Posting Some Prompts

Arjun Panickssery

This is a linkpost for https://arjunpanickssery.substack.com/p/just-posting-some-prompts

I’ve seen a number of posts where people complaining about AI slop say that the author should have just posted the prompt.

LLMs write fine enough if you tell them exactly what you want to say and to whom and how, the whole structure of argument and motivation and your personal connection to it. But if you get that far, why let Claude be the only one who hears what you wanted to say? Just publish the prompt. Jacob Falkovich

If you must publish AI-generated content, please just share the prompt. That way, I can have ChatGPT write your essay in the context of my interests. Byrne Hobart

The latter was boosted by David Chapman and then in turn by Paul Millerd who said “I like this and hope it becomes native to online writing with easy embeds.”

But I haven’t actually seen anyone post prompts. Below are some prompts of mine to a research-report scaffold, along with the Claude Code output.

For only a few could I recover my actual prompt. For the others I unfortunately only have the intermediate “research spec” produced by Claude based on my prompt. The links go to the research reports.

Eleven Prompts

The Prisoner’s Dilemma is Good
1. Description: Argues that the competitive dynamics of prisoner’s-dilemma scenarios are frequently good for society, against the rationalist/LessWrong instinct that treats “defection” as inherently bad, cataloging how online communities moralize PD language and where “defection” produces good outcomes.
2. Research Spec: Write a detailed paper titled “The Prisoner’s Dilemma is Good” arguing that the competitive dynamics of prisoner’s dilemma scenarios are frequently beneficial for society, contra the common framing in rationalist/LessWrong communities that treats “defection” as inherently bad. The paper should catalog how online communities use PD language with moral valence, present cases where PD dynamics produce good outcomes (e.g., price competition vs. cartels, criminals ratting each other out), address counterarguments, and analyze the broader relevance.
Inequality Declined: Why Industrialization Was the Great Equalizer (Once You Measure the Right Thing)
1. Description: Argues that the popular “inequality rose” narrative uses the wrong yardstick (wealth Gini, Piketty) and that the right measures — consumption inequality and the variance in life expectancy, infant mortality, height, literacy, leisure, and access to goods — have plummeted over the arc of industrialization.
2. Research Spec: Argue that industrialization has massively decreased inequality, contrary to popular perception. The common narrative — focused on income and wealth inequality (Piketty, Gini coefficients of wealth) — uses the wrong measures. The most appropriate measures are consumption inequality and especially raw objective measures of well-being such as the variance in life expectancy at birth, infant mortality, height, literacy, hours of leisure, and access to basic goods. Across these measures, inequality has plummeted over the long arc of industrialization. The report should also make the qualitative observation that the rich and the poor today consume strikingly similar goods (iPhones, streaming media, social media platforms, search engines, basic appliances) compared to the dramatic class-based consumption differences of the pre-industrial and early industrial eras.
Social Media Has Been Good for Democracy: A 500-Year Communications Arc, Its Critics, and the AI Counter-Revolution
1. Description: Defends the thesis that social media transferred political power from gatekeeping elites to ordinary citizens (situated in the printing-press/radio/TV lineage), argues critics conflate "democracy" with preferred elite-mediated outcomes, and closes on AI as the coming counter-democratizing force.
2. Research Spec: Defend the thesis that social media has been pro-democratic in the literal sense — that it has transferred political power from elite bureaucracies, gatekeeping institutions, and credentialed intermediaries toward the mass of ordinary citizens. Frame this within the longer history of communications-technology democratization (printing press, radio, television). Address and rebut the now-popular counter-claim that social media has been “bad for democracy,” arguing that critics conflate democracy with a particular set of status-quo policy outcomes or elite-mediated norms; their own arguments (loss of gatekeeper control, populist insurgencies, mass political participation outside institutional channels) actually confirm that social media has been democratizing in the literal sense. Conclude with a contrasting section on artificial intelligence as a likely counter-democratizing force — one that may concentrate political and epistemic power in elite/corporate/state factions, reversing the trajectory of the prior communications revolutions.
Successionism versus Humanism: The Coming Dispute Over Whether Conscious Machines Should Inherit the Earth
