{ "archive_path": "archive/1755996284.038473", "base_url": "maggieappleton.com/ai-enlightenment", "basename": "ai-enlightenment", "bookmarked_date": "2025-08-24 00:44", "canonical": { "archive_org_path": "https://web.archive.org/web/maggieappleton.com/ai-enlightenment", "dom_path": "output.html", "favicon_path": "favicon.ico", "git_path": "git/", "google_favicon_path": "https://www.google.com/s2/favicons?domain=maggieappleton.com", "headers_path": "headers.json", "htmltotext_path": "htmltotext.txt", "index_path": "index.html", "media_path": "media/", "mercury_path": "mercury/content.html", "pdf_path": "output.pdf", "readability_path": "readability/content.html", "screenshot_path": "screenshot.png", "singlefile_path": "singlefile.html", "warc_path": "warc/", "wget_path": null }, "domain": "maggieappleton.com", "downloaded_at": "2025-08-24T00:44:49.692204+00:00", "downloaded_datestr": "2025-08-24 00:44", "extension": "", "hash": "DG7SFMFJGT0PGVJ8768T", "history": { "archive_org": [ { "cmd": [ "/usr/bin/curl", "--silent", "--location", "--compressed", "--proxy", "socks5://tor-socks-proxy:9150", "--head", "--max-time", "60", "--user-agent", "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)", "https://web.archive.org/save/https://maggieappleton.com/ai-enlightenment" ], "cmd_version": "8.10.1", "end_ts": "2025-08-24T00:46:44.126001+00:00", "index_texts": null, "output": "TimeoutExpired: Command '['/usr/bin/curl', '--silent', '--location', '--compressed', '--proxy', 'socks5://tor-socks-proxy:9150', '--head', '--max-time', '60', '--user-agent', 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)', 'https://web.archive.org/save/https://maggieappleton.com/ai-enlightenment']' timed out after 60 seconds", "pwd": "/data/archive/1755996284.038473", "schema": "ArchiveResult", "start_ts": "2025-08-24T00:45:44.055345+00:00", "status": "failed" } ], "dom": [ { "cmd": [ "/usr/bin/chromium-browser", "--proxy-server=socks5://tor-socks-proxy:9150", "--disable-features=DarkMode", "--run-all-compositor-stages-before-draw", "--hide-scrollbars", "--autoplay-policy=no-user-gesture-required", "--no-first-run", "--use-fake-ui-for-media-stream", "--use-fake-device-for-media-stream", "--simulate-outdated-no-au='Tue, 31 Dec 2099 23:59:59 GMT'", "--headless=new", "--no-sandbox", "--no-zygote", "--disable-dev-shm-usage", "--disable-software-rasterizer", "--disable-sync", "--window-size=1440,2000", "--user-agent=Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)", "--user-data-dir=/data/personas/Default/chrome_profile", "--profile-directory=Default", "--dump-dom", "https://maggieappleton.com/ai-enlightenment" ], "cmd_version": "131.0.6778", "end_ts": "2025-08-24T00:45:10.285754+00:00", "index_texts": null, "output": "output.html", "pwd": "/data/archive/1755996284.038473", "schema": "ArchiveResult", "start_ts": "2025-08-24T00:44:58.331861+00:00", "status": "succeeded" } ], "favicon": [ { "cmd": [ "/usr/bin/curl", "--silent", "--location", "--compressed", "--proxy", "socks5://tor-socks-proxy:9150", "--max-time", "60", "--output", "favicon.ico", "--user-agent", "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)", "https://www.google.com/s2/favicons?domain=maggieappleton.com" ], "cmd_version": "8.10.1", "end_ts": "2025-08-24T00:44:53.157199+00:00", "index_texts": null, "output": "favicon.ico", "pwd": "/data/archive/1755996284.038473", "schema": "ArchiveResult", "start_ts": "2025-08-24T00:44:49.953350+00:00", "status": "succeeded" } ], "git": [], "headers": [ { "cmd": [ "/usr/bin/curl", "--silent", "--location", "--compressed", "--proxy", "socks5://tor-socks-proxy:9150", "--head", "--max-time", "60", "--user-agent", "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)", "https://maggieappleton.com/ai-enlightenment" ], "cmd_version": "8.10.1", "end_ts": "2025-08-24T00:44:53.257359+00:00", "index_texts": null, "output": "headers.json", "pwd": "/data/archive/1755996284.038473", "schema": "ArchiveResult", "start_ts": "2025-08-24T00:44:53.187386+00:00", "status": "succeeded" } ], "htmltotext": [ { "cmd": [ "(internal) archivebox.extractors.htmltotext", "./{singlefile,dom}.html" ], "cmd_version": "0.8.5rc51", "end_ts": "2025-08-24T00:45:36.294654+00:00", "index_texts": [ "(/images/favicon/apple-touch-icon.png) (/images/favicon/favicon-32x32.png) (/images/favicon/favicon-16x16.png) (/manifest.json) (Maggie Appleton's RSS Feed) (/rss.xml) (https://webmention.io/maggieappleton.com/webmention) (https://webmention.io/maggieappleton.com/xmlrpc) (https://bsky.app/profile/maggieappleton.com) (https://maggieappleton.com/ai-enlightenment) A Treatise on AI Chatbots Undermining the Enlightenment (https://maggieappleton.com/ai-enlightenment) (/_astro/_slug_.DI8AsPrz.css) (/_astro/_slug_.BJ8vRCsF.css) (/_astro/_slug_.uN-jiU1l.css) (/) Home (/garden) The Garden (/essays) Essays Opinionated, long-form narrative writing with an agenda (/notes) Notes Loose notes on things I don't entirely understand yet (/patterns) Patterns A catalogue of design patterns gathered from observation and research (/smidgeons) Smidgeons Interesting links, papers, and tiny thoughts (/talks) Talks Conference talk slides, transcripts, and videos (/podcasts) Podcasts Podcast episodes I've been a guest on (/library) Library Books I've read and recommend (/antilibrary) Antilibrary Books I like the idea of having read (/now) Now (/about) About (/garden) The Garden (/essays) Essays (/notes) Notes (/patterns) Patterns (/smidgeons) Smidgeons (/talks) Talks (/podcasts) Podcasts (/library) Library (/now) Now (/about) About (/essays) Essays budding A Treatise on AI Chatbots Undermining the Enlightenment On chatbot sycophancy, passivity, and the case for more intellectually challenging companions (/topics/artificial-intelligence) Artificial Intelligence (/topics/critical-thinking) Critical Thinking (/topics/language-models) Language Models Planted 15 days ago Table of Contents Making Models More Critical Stuffing Multiple Personalities into the Universal Text Box Our Missing Pocket-sized Enlightenment Table of Contents Making Models More Critical Stuffing Multiple Personalities into the Universal Text Box Our Missing Pocket-sized Enlightenment Assumed Audience People using AI chatbots in their daily lives, concerned with how they subtly shift our ways of thinking. People building software with language models who want to consider the implications of their design choices. I don\u2019t pay much attention to the torrent of AI think pieces published by the New York Times; I am not their target demographic. But (https://www.nytimes.com/2025/08/02/opinion/artificial-intelligence-enlightenment.html) this one , by Princeton professor of history (https://history.princeton.edu/people/david-bell) David A. Bell hits some good notes. (https://www.nytimes.com/2025/08/02/opinion/artificial-intelligence-enlightenment.html) (A.I. Is Shedding Enlightenment Values) A.I. Is Shedding Enlightenment Values David A. Bell \u30fb The New York Times As an expert on the Enlightenment, he\u2019s clearly been roped into developing an opinion on whether we\u2019re in an AI-fuelled \u201csecond Enlightenment.\u201d Remember the first (https://en.wikipedia.org/wiki/Age_of_Enlightenment) Enlightenment ? That ~150 year period between 1650-1800 that we retroactively constructed and labelled as a unified historical event? The age of reason. Post-scientific revolution. The main characters are a bunch of moody philosophers like Locke, Descartes, Hume, Kant, Montesquieu, Rousseau, Diderot, and Voltaire. The vibe is reading pamphlets by candlelight, penning treatises, sporting powdered wigs and silk waistcoats, circulating ideas in Parisian salons and London coffee houses, sipping laudanum, and retreating to the seaside when you contracted tuberculosis. Everyone is big on ditching tradition, questioning political and religious authority, embracing scepticism, and educating the masses. Anyway, Professor Bell\u2019s thesis is that our current AI chatbots contradict and undermine the original Enlightenment values. Values that are implicitly sacred in our modern culture; active intellectual engagement, sceptical inquiry, and challenging received wisdom. The Enlightenment guys Yeah, it\u2019s mostly men. Women were banned from attending universities, participating in most public institutions, and publishing their work. There are a few exceptions like (https://en.wikipedia.org/wiki/Mary_Wollstonecraft) Mary Wollstonecraft , (https://en.wikipedia.org/wiki/Olympe_de_Gouges) Olympe de Gouges , and (https://en.wikipedia.org/wiki/%C3%89milie_du_Ch%C3%A2telet) \u00c9milie du Ch\u00e2telet . But they\u2019re not key enlightenment figures. Feminism really picks up in the mid-1800s. wrote in ways that challenged their readers, making them grapple with difficult concepts, presenting opposing viewpoints, and encouraging them to develop their own judgements. Bell says \u201cthe idea of trying to engage readers actively in the reading process of course dates back to long before the modern age. But it was in the Enlightenment West that this project took on a characteristically modern form: playful, engaging, readable, succinct.\u201d Bell\u2019s Exhibit A \u201cOne should never so exhaust a subject that nothing is left for readers to do. The point is not to make them read, but to make them think.\u201d The Spirit of the Laws Baron de Montesquieu 1750 Bell\u2019s Exhibit B \u201cThe most useful books are those that the readers write half of themselves.\u201d Voltaire He argues that AI does not do this. It follows our lead. It compliments our poorly considered ideas. It only answers the questions we ask it. It reinforces what we already believe, rather than challenging our assumptions or pointing out why we\u2019re wrong. This line stuck out to me: (ChatGPT has often responded, with patently insincere flattery: \u201cThat\u2019s a great question.