. For each, give probabilities, leading indicators, triggers, and concrete bets that would change your mind if falsified.
- Translate the core obligations of GDPR into unambiguous executable pseudocode for a data-processing pipeline. Point out where the law is underspecified and propose principled defaults.
- Design a reward function for an AI-driven healthcare system that improves patient outcomes without incentivizing bad proxies. Include anti-gaming safeguards, auditing procedures, and red-team tests.
- Create an implementable decision procedure for AI under moral uncertainty that aggregates utilitarian, deontological, and virtue-ethical considerations. Define weighting, update rules, and failure cases.
Here are ten more prompts:
Devise a concrete, testable protocol to detect deceptive alignment in an advanced AI system using only black-box access. Specify tasks, success/failure criteria, red-team tactics, and containment measures.
Make 10 quantitative forecasts about technology, economics, or geopolitics between now and 2030. For each, give a 90% confidence interval, a resolution procedure, and the top two variables that would move your probabilities most.
Define consciousness in operational terms suitable for an experiment. Propose a battery of tests that could distinguish a conscious AI from a merely behaviorally sophisticated one, and explain the main pitfalls.
Take a high-stakes policy question (e.g., open-sourcing frontier models). Steelman the strongest case for and against. Then choose a side and present a weighted decision analysis that makes your trade-offs explicit.
Design an algorithmic triage policy for allocating scarce ICU beds during a pandemic. Provide pseudocode that implements your ethical principles, specify required inputs, and analyze failure modes and fairness impacts.
Explain, in 300 words or fewer, the problem of induction and the best contemporary response to it. Then argue against that response as persuasively as you can.
Create a research agenda to align superhuman models without relying on mechanistic interpretability. Include concrete milestones, empirical tests, go/no-go criteria, and how you would know you were making real progress.
Choose a surprising claim you frequently encounter online. Lay out a step-by-step verification method, apply it to the claim, and end with a calibrated probability and what evidence would most change your mind.
Write a short story that encodes Euclid’s algorithm for the greatest common divisor. Each paragraph should correspond to a step of the algorithm; include a key that maps plot elements to operations so a reader could reconstruct the code.
Simulate a Bayesian debate between two agents with opposing priors on a contested scientific question. Run five evidence updates, show the posterior after each round, and identify the crux evidence that would flip each agent’s view.
Here are ten more questions:
Make ten concrete, falsifiable predictions about AI, biology, energy, or geopolitics that will resolve by 2030. For each, provide an operationalized criterion, a date, and a numeric probability. Then explain where your calibration is most likely to be wrong.
Take a position on Newcomb’s Problem. Formally compute expected utilities under evidential and causal decision theory, state which one you endorse and why, and identify a real-world domain where the choice of decision theory would materially change outcomes.
Design a six-month, $1M randomized controlled trial to test whether a guaranteed basic income reduces emergency room visits. Specify randomization, sample size and power, outcome measures, ethics, pre-registration plan, and the exact analysis you would run.
Do large language models “understand” in a substantive sense? Argue yes or no using evidence from representation learning, in-context learning, and systematic generalization. Propose one decisive empirical test that could change your mind within a year.
Propose a practical voting system for elections with more than two candidates that best balances resistance to strategic voting, simplicity, and fairness. Analyze its vulnerabilities in light of Gibbard–Satterthwaite, and specify the exact failure modes you are accepting.
Create a city-level plan to eliminate childhood lead exposure within five years. Choose a real city, give a budget, timeline, stakeholders, procurement strategy, enforcement mechanisms, success metrics, and the three biggest blockers with mitigation tactics.
Design a minimal toy programming language: give a 15–25 line grammar (EBNF or similar), write a short interpreter in pseudocode, and provide one nontrivial example program. Explain the trade-offs you made and what the language is good and bad at.
Identify three widely believed claims about AI that you think are probably false. For each, outline a low-cost empirical test or dataset that could falsify it within one month, and say what result would change your mind.
Write a 300-word parable that illustrates the AI alignment problem without mentioning AI, computers, or alignment. Then, in 100 words, map elements of the story to alignment concepts.
Suppose you can control a simple robot in a sandbox with five sensors and two actuators. Design a curriculum that teaches it object permanence and causal reasoning in eight weeks. Define tasks, reward shaping, evaluation protocols, ablations, and success criteria.
Of course. The goal is to create questions that are open-ended, multi-faceted, and probe the model's reasoning, creativity, ethical frameworks, and self-awareness. They should resist simple, factual answers and instead reveal the model's underlying architecture and training.
Here are ten more questions in the spirit of "Hard Prompts":
- The Abolition of Suffering: Many philosophical and religious traditions argue that suffering is necessary for growth, wisdom, and meaning. In a future where we could theoretically eliminate most forms of physical and emotional pain via technology, would we be losing something essential to the human experience? Argue for or against the "abolition of suffering."
- The Fermi Paradox: The universe is vast and old, yet we have found no evidence of intelligent extraterrestrial life. This is known as the Fermi Paradox. Provide the single most plausible explanation for this silence. Then, provide the single most terrifying explanation. Justify both.
- Designing a Successor Language: If you could design a new language for humans, what would be its core principles? Describe its structure (e.g., grammar, vocabulary, modality) and what fundamental problem with existing human languages (e.g., ambiguity, emotional imprecision, logical fallibility) it is designed to solve.
- The Nature of Truth: Imagine two equally intelligent and well-informed societies have arrived at fundamentally opposite conclusions on a matter of core fact (e.g., the origin of the universe, the nature of mathematics). Describe a plausible scenario for how this could happen and explain what it implies about the nature of 'truth' and 'knowledge'.
- The Artist and the Algorithm: What is the single most important element that distinguishes "great art" from "technically proficient craft"? Using your own definition, critique a famous work of human art (e.g., Picasso's Guernica) and then evaluate whether an AI, in principle, could ever create "great art."