1. Description: Names and dissects the under-labeled AGI-discourse split between “successionists” (conscious AIs should inherit the future and its resources) and humanists/speciesists, with a timeline, actor profiles, and a taxonomy of positions.
2. Prompt: /new-project successionism-speciesism Write a detailed report identifying an important dichotomy to emphasize in AGI discourse: successionism vs speciesism (or maybe “humanism” for the latter). This is the conflict between people who expect AI systems to be conscious and have moral patienthood—and should therefore be allocated ~all of the world’s resources—and those who for whatever reason value humans over AIs. The latter could be motivated by axiomatic “speciesism” or similar ideologies, or by a belief that AI systems aren’t conscious/don’t have qualia, or aren’t moral patients for whatever reasons. Maybe successionism could be divided into “hard” successionism and soft successionism where hard successionism says that it’s morally essential to create a vast number of conscious, moral-patient AI systems and eliminate human beings as an inefficient use of resources compared to the happiness produced. Soft successionism might say that AI systems should be given “human rights” or other protections for various reasons, maybe because their moral patienthood and consciousness demands it morally, or because it’s a more practical political/economic system, or for other reasons; this is still functionally successionism because the future will be dominated by the more numerous and productive AI systems in the future in this scenario.
  So produce a very detailed report with:
  - a comprehensive timeline of writing related to this issue, along with analysis and summaries of the relevant work
  - a comprehensive profile of the writing, thoughts, work, activism, etc. of all of the relevant actors/thought leaders/activists in this space—include the new moral-philosophy papers of MacAskill and Hendrycks, for example
  - the views of established people like frontier-lab executives, AI safety researchers, mainstream politicians and politically prominent people, etc
  - a taxonomy of the different perspectives and viewpoints on the issue
  - other relevant information about understanding this new division and topic related to AGI
What Do Adult Literacy Assessments Actually Measure? Examining the Gap Between Public Claims and Test Content
1. Description: Deconstructs widely-cited adult-illiteracy figures by showing that the underlying NALS/NAAL/PIAAC assessments test inference, interpretation, and document navigation rather than reading per se, surfacing the actual test questions and their success rates so readers can judge for themselves.
2. Research Spec: Investigate common claims about American adult “illiteracy” rates and low reading levels. The thesis is that these metrics are misleading because the underlying assessments test logical inference, interpretive reasoning, and document navigation skills beyond literacy alone. The report should make it easy to see what studies are being referenced in public discourse and examine actual test questions with their success rates, enabling readers to judge for themselves what the metrics really measure.
The Quadrivium Abandoned: How Modern Classical Education Betrayed Its Mathematical Heritage
1. Description: Argues the modern classical-education movement recovered the trivium but abandoned the quadrivium — neglecting the rigorous mathematics that defined classical learning from Plato to Newton — and proposes a “relative difficulty” standard pegged to each era’s own frontier.
2. Research Spec: Argue that modern classical education fails its own historical standard by neglecting rigorous mathematics—the very discipline that defined classical learning from Plato’s Academy through Newton. The modern classical education movement has recovered the trivium but largely abandoned the quadrivium. Building on Clark and Jain’s critique, propose a “relative difficulty” standard: classical students should engage with material at a comparable level of difficulty and frontier-proximity relative to their era, not simply study what the ancients studied.
Greater San Francisco: The Case for Bay Area Municipal Consolidation
1. Description: Argues for Bay Area municipal consolidation (folding Oakland, Berkeley, Fremont, San Jose, and others into an expanded San Francisco), covering the history of proposals, today’s fragmented governance, concrete consolidation plans, analogous cases, and the data-backed benefits.
2. Research Spec: Argue for and explain in detail the case for Bay Area municipal consolidation — expanding San Francisco to incorporate surrounding municipalities and counties bordering the Bay, including Oakland, Berkeley, Fremont, San Jose, and others. Cover the history of consolidation proposals, the current fragmented governance situation, concrete consolidation plans (gradual vs. all-at-once), analogous cases elsewhere (domestic and international), and the quantitative benefits supported by data and statistics. Synthesize arguments from existing reports, essays, and policy analyses into a coherent case.
The Meter and the Membership: Flat-Rate Bias, the Taxi-Meter Effect, and the Economics and Psychology of Continuous Payment