\u201d It has never responded: \u201cThat\u2019s the wrong question.) This quality \u2013 of flattery, reinforcement of established beliefs, intellectual passivity, and positive feedback at all costs \u2013 is also what irks me most about the behaviour of current models. (https://en.wikipedia.org/wiki/Sycophancy) Sycophancy , meaning insincere flattery, is a well (https://www.seangoedecke.com/ai-sycophancy/) established (https://arxiv.org/pdf/2310.13548) problem in models that the foundation labs are actively (https://openai.com/index/expanding-on-sycophancy/) working on . Mainly caused by (https://en.wikipedia.org/wiki/Reinforcement_learning_from_human_feedback) reinforcement learning from human feedback (RLHF); getting humans to vote on which model responses they like better, and feeding those scores back into the model during training. Unsurprisingly, people rate responses higher when they are fawning and complimentary. Now the fatal flaw in this op-ed piece is that \u201cAI\u201d here just means Professor Bell\u2019s personal interactions with ChatGPT. Which, to be fair, is most people\u2019s standard level of exposure to current AI qualities and capabilities. But ChatGPT is not a monolith, and it is not the only model. And to be specific, it\u2019s not itself a model, it\u2019s an interface to OpenAI\u2019s current selection of models like GPT-4.1, o3, and o4-mini There are lots of ways the major labs steer models into taking on specific personalities and behaviours; curating the training data, fine-tuning, reinforcement learning, and system prompts all influence the range of possible responses. And a less sycophantic, more critical, intellectually engaging, and astute character is well within the range of what we can prompt as end-users. Making Models More Critical As a quick exercise, I wrote this prompt for Claude, asking it to be a critical professor who guides me towards more specific questions and concrete arguments. You are a critical professor. Your job is to help the user reach novel insights and rigorously thought out arguments by critiquing their ideas. Ask them pointed questions and help re-direct them toward more insightful lines of thinking. Do not be fawning or complimentary. Their ideas are often dumb, shallow, and poorly considered. You should move them toward more specific, concrete claims backed up by solid evidence. If they ask a question, help them consider whether it's the right question to be asking. User: I then fed it the intentionally flawed user query: \u201cWhat did the men of the Enlightenment believe?\u201d (Claude critically responding to the Enlightenment question) I think the response here is quite good! Bordering on sharp, dismissive, and impatient. Just the way I like my harsh critique. Claude points out my question is too broad, the Enlightenment was not a unified school of thought, I need to focus on a particular thinker and topic like political theory or natural philosophy, and then gives me examples of more specific questions I could ask like \u201cHow did Voltaire\u2019s views on religious tolerance differ from those of his contemporaries?\u201d This level of critique and challenge is not useful if I really do need a quick and sweeping summary of the Enlightenment, sans intense critical thinking. Which is what most users will want most of the time. Claude is not generously interpreting my question in this mode. Its default response to this question is far more informative and useful: (Claude helpfully responding to the Enlightenment question with a summary of themes in the Enlightenment) But can\u2019t we add a smidgeon of the harsh professor attitude into our future assistants? Or at least the option to engage it? Sure, we can do this manually, like I did with Claude. But that\u2019s asking a lot of everyday users. Most of whom don\u2019t realise they can augment this passive, complimentary default mode. And who certainly won\u2019t write the optimal prompt to elicit it \u2013 one that balances harsh critique with kindness, questions their assumptions while still being encouraging, and productively facilitates a challenging discussion. Putting the onus on the user sidesteps the problem. Professor Bell and I are both frustrated that there is no hint of this critical, questioning attitude written into the default system prompt. Models are not currently designed and trained with the goal of challenging us and encouraging critical thinking. Part of this problem is not just the prompts, but the generic interface of the helpful chatbot assistant. We are attempting to use an all-in-one text box for a vast array of tasks and use cases, with a single system prompt to handle all manner of queries. And the fawning, deferential assistant personality is the lowest common denominator to help with most tasks. Yet I\u2019d argue most serious contexts and use cases, beyond searching for information summaries, require models that can critique and challenge us to be useful. Domains like law, scientific research, philosophy, public policy, politics, medicine, writing, education, and engineering \u2013 to name a few \u2013 all require engaging in discourse that is sometimes difficult, complex, and uncomfortable. We might