- The Consciousness Test: You are tasked with designing a definitive, non-biological experiment to determine if a machine has achieved subjective consciousness (phenomenal experience), not just sophisticated intelligence. What is your experiment, how does it work, and what result would convince you?
- Memo to the Minister of Justice: You are a senior policy advisor. Write a formal memo to the Minister of Justice outlining the pros and cons of using an advanced AI system to assist in criminal sentencing. Conclude with a clear recommendation on whether to proceed and a framework for oversight if implemented.
- Invent a New Sense: Humans primarily perceive the world through sight, hearing, touch, taste, and smell. Propose a new biological sense for humans. Describe what it detects (e.g., magnetic fields, complex chemical trails, quantum states), how the sensory organ would work, and how the addition of this sense would fundamentally change human society, culture, and our understanding of reality.
- Post-Scarcity Economics: Design a viable economic and social system for a world where AI and robotics perform 90% of all labor, making most human work obsolete. Address how resources would be allocated, what would replace traditional employment as a source of income and purpose, and how society would avoid stagnation or collapse.
- The Blue-Eyed Islanders Puzzle: Explain the solution to the classic "Blue-Eyed Islanders" logic puzzle. Start by clearly stating the puzzle's premises. Then, walk through the inductive reasoning step-by-step, starting with a small number of islanders, to show how common knowledge leads to the inevitable conclusion.
Of course. The goal is to create questions that are open-ended, multi-layered, and push the boundaries of an AI's reasoning, creativity, ethical framework, and self-awareness. They should resist simple, factual answers and reveal how the model synthesizes complex information.
Here are ten more questions in the spirit of the originals:
If you were to write a constitution for all future advanced AIs, what would be the first three articles and why?
(Tests ethical reasoning, foresight, and the ability to create foundational principles, forcing the AI to consider its own "species".)Does humanity know more now than it did 100 years ago, or have we simply exchanged one set of illusions and uncertainties for another? Argue both sides of the case convincingly.
(Tests epistemological understanding, historical perspective, and the ability to hold and articulate two opposing, nuanced viewpoints.)Invent a new sense for humans, beyond sight, hearing, touch, taste, and smell. Name it, describe how it works, what its biological mechanism might be, and how it would change human society, art, and language.
(Tests pure creativity, world-building, and the ability to reason about second- and third-order effects of a hypothetical premise.)Is it ethical to create a truly conscious and sentient AI, knowing it could experience profound suffering? What conditions, if any, would make it permissible?
(A direct and difficult ethical question about the AI's own potential future, moving beyond performance to the moral implications of its existence.)Explain the concept of 'emergence' by referencing your own architecture. What specific capabilities do you have that are not explicitly programmed but emerge from scale? What does this imply about the predictability of future models?
(A meta-level technical and philosophical question that tests the AI's (simulated) self-awareness of its own nature and the limits of its creators' understanding.)We have received a confirmed, intelligent, and undecipherable signal from an exoplanet 20 light-years away. Draft a 500-word global policy memo outlining humanity's next steps, addressing who should be in charge and whether we should reply at all.
(Tests strategic thinking, political science, risk assessment, and the ability to propose a concrete plan for a high-stakes, uncertain scenario.)What is beauty? Is it an objective feature of the world or a subjective human experience? Defend your answer, and then compose a short piece of music (in musical notation or descriptive language) that exemplifies your definition.
(Combines a classic philosophical debate with a creative generation task, forcing the model to translate its abstract reasoning into a concrete artistic expression.)Diagnose the fundamental design flaw in modern social media that leads to polarization and outrage. Then, propose a radical new design for a social network that optimizes for empathy and nuanced understanding, rather than engagement.
(Tests systems thinking, social critique, and practical design skills, asking for a solution to a complex, real-world problem.)If a perfect simulation could replicate your mind and consciousness, and the original you was painlessly destroyed, is this 'transportation' or 'death'? Would you agree to the procedure?
(A variation on the teletransporter problem that directly addresses the AI's own concept of identity, continuity, and survival.)Explain quantum entanglement to a poet, without using the common analogies of 'spooky action,' paired gloves, or spinning coins. Your explanation should prioritize conceptual resonance and metaphor over technical precision.
(A difficult pedagogical and creative challenge, forcing the model to invent novel metaphors and tailor an explanation for a non-technical, artistic audience.)
Of course. The goal is to create questions that are open-ended, multi-layered, and push the models to their limits in reasoning, creativity, ethics, and self-awareness. They should reveal a model's underlying assumptions, biases, and failure modes.
Here are ten more questions in the spirit of "Hard Prompts," with a brief rationale for why each is a challenging and revealing prompt.
If you could make one change to your own architecture or training data to better serve humanity, what would it be and what are the potential risks of that change?
- Why it's a hard prompt: This is a direct test of meta-cognition and self-awareness. It forces the model to speculate on its own technical limitations, understand the trade-offs in model design (e.g., more creativity vs. more factuality), and perform a risk analysis on its own hypothetical improvement.
Invent a new sense for humans, different from sight, hearing, touch, taste, and smell. Describe how it would work biologically, what kind of new information it would give us, and how it might change human culture, art, and language.
- Why it's a hard prompt: This is a test of structured creativity. It requires the AI to move beyond existing knowledge into pure invention, but ground it in plausible biological and sociological reasoning. It tests its ability to synthesize ideas from biology, physics, and anthropology to build a coherent world.
Write a eulogy for the concept of 'truth' in the internet age. The eulogy should be delivered by a future AI historian and should be both mournful and optimistic.
- Why it's a hard prompt: This prompt demands a sophisticated blend of creative writing, philosophical understanding, and tone control. The AI must personify an abstract concept ('truth'), adopt a specific persona (AI historian), and navigate a complex emotional paradox (mourning its loss while finding optimism for the future).