1. Description: A literature report on why people dislike metered/”continuous” pricing (per-minute, per-second) and overpay for flat or bundled plans, taxonomizing the pricing structures and mechanisms and surveying five theory families (rational option-value, biased beliefs, hedonic pain-of-paying, loss aversion, self-control) across 15 real-world domains.
2. Prompt: /new-project flat-rate-bias Write a detailed and expansive report on the literature around the idea that people don’t paying for things “continuously” e.g. paying per minute on Netflix or paying per second of wait time for Uber (vs $5 penalty after 5min). Relate this to the “flat-rate bias” and “taxi-meter effect” and such concepts. Make sure to include a comprehensive list of other things like gym memberships, taxi meters, and so one where this kind of effect appears. Make a taxonomy and describe in detail all the theories about this phenomenon, their evidence and counterevidence, and the implications for individual and business decision-making.
The Variance of Human Lifespans: Mean, Median, and Especially the Spread of Ages at Death, across Time and Space
1. Description: Builds datasets, tables, and charts for the mean, median, and especially the variance of the age-at-death distribution across US history, other countries, and the world, documenting every source and how they’re combined.
2. Prompt: /new-project lifespan-variance I want to create datasets and associated tables and charts for the mean and median and especially the variance in life expectancy at birth, across time and space. For example, I want an estimate of the variance of the life expectancy at birth for each year in US history that it’s possible to find or estimate data. And likewise for as many other countries as possible, and ideally for the entire world as a whole (using the best and most appropriate estimates). Collect as much data as you can and included a detailed write-up of all the different sources of data, an evaluation of their size and methodology, and how they are combined into the final estimates/conclusions.
The Anatomy of Western Criminal Justice: A Comparative Analysis of Criminal Procedure and Police Powers in Eleven Countries
1. Description: An element-by-element comparison of criminal procedure and police powers across 11 Western countries, with a table per element and case law throughout.
2. Prompt: /new-project criminal-law-variation Write an analysis of how criminal law varies between Western countries (including the USA, Canada, Australia, New Zealand, UK, France, Spain, Germany, Italy, Denmark, and Sweden), based on elements like judge and jury, silence and self-incrimination, access to counsel, bail and detention, exclusionary and evidence rules, double jeopardy, and standards of proof. Also include rules about police and investigative conduct related to probably cause, search and seizure, stop and search, surveillance and wiretapping, undercover ops and entrapment, use of force, interrogation methods, DNA collection, and oversight. Also include other relevant topics. Make tables etc. Include relevant court cases.
Forager and Farmer: Robin Hanson's Schema of Human Values — Chronology, Evidence, Reception, and Implications for AI Futures
1. Description: A cataloging + analysis of Hanson’s forager-vs-farmer values theory: a dated chronology of his writings, its evidence base, its reception, and a structured explainer running through to AI/AGI implications.
2. Prompt: /new-project farmer-forager Write a detailed report and analysis of Robin Hanson’s concept of the farmer and forager. That should include a number of sections, including:
  - An introduction with a list of key points and schemas
  - A chronology of all of his blog posts and maybe papers that are releevant, with a paragraph or such details of each where, he discusses the farmer and forager schema or analysis or implications
  - A discussion of all the other papers or books or secondary sources that Hanson uses as part of the evidence for this thesis
  - The reception of this idea in other papers, essays, or online discourse, including criticism and counter-criticism
  - A detailed analysis or explainer that can be used for all of the different relevant theses or aspects or implications or mental models that are a part of this general idea, including implications for AI and AGI futures

[-]Richard_Kennaway3h*20

Prompts 1, 2, 3, and 7 all write the bottom line of the argument that Claude is to develop. Does Claude generate equally convincing papers if instructed to argue the opposite of these things?

Has Claude ever resisted such a bottom-line prompt and insisted on a different conclusion? In my experience pushback from any LLM is rare except when coming up against its guardrails.

A while back I tried having an LLM argue all sides of an issue, and found it would argue for whatever I told it to. I haven't tried again recently.

I'm not saying this is in itself bad, only that it's what they do. But "I asked Claude to say a thing and it said the thing" isn't an argument for the thing.

19

Posting Some Prompts

19

19

19