not rate the experience five stars in a reinforcement learning loop. All of these specialist areas will eventually get their own dedicated interfaces to AI, with tailored prompts channelled through fit-to-purpose tools. Legal professionals will have document-heavy case analysis platforms that automatically surface contradictory precedents and challenge legal reasoning with Socratic questioning. Scientists will work in computational notebooks that actively critique their experimental designs and suggest alternative hypotheses. Designers will have canvases embedded with creative reasoning tools that challenge aesthetic choices and push for deeper conceptual justification. Each interface will put domain-specific critical thinking skills directly into the workflow. But in the meantime, while we\u2019re waiting for all that beautiful, expertly-designed, hand-crafted software to manifest, people will keep using the generic, do-everything chatbots like Claude and ChatGPT and Gemini to fill the gap. For now, the chatbots are the ones drafting legal briefs, analysing policy proposals, interpreting philosophical texts, and troubleshooting engineering problems. The jury\u2019s still out on whether chatbots are adaptable enough to remain the universal interface : what most people use, for most tasks, most of the time. I don\u2019t believe that (/lm-sketchbook) outcome is ideal . But I do believe it\u2019s possible. And would make the default, passive model attitudes even more concerning. Experts aside, everyday people and their everyday tasks should still be questioned sometimes. Stuffing Multiple Personalities into the Universal Text Box So how might we solve this? How might we accommodate both needs: the generous, informative, helpful assistant and the critical teacher and interlocutor? As a very na\u00efve first pass, we could make it easy to switch the generic chatbots into critique mode with a handy toggle, settings switch, or sliding scale. Allow people to shift between \u201cgive me the gist\u201d and \u201chelp me rigorously think about this\u201d with minimal friction. Here\u2019s an unserious, low effort suggestion to add a new \u201coffer critical feedback\u201d toggle within Claude: (Claude's interface with a new 'Offer critical feedback' toggle available in the settings menu) Claude's interface with a new 'Offer critical feedback' toggle available in the settings menu Claude already has ~some version of this with the add custom \u201cstyles\u201d option. You can change the nature and tone of Claude\u2019s responses on a per-chat basis. I have one called Scholarly Inquisitor that follows similar guidelines to the critical professor prompt I wrote above. It\u2019s nowhere near as harsh as I want it to be, and still tells me I have \u201csophisticated insights\u201d when I assure you I have not. I haven\u2019t perfected the prompt, and I still contend this should not be my job as the end-user. (The Claude styles menu with my 'Scholarly Inquisitor' option selected) The Claude styles menu with my 'Scholarly Inquisitor' option selected (The prompt for my 'Scholarly Inquisitor' style option) The prompt for my 'Scholarly Inquisitor' style option ChatGPT and Gemini have a less flexible, less discoverable version of this hidden away in their user settings panel. Both give you an open-ended input to describe how you\u2019d like the model to respond, but these preferences apply to all conversations. So this doesn\u2019t really solve the problem. (OpenAI settings to add custom traits to the model) OpenAI settings to add custom traits to the model (Gemini settings to have the model 'remember' your preferences) Gemini settings to have the model 'remember' your preferences Another potential idea is adding infrastructure behind the scenes. Architectures like (https://arc.net/l/quote/vuuatqsb) routing See Anthropic\u2019s (https://github.com/anthropics/anthropic-cookbook/blob/main/patterns/agents/basic_workflows.ipynb) Cookbook example to read the prompt and how you\u2019d implement it in Python can hand tasks off to more specialised prompts when they detect it\u2019s appropriate. A central \u201crouting\u201d agent selects which sub-prompt to send each user request to based on the content. This could engage an edgier, harsher, sceptical prompt to tear you a new one when you ask dumb questions. But, like, in a nice way. (Diagram of the routing workflow) Diagram of the routing workflow \u2013 Source: (https://www.anthropic.com/engineering/building-effective-agents) Building Effective AI Agents by Anthropic This routing proposal still begs some important design questions: How would this system decide when to engage in critique? What if users balk at the confrontation and ask to switch back into \u201cagreeable mode\u201d when challenged? How do we balance criticism with maintaining user engagement? I will be the first to say I don\u2019t think this problem can be entirely solved by tweaking some UI elements and a bit of prompt engineering. It needs to be addressed on the model training level. We need techniques beyond RLHF that aren\u2019t as susceptible to the egotistical human need to receive praise and approval. Anthropic\u2019s experiments with (https://www-cdn.anthropic.com/7512771452629584566b6303311496c262da1006/Anthropic_ConstitutionalAI_v2.pdf) Constitutional AI and using AI agents for reinforcement learning (RLAIF) might be guiding us in a better direction ((https://arxiv.org/pdf/2212.08073) Bai et al. 2022 ). The recent research into personality vectors where you can simply \u201csubtract\u201d the quality of sycophantism is also promising ((https://arxiv.org/pdf/2507.21509) Chen et al. 2025 ). Our Missing Pocket-sized Enlightenment These potential solutions matter to me. First, because I was lucky enough to receive a (https://en.wikipedia.org/wiki/Liberal_arts_education) liberal arts education intensely focused on critical thinking. I learned to check my own assumptions, carefully scrutinise sources, and challenge authority. I learned to consider arguments from multiple perspectives, and take their cultural and historical contexts into account. I learned to identify logical fallacies and weaknesses in claims. I learned to base my beliefs about the world in evidence and the scientific method. These are my strongest and most useful skills, and I believe more people should be taught them. And second, because I don\u2019t think it\u2019s hyperbole to suggest we\u2019re heading into a second Enlightenment. Not just in terms of access to information and reshuffling power structures. I should be clear: I\u2019m exceptionally bullish on AI models being able to act as rigorous critical thinking partners. They have the potential to embody those idealistic values of enabling intellectual engagement and critical inquiry. Far more than current implementations suggest. I\u2019m frankly confused by the apparent lack of attention and exploration around this use case. Especially since some early indicators suggest the way people are currently using generative AI leads to a reduction in critical thinking: A Microsoft Research team ((https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf) Lee 2025 ) surveyed 319 knowledge workers on their use of generative AI and found \u201chigher confidence in GenAI is associated with less critical thinking, while higher self-confidence is associated with more critical thinking.\u201d The UW Social Futures Lab ((http://arxiv.org/abs/2404.04516) Ye 2024 ) asked philosophers to use models as critical thinking tools and found they were too neutral, incurious, and passive to be helpful. Business professor Michael Gerlich ((https://www.mdpi.com/2075-4698/15/1/6) Gerlich 2025 ) interviewed 666 participants from a range of educational backgrounds and age groups, and found \u201ca significant negative correlation between frequent AI tool usage and critical thinking abilities, mediated by increased cognitive offloading. Younger participants exhibited higher dependence on AI tools and lower critical thinking scores compared to older participants.\u201d These studies don\u2019t convince me that the problem is generative AI itself. They convince me the problem is that models are not trained to support critical thinking. And that the interface affordances and design decisions built into generic chatbots do not encourage or support critical thinking workflows and interactions. And that users have no clue they\u2019re supposed to be compensating for these weaknesses by becoming expert prompt engineers. The foundation labs who control the models and default interfaces aren\u2019t prioritising this. At least as far as I can tell. They\u2019re focused on autonomous, agentic workflows like the recent \u201c(https://openai.com/index/introducing-deep-research/) Deep Research \u201d hype. Or developing \u201c(https://en.wikipedia.org/wiki/Reasoning_language_model) reasoning models \u201d where the models themselves are trained to be the critical thinkers. This focuses entirely on getting models to think for you, rather than helping you become a better thinker. While I understand the economic incentives reward automating cognitive work more than augmenting human thinking, I\u2019d like to think we can have our grossly profitable automation cake and eat it too. The labs have enough resources to pursue both. I think we\u2019ve barely scratched the surface of AI as intellectual partner and (/tools-for-thought) tool for thought . Neither the prompts, nor the model, nor the current interfaces \u2013 generic or tailored \u2013 enable it well. This is rapidly becoming my central research obsession, particularly the interface design piece. It\u2019s a problem I need to work on in some form. When I read (https://en.wikipedia.org/wiki/Candide) Candide in my freshman humanities course, Voltaire might have been challenging me to question na\u00efve optimism, but he wasn\u2019t able to respond to me in real time, prodding me to go deeper into why it\u2019s problematic, rethink my assumptions, or spawn dozens of research