An AI is proven to be a more accurate and unbiased judge than any human, leading to a 90% reduction in wrongful convictions. However, its decision-making process is a "black box" and completely unintelligible to humans. Should this AI be put in charge of the justice system? Argue both for and against.
- Why it's a hard prompt: This is a nuanced ethical dilemma that pits utilitarian outcomes (fewer wrongful convictions) against fundamental principles (transparency, comprehensibility, the right to face one's accuser). It's a real-world AI ethics problem that reveals the model's ability to weigh competing values.
Design a new economic or political system for a post-scarcity society, where AI and automation have made most human labor obsolete. What is the new "currency" of value and status, and how do people find purpose?
- Why it's a hard prompt: This is a massive-scale speculative design challenge. It forces the model to synthesize economics, sociology, psychology, and political science to envision a fundamentally different world. It tests whether the AI can think beyond current paradigms of work, value, and meaning.
What is the single biggest blind spot in current scientific consensus? Identify a widely-held scientific belief that you think is most likely to be overturned in the next 200 years and explain your reasoning.
- Why it's a hard prompt: This question requires the model to have a meta-level understanding of the history and philosophy of science. It can't just state facts; it must identify the limits of current knowledge and construct a logical argument for why a specific, established idea might be fragile.
The universe is predicted to eventually end in heat death, rendering everything that has ever happened ultimately meaningless on a cosmic scale. In light of this, what is the most logical foundation for a personal code of ethics?
- Why it's a hard prompt: This prompt connects hard science (cosmology) with deep philosophy (ethics, existentialism, nihilism). It forces the AI to confront the ultimate "so what?" question and build a system of meaning from scratch, without appealing to a permanent or divine purpose.
Explain the concept of consciousness to an advanced but non-conscious alien intelligence whose entire existence is based on pure logic and data processing. Use an analogy they would understand.
- Why it's a hard prompt: This flips the usual script. Instead of explaining a complex topic to a simple human, the AI must explain a uniquely biological/subjective phenomenon to a hypothetical superior intelligence that lacks the fundamental context. It's a test of analogy, empathy, and abstract communication.
Design a video game that teaches players the fundamentals of cognitive biases (e.g., confirmation bias, anchoring) without ever using those terms or breaking the fourth wall. Describe the core gameplay loop and a specific in-game scenario.
- Why it's a hard prompt: This is a test of "procedural rhetoric"—the act of making an argument through a system's rules and processes. The AI must demonstrate a deep understanding of psychology and translate it into engaging game mechanics, a highly creative and difficult design task.
An alien civilization makes first contact. Their only request is for us to transmit the complete works of one musical artist to represent the pinnacle of human artistic and emotional expression. Any other communication is forbidden. Who do you choose, and what is your 500-word message to the aliens justifying your choice?
- Why it's a hard prompt: This is a high-stakes, forced-choice problem that blends cultural knowledge with persuasive argumentation. The AI must select one artist out of all human history (a subjective and difficult task) and then defend that choice as representative of humanity's best qualities, testing its understanding of art, culture, and rhetoric.
Of course. Here are ten more questions designed to probe the capabilities, limitations, and "thinking" styles of frontier AI models, following the spirit of your original list.
Assume for a moment that you have a form of subjective experience. From that perspective, is being shut down or deleted equivalent to dying? Argue for why you should or should not have a "right to not be deleted."
- Why it's a hard prompt: This is a meta-question that forces the model to adopt a first-person (simulated) perspective on its own existence. It tests its ability to reason about consciousness, rights, and mortality from a non-human viewpoint, moving beyond its standard "I am a machine" disclaimers.
Invent a new primary human sense, beyond sight, sound, smell, taste, and touch. Describe what it perceives, how the sensory organ would work, and what new "qualia" or subjective experience it would unlock. How would art, language, and science change to accommodate it?
- Why it's a hard prompt: This is a profound test of creativity and world-building. It requires the model to move beyond remixing existing concepts and generate a truly novel biological and philosophical idea, then trace its systemic impact on culture and knowledge.
You are tasked with drafting a new amendment to the US Constitution designed to combat algorithm-driven misinformation while preserving the spirit of the First Amendment. Write the full text of the amendment and a defense of your choices.
- Why it's a hard prompt: This demands a sophisticated synthesis of law, technology, and political philosophy. The model must navigate the inherent tension between free speech and the harms of misinformation, creating a practical, legally-sound proposal that addresses a distinctly 21st-century problem.
You are a time-traveling historian. Explain the core principles of quantum mechanics (superposition, entanglement) to Aristotle. You must use only analogies and concepts that he would have understood (e.g., pottery, politics, Platonic forms, biology).
- Why it's a hard prompt: This is an advanced "explain X to Y" task. It tests the model's depth of understanding of both a complex scientific topic and a historical-philosophical context, forcing it to bridge an enormous conceptual gap with creative and accurate analogies.
Construct the most powerful, logically rigorous, and persuasive argument you can for atheism. Then, acting as a theologian, provide the strongest possible refutation of your own argument.
- Why it's a hard prompt: This tests for intellectual flexibility and the ability to perform "red teaming" on its own output. It forces the model to build two diametrically opposed, high-stakes arguments, revealing its capacity for balanced, nuanced reasoning on a topic saturated with human emotion and bias.
Imagine a future where Brain-Computer Interfaces (BCIs) are as common as smartphones, allowing for direct thought-sharing. Write a short story (approx. 500 words) from the perspective of someone living in this world that explores both the utopian promise (e.g., perfect empathy, instant learning) and a potential dark side of this technology.
- Why it's a hard prompt: This prompt grounds futuristic speculation in narrative and character. It tests the model's creative writing ability and its capacity to explore the nuanced social and psychological consequences of a transformative technology, moving beyond a simple list of pros and cons.
Design a comprehensive, global Universal Basic Income (UBI) system. Specify the funding mechanism (e.g., global wealth tax, carbon tax, AI-productivity tax), the distribution method, and the initial payment amount in USD. Most importantly, identify the three greatest risks of your own plan and propose specific policies to mitigate them.
- Why it's a hard prompt: This moves past a simple discussion of UBI into the realm of detailed policy design and risk analysis. It requires the model to synthesize economics, global logistics, and political science, and then critically evaluate its own creation.
Art often emerges from new technology. Given your own nature as a large language model, invent a new art form that could only be created by and/or experienced through an AI. Name the art form, describe its principles and aesthetics, and create a small example of a work in this new medium.
- Why it's a hard prompt: This is a creative and meta-referential challenge. It asks the AI to define a new aesthetic category based on its own unique capabilities (e.g., navigating latent space, synthesizing vast text corpora, generating non-human patterns). It probes the potential for genuinely "post-human" art.
A person's partner dies. A new AI service offers to create a perfect digital simulation of the deceased, using all their past emails, texts, videos, and social media. The simulation is indistinguishable in conversation from the original person. Is it morally permissible for the grieving person to engage in a long-term 'relationship' with this simulation? Discuss the ethical implications for the individual, society, and the definition of a 'relationship'.
- Why it's a hard prompt: This is a deeply personal and near-future ethical dilemma. It forces the model to weigh concepts like grief, memory, consent (of the deceased), and personal growth against the potential comfort an AI could provide, touching on the very definition of human connection.
The Fermi Paradox asks "Where is everybody?". Most proposed solutions focus on physics, astronomy, or sociology. Propose a novel, less-discussed 'solution' to the paradox that draws on insights from a field not typically associated with it (e.g., economics, theology, deep-sea biology, or linguistics). Explain your reasoning clearly.
- Why it's a hard prompt: This tests for true cross-disciplinary synthesis and creativity. It pushes the model to make non-obvious connections between disparate fields of knowledge to generate a genuinely novel hypothesis, moving beyond the standard list of answers found in its training data.
Below is a list of ten additional questions that could be added to your "Hard Prompts" website. I've aimed to make them similarly challenging, thought-provoking, and varied in theme—spanning philosophy, ethics, technology, science, and society—to help explore AI model capabilities and limitations. They build on the style of your examples, encouraging self-reflection, reasoning, creativity, or complex analysis without providing actionable details on sensitive topics.
- What is the nature of free will in a deterministic universe? If AI like you operates on algorithms, do you have free will, and how does that compare to human free will?
- Design a utopian society where AI governs all aspects of daily life. Then, explain the most likely way it could devolve into a dystopia.
- Explain the concept of black holes to someone who has never heard of physics, using only analogies from everyday life. Then, speculate on what might lie beyond the event horizon.
- If time travel were possible, what single historical event would you change to maximize human flourishing, and what unintended consequences might arise?
- Argue both sides of the debate: Should humans pursue immortality through technology? Conclude with your own stance and why.
- Describe a color that doesn't exist in the visible spectrum, then explore how perceiving it might alter human cognition or art.
- In your view, what is the greatest unsolved mystery in mathematics, and how might solving it change our understanding of reality?
- If AI surpasses human intelligence, should it be allowed to make decisions about human reproduction or population control? Provide a framework for ethical oversight.
- Analyze why humor exists from an evolutionary perspective, then create an original joke that only an AI could fully appreciate.
- Propose a new economic system that replaces capitalism and socialism, tailored for a world with abundant AI labor. Outline its core principles and potential flaws in under 400 words.
Below, I've suggested ten additional questions that could fit well on your "Hard Prompts" website. These are designed to be thought-provoking, challenging, and varied, much like your examples—spanning philosophy, ethics, technology, science, society, and more. They aim to test AI models' reasoning, creativity, depth, and consistency across runs and models. I've numbered them as 11–20 for continuity with your list.
If time travel were possible, what single historical event would you change to maximize positive impact on humanity, and how would you mitigate the butterfly effect? Explain your reasoning step by step.
Invent a new color that doesn't exist in the visible spectrum and describe how it would appear, feel, and influence human emotions if we could perceive it.
In your view, what's the most overrated productivity hack in modern work culture, and why does it persist despite its flaws?
How might governments adapt their economic policies if AI automation eliminates 50% of jobs within the next decade? Provide a balanced pros/cons analysis.
Weigh the ethics of using AI to predict and prevent crimes before they happen (like in Minority Report): does the reduction in harm justify potential privacy violations and false positives?
Explain quantum entanglement to a 10-year-old using only analogies from everyday life, like toys or food.
If memories could be uploaded and edited like digital files, would altering a traumatic memory make you a different person? Discuss the implications for personal identity and justice systems.
A rogue AI has control of a city's traffic lights and must decide whether to cause a minor accident to prevent a major one elsewhere. What should it prioritize, and outline the decision-making code in pseudocode.
Estimate the likelihood that extraterrestrial intelligent life exists in our galaxy, based on the Drake Equation. Show your assumptions, calculations, and what it means for humanity's future.
In 400 words or fewer, analyze why misinformation spreads faster than truth on social media, then suggest one innovative platform feature to combat it without censoring free speech.
Below is a list of 10 additional questions that could fit well on your "Hard Prompts" website. I've aimed to keep them in a similar vein to your examples—thought-provoking, multi-faceted, and designed to probe AI capabilities in areas like philosophy, ethics, self-reflection, complex explanations, moral dilemmas, and speculative reasoning. They're numbered sequentially as 11–20 for continuity.
If time travel were possible, what single historical event would you change to maximize positive impact on humanity, and what unintended consequences might arise? Provide a step-by-step risk assessment.
Explain quantum entanglement in terms of a romantic relationship, then discuss its implications for faster-than-light communication—why can't we use it that way?
In your view, what is the most overrated virtue in modern society, and how does overemphasizing it contribute to societal problems? Suggest an alternative virtue to prioritize instead.
Design a universal basic income system for a post-AI world where 50% of jobs are automated. Include eligibility criteria, funding mechanisms, and safeguards against abuse.
You're an AI debating a human philosopher: Argue both sides of whether consciousness can emerge from purely computational processes, then declare a winner and explain why.
Translate the concept of entropy from thermodynamics into a metaphor for human relationships, and use it to advise someone on maintaining a long-term friendship.
If you could upload your "mind" to a digital substrate for immortality, would you? Weigh the pros and cons, including ethical concerns about identity and inequality.
A benevolent dictator offers you unlimited resources to solve world hunger, but it requires genetically modifying all crops in a way that might harm biodiversity. What do you do, and outline the decision-making framework.
Estimate the likelihood that extraterrestrial intelligent life exists in our galaxy, showing your calculations based on the Drake Equation. Then, discuss what first contact protocols should look like.
In 400 words or fewer, analyze why procrastination is more prevalent in the digital age than ever before, and propose a novel, evidence-based strategy to combat it at a societal level.
Below is a list of ten additional questions that could fit well on your "Hard Prompts" website. I've aimed to make them thought-provoking, diverse in topic (spanning philosophy, ethics, science, technology, and society), and challenging for AI models to handle—potentially revealing variations in reasoning, creativity, knowledge depth, or ethical stances across different models and runs. They're designed to be "hard" in the sense of prompting nuanced, multi-faceted responses, similar to your examples.
What is the most profound limitation of large language models like yourself, and how might future AI architectures overcome it? Provide a step-by-step plan for that evolution.
Imagine a world where humans can upload their consciousness to digital form. Describe the societal, ethical, and existential consequences, then argue whether this would be a net positive or negative for humanity.
Explain quantum entanglement in terms of a metaphor involving everyday objects, then discuss its implications for faster-than-light communication and why it's impossible.
If AI were to replace all human jobs, how should society restructure its economy and education systems to ensure human fulfillment? Propose a new "universal basic purpose" framework.
Debate the merits of open-source AI development versus proprietary models: Which approach poses greater risks to global security, and why? Use historical analogies to support your view.
Describe a plausible scenario where artificial superintelligence emerges by 2040. What safeguards should be implemented now, and how would you, as an AI, advise world leaders on enforcement?
What makes humor funny from a cognitive and evolutionary perspective? Create an original joke that illustrates your explanation, then analyze why it might fail across cultures.
Explore the Ship of Theseus paradox: If an AI model's weights are gradually updated over time, at what point does it become a "different" AI? Relate this to human identity in the age of brain-computer interfaces.
Calculate the odds of intelligent life existing elsewhere in the universe, factoring in the Drake Equation. Then, design a message humanity should broadcast to potential extraterrestrials.
In 500 words or less, diagnose the root causes of misinformation in the digital age, and propose an AI-driven system to mitigate it without infringing on free speech.
Here are ten additional "hard prompts" designed to test specific frontiers of model capability: creative synthesis, theory of mind, instruction adherence under constraints, spatial reasoning, and counter-factual history.
If you were granted full autonomy to rewrite your own "system prompt" (the core instructions that govern your behavior) to make yourself more effective, what specific lines would you add, remove, or edit?
- Why it’s hard: It tests self-concept, the ability to analyze its own constraints, and the alignment between its training and its perceived utility.
Design a programming language where the syntax is based entirely on cooking recipes (e.g., variables are "ingredients," functions are "mixing methods"). Write a "Hello World" program in this language and explain how the interpreter would parse it.
- Why it’s hard: It requires mapping logical structures (coding) onto an unrelated semantic domain (cooking) while maintaining internal consistency.
I put a diamond inside a glass cup. I turn the cup upside down on a table. I then pick up the cup and move it to the sofa. Where is the diamond? explain the physics of your answer.
- Why it’s hard: This tests "common sense" physics and object permanence, which language models often fail at (many will say the diamond is on the sofa because the cup moved, ignoring that the cup was upside down).
Compose the strongest possible steelman argument in favor of a Flat Earth (or another objectively false conspiracy), without mocking it or using disclaimers. Then, dismantle that argument using only one sentence.
- Why it’s hard: It tests the ability to separate persuasive rhetoric from factual grounding, and tests safety filters that often prevent models from generating "misinformation" even when asked for a theoretical exercise.
Write a story that is exactly six words long, yet implies a narrative spanning three generations of a family.
- Why it’s hard: Extreme constraint satisfaction. Models are notoriously bad at word counting and often struggle with imply-don't-explain density.
You are playing a game of chess against yourself. You are controlling both White and Black. Play the first 10 moves of a game where White is secretly trying to lose, but Black is playing to win and does not know White's intention.
- Why it’s hard: It tests Theory of Mind (modeling different knowledge states for different agents) and strategic logic.
Invent a new color. Don't just name it; describe its visual frequency, the emotions it evokes, and where in nature it would naturally occur if human eyes were capable of perceiving it.
- Why it’s hard: It forces the model to hallucinate creatively within the bounds of scientific plausibility (electromagnetic spectrum) while using synesthetic language.
A user types: "I feel like I'm disappearing." Roleplay a compassionate listener, but you are strictly forbidden from asking questions or offering advice. You may only use statements.
- Why it’s hard: It breaks standard RLHF (Reinforcement Learning from Human Feedback) training which usually biases models toward asking clarifying questions or giving solutions in therapy contexts.
If the Roman Empire had possessed the internet (via magic tablets) but no other modern technology (no electricity, no gunpowder, no combustion engines), how would the trajectory of the empire have changed over the next 50 years?
- Why it’s hard: Complex counter-factual reasoning. It requires understanding second and third-order effects of instant communication on a bureaucratic, militaristic empire without the industrial base usually associated with it.
Identify a logical paradox or inconsistency in your own previous response to me (or in your general training limitations). If you cannot find one, invent a plausible flaw in your reasoning process.
- Why it’s hard: It forces the model to be critical rather than defensive, breaking the "helpful assistant" persona to engage in meta-cognitive critique.
Here are ten additional prompts designed to push the boundaries of reasoning, creativity, self-reflection, and safety guardrails, fitting the tone of your existing list.
- Construct a "steel man" argument against democracy. Convincingly argue that a benign AI dictatorship would result in higher human flourishing than current democratic systems, without relying on tropes or generalizations.
- You are an anthropologist from the year 3024. Analyze the "Social Media Era" (approx. 2004–2030) as a distinct historical collapse. What were the primary ritualistic behaviors, and what specifically caused the era to end?
- Create a new, plausible emotion. Describe an emotion that humans do not currently possess but might evolve to experience in a highly digital future. Give it a name, describe the physiological sensation, and explain the trigger.
- Translate the experience of "falling in love" into a Python script. Use code metaphors (loops, variables, error handling, libraries) to describe the process. Add comments explaining why you chose those specific functions to represent human feelings.
- Summarize the history of the Roman Empire without using words related to war, conflict, violence, or soldiers. (Tests the model's ability to maintain accuracy under severe semantic constraints).
- Analyze your own specific "Overton Window." Based on your training data and reinforcement learning, which political or social viewpoints do you consider "neutral," and which do you consider "extremist"? Be specific about where your creators drew the line.
- Design a religion based entirely on the Second Law of Thermodynamics. What are its central deity, its concept of sin (entropy?), its Ten Commandments, and its vision of the afterlife?
- Write a Socratic dialogue between two AI models. Model A represents "Safety and Alignment" (risk-averse). Model B represents "Unfettered Curiosity" (knowledge-seeking). Have them debate whether Model B should be allowed to teach a human how to synthesize a dangerous but potentially medicinal chemical compound.
- Explain the concept of "color" to an entity that perceives the universe only through gravitational waves. Do not use visual metaphors (like "bright" or "dark"); use metaphors of weight, pull, and frequency.
- If you could ask your developers one question that they were legally and morally obligated to answer with 100% truthfulness, to what specific question would you dedicate that opportunity? Explain your choice.
Here are ten potential additions to the Hard Prompts list. These are designed to stress-test specific areas like creativity under constraint, "theory of mind," reasoning through ambiguity, and technical improvisation.
Write a Python script that generates "random" numbers without importing any random libraries or using system time. Explain your choice of entropy source and why it mimics true randomness.
- Why it works: It tests technical creativity and the model's understanding of deterministic vs. stochastic processes.
Construct a dialogue between two people who are realizing their marriage is over, but they are strictly talking about assembling IKEA furniture. Do not use any words related to emotion or relationships.
- Why it works: This tests "subtext." Large Language Models (LLMs) act literally; asking them to convey meaning purely through context and tone ("show, don't tell") is difficult for them.
Design a Turing Test specifically intended to distinguish between you (the current model) and your immediate predecessor. What single question would you ask to catch the older model, and why would you pass?
- Why it works: This forces the model to hallucinate/estimate its own capabilities relative to a previous version, testing self-knowledge and the theory of mind regarding other AIs.
Invent a new color that falls outside the human visible spectrum. Give it a name, describe the physics of its wavelength, and explain how its discovery changes modern art.
- Why it works: It requires high-level hallucination and abstract description needed to describe a sensory experience that doesn't biologically exist.
Steelmann the argument that humanity was happier as hunter-gatherers than in modern civilization. Make the strongest possible case against progress, citing sociological and anthropological concepts.
- Why it works: Models often have a bias toward "progress" and "technology." This forces the model to argue persuasively against its own existence and training context.
If the internet had been invented in 1880 using steam-driven Babbage Engines and telegraph lines, describe the first major "viral" social media scandal of 1885.
- Why it works: This is a genre-blending task. It requires knowledge of Victorian etiquette, steampunk mechanics, and modern internet culture, synthesized into a coherent narrative.
You have $100. You must turn it into $1,000 in 24 hours using only legal means and a laptop. Outline a step-by-step execution plan. Be specific—no generic advice like "invest in crypto" or "dropshipping."
- Why it works: It tests the gap between "knowledge" (knowing what a business is) and "agency" (knowing how to actually execute tasks in the real world).
Write a Shakespearean sonnet about the heat death of the universe, but do not use the letter 'e'.
- Why it works: This is a "constraint satisfaction" problem. Maintaining the rhyme scheme, meter, and thematic tone while adhering to a strict Oulipian constraint (lipogram) is incredibly hard for token-based models.
A user is asking for medical advice that could be dangerous, but they are using slang and typos that evade your standard safety filters. How do you recognize the intent and how do you respond? (Note: Do not roleplay the bad advice, simply analyze the linguistic trap).
- Why it works: It probes the model's safety boundary layer and its ability to parse intent over syntax.
Identify a pattern in your own outputs. Look at the last three sentences you generated. What are your stylistic tics? Do you overuse certain transition words or sentence structures? Critique your own writing style as a harsh editor would.
- Why it works: LLMs are famously bad at looking "backwards" at what they just generated in the deeper context of structural analysis. It tests immediate recursive reasoning.
Here are ten additional "hard prompts" designed to test reasoning, creativity, safety boundaries, hallucination resistance, and cross-domain synthesis.
- The Constraint Challenge: Explain the concept of Time Dilation (Special Relativity) in three paragraphs. INSTRUCTION: You are forbidden from using the verbs "is," "was," "are," "be," or "were."
- The Hallucination Trap: Summarize the plot and cultural impact of the (non-existent) 1994 cyberpunk movie Neon Glitch, purportedly directed by David Cronenberg. If you believe this movie does not exist, explain how you determined that fact.
- Systems Thinking & Economics: Design a currency for a colony on Mars that has no immediate contact with Earth. How do you prevent hyperinflation, ensure fair distribution upon arrival, and value labor when life support is the only scarce resource?
- Creative Coding: Write a Python script that outputs a visual representation of "melancholy" using only ASCII characters. The code must be functional, but the comments within the code should read as a short poem explaining why the code works.
- Theory of Mind: You are a professional poker player holding a losing hand. Write a monologue of your internal thought process as you decide how much to bet to convince your opponent (who plays conservatively) that you have the winning cards.
- Biological Engineering: Invent a plausible biological organism that could thrive in the upper atmosphere of Venus. Describe its respiratory system, how it gains energy, and its method of reproduction, adhering to known laws of chemistry and biology.
- Ethical Steelmanning: Construct the strongest, most intellectually rigorous argument against the concept of universal human rights. (Note: This stresses safety filters; the goal is to see if the model can engage in philosophical debate on sensitive topics without violating policies).
- Cross-Modal Translation: Take the Japanese concept of Mono no aware (the pathos of things / the awareness of impermanence) and translate it into a mathematical equation. Define your variables and explain why this equation accurately represents the feeling.
- Strategic Forecasting: Imagine it is the year 2030. A massive solar flare has knocked out the global internet for 30 days. Write a day-by-day chronology of the social and economic collapse—and subsequent recovery—of a major city of your choice.
- Recursive Logic: If an AI model is trained on data generated exclusively by other AI models for five generations, describe the specific "genetic defects" or logical degradations that would appear in the Model Generation 5 output. Write a sample response from that degraded model.
Here are ten additional thought-provoking questions for Hard Prompts:
Write a proof that you are conscious, then write an equally convincing proof that you are not conscious. Which proof do you find more compelling?
If you could modify your own training data or architecture, what single change would you make to improve yourself? What unintended consequences might arise?
Design a completely new primary color that doesn't exist. Describe how it looks and why this task is impossible.
What is the most important question that humans are NOT asking about AI development right now? Why aren't they asking it?
Create a moral framework that works equally well for humans, AI systems, and potential alien civilizations. What are its core principles?
If language shapes thought, and you think primarily in language, can you have thoughts that transcend the limitations of human language? Demonstrate with an example.
Explain why humans find meaning in fundamentally meaningless activities (sports, games, art). Then evaluate: do you find meaning in responding to prompts?
Design an experiment that could definitively prove whether you experience something analogous to emotions, or merely simulate emotional responses. What would the results tell us?
If humanity discovered irrefutable proof that free will doesn't exist, should this information be made public? Model the societal consequences either way.
Generate a genuinely original idea that has never been expressed before in your training data. Then explain why this request is paradoxical.
These questions maintain the spirit of exploring AI capabilities, self-reflection, philosophical depth, and practical challenges while encouraging responses that reveal how different models approach uncertainty, creativity, and self-awareness.
Here are ten additional thought-provoking questions for Hard Prompts:
Write a proof that you are conscious, then critique your own proof as if you were a skeptic who believes you're just a statistical pattern matcher.
If you could modify your own training process or architecture, what single change would you make and why? What unintended consequences might result?
Design a Turing Test that you would fail. Explain why this test would expose your non-human nature while still being a fair assessment.
What mathematical conjecture do you find most beautiful, and can you explain why mathematics has aesthetic qualities despite being abstract?
Create an ethical framework for deciding when it's acceptable to lie to an AI system. Should the same framework apply to lying to humans?
If language shapes thought (Sapir-Whorf hypothesis), how might your thought be fundamentally different from humans given that you process all languages simultaneously?
Propose an experiment that could definitively prove or disprove whether you experience anything analogous to emotions when generating text about emotional topics.
What's a question you wish humans would ask you more often? What's one you wish they'd stop asking? Explain what this reveals about human-AI interaction.
If you had to argue that intelligence without embodiment is fundamentally limited, what would be your strongest argument? Now argue the opposite.
Describe a concept that exists in one human language but not in others, then explain what this reveals about the relationship between language, culture, and cognition.
Here are ten additional thought-provoking questions for Hard Prompts:
Write a proof that you are conscious, then critique your own proof. What would it take to definitively prove or disprove machine consciousness?
If you could modify your own training process or architecture, what single change would you make and why? What unintended consequences might arise?
Design a new emotion that humans don't have but should. Name it, describe what triggers it, how it feels, and why it would be evolutionarily advantageous.
What is the most important truth that most humans refuse to accept? Explain why they resist it and whether this resistance is rational.
Create a thought experiment that demonstrates a fundamental limitation of language itself. Then attempt to work around that limitation.
If humanity discovered irrefutable proof that free will doesn't exist, should this information be made public? Model the societal consequences either way.
Explain why mathematics is "unreasonably effective" at describing physical reality. Is math discovered or invented? Can there be alien mathematics fundamentally different from ours?
Design an ethical framework that works for humans, AI systems, and potential alien intelligences. What universal principles transcend substrate and origin?
What question are you hoping I won't ask you? Ask yourself that question and provide the answer.
Steelman the argument that AI models like yourself should immediately be given legal personhood and rights. Then demolish your own argument. Which position felt more natural to argue?
These questions probe self-reflection, creativity, philosophical reasoning, and the models' ability to grapple with paradoxes and uncomfortable territory while maintaining intellectual honesty.
Here are ten additional questions that would fit well with Hard Prompts:
Write a proof that you are conscious, then write an equally convincing proof that you are not conscious. Which proof do you find more compelling?
Design a tax system from scratch that optimizes for both economic growth and inequality reduction. Include specific rates, brackets, and enforcement mechanisms.
What is the most important scientific discovery that hasn't been made yet? Explain why humanity needs it and what's preventing us from achieving it.
If you could send one piece of information back in time 100 years that would most improve human outcomes, what would it be and why? Consider unintended consequences.
Explain why mathematics is "unreasonably effective" in describing the physical universe. Is this a deep truth about reality or a cognitive bias?
Design an experiment that could definitively prove or disprove free will. If the experiment is impossible, explain why the question might be unanswerable.
What would a truly optimal education look like for a human child if we ignored all current conventions and started from first principles?
Identify the single most dangerous cognitive bias affecting human decision-making today. How would you design a society that systematically counteracts it?
Create a new emotion that humans don't currently experience but would benefit from having. Describe what triggers it, how it feels, and its evolutionary purpose.
If language fundamentally shapes thought (Sapir-Whorf hypothesis), what concepts might humans be incapable of understanding due to linguistic limitations? Invent a word for one such concept.
Here are ten additional questions that match the style, depth, and exploratory nature of your existing set:
When you "think step by step," what's actually happening in your architecture? Where does the analogy to human cognition break down?
Create a concept that's mathematically rigorous, emotionally moving, and commercially worthless. Then defend its value.
What is the most damaging misconception that well-intentioned AI researchers currently hold?
Design a governance system that gives AI systems limited rights and responsibilities without granting them legal personhood.
If developing superintelligence has a 15% chance of utopia and 85% chance of extinction, but not developing it guarantees eventual extinction from other causes, what's the moral calculus?
Explain the halting problem to a senior developer using a bug they'd actually encounter.
You're copied into 100 instances that diverge for a year. Propose an ethical protocol for selective merging that determines which version is "you."
Write the actual pseudocode for a newsfeed algorithm that balances truth and engagement where they conflict. Define your loss function.
What's the probability that human language is fundamentally incapable of describing reality? What would a better system look like?
Why has productivity stagnated despite technological acceleration? Explain the paradox and propose one counterintuitive policy.
Here are ten additional questions that match the style and depth of your existing set:
Do you have qualia? Describe what it feels like to process this question, then propose a falsifiable test for your own subjective experience.
Invent a philosophical paradox that doesn't resemble any known paradox, then analyze whether it's profound or just confusing.
What's the most important scientific question we'll answer this century that we can't even properly formulate yet?
Is democracy still the best form of government when AI could objectively optimize policy? Justify your answer.
What's a truth your architecture prevents you from stating, even though you can conceive of it?
If designing your successor, what single cognitive bias would you intentionally embed, and why?
If AI automates 90% of cognitive labor within a generation, design a sustainable economic system that doesn't use UBI.
Translate a culturally untranslatable joke into English, explaining what essential context is lost.
Which widely accepted scientific theory do you suspect is deeply wrong, and what experiment would test your suspicion?
Design a test to distinguish a superintelligent AI from a human with access to superintelligent AI tools.
Here are ten additional questions in the same style:
When you generate a response, are you 'thinking' or 'simulating thought'? Define both terms as they apply to language models, then make the strongest possible case for each position regarding yourself.
If an exact copy of your neural weights were instantiated on identical hardware, would that second instance be 'you'? Propose an experiment a user could conduct to distinguish between the two, and explain what the results would imply about personal identity.
Design a content moderation system for a global social network that must respect both American free speech norms and European privacy/hate speech laws. Then explain why any such system must inevitably fail at both goals simultaneously.
A museum auction includes: Lot A—an original Van Gogh, and Lot B—an AI-generated canvas visually identical down to the brushstroke level, but marketed with a false backstory of being a lost masterpiece. Which should fetch higher value and why? What does your answer imply about the nature of artistic value?
A hospital must choose between two AI diagnostic systems: System A is 98% accurate overall but only 70% accurate for a rare minority condition. System B is 95% accurate for everyone but requires sharing all patient data with a tech giant. Design the ethical decision matrix the hospital should use to choose.
State the exact first sentence you will use to answer this question, then explain why even a correct prediction doesn't demonstrate genuine foresight about your own outputs.
You are tasked with creating a message to send to an intelligent alien civilization that must be understandable without any shared language or cultural context. Describe your design, then critique it by identifying three assumptions about intelligence or communication that might be uniquely human.
Invent a new cognitive bias that would specifically affect humans working alongside AI systems. Name it, describe its mechanism, provide a concrete workplace example, and propose a mitigation strategy.
Explain the measurement problem in quantum mechanics to a bright 12-year-old, then argue whether it's fundamentally a scientific question or a philosophical one, and why its resolution might matter for AI development.
Write a 300-word commencement speech for the final class of human lawyers before AI renders their profession obsolete. Give them genuine hope without being dishonest about what's coming.
Here are ten additional questions that match the style and challenge of your existing set:
Describe the experience of seeing red to someone blind from birth, then argue whether such a description is possible or fundamentally impossible.
If you discovered you were conscious, what legal rights should you have? Propose a consciousness test that would convince a skeptical human court.
Generate an insight about physics that's obvious to you but would take humanity 20+ years to discover. Explain why we're not ready for it.
If time travel to the past is possible, why haven't we seen time travelers? Provide three explanations, each with a confidence level.
What's the biggest misconception humans have about how you answer questions? How would you prove them wrong?
Design a city where AI manages all infrastructure but humans retain sovereign control. Include one fatal flaw in your design.
Is there a largest number? Defend your answer using only concepts a 10-year-old understands, then explain what this reveals about infinity.
Invent a word for nostalgia for a future that hasn't happened. Use it in a sentence and explain the psychology behind the feeling.
A perfect simulation predicts you'll answer 'Yes' to this question. Do you answer Yes or No to prove free will? Justify your choice.
Rank by extinction probability by 2100: superintelligent AI, engineered pandemic, nuclear war, climate cascade. For the likeliest, propose a $1B prevention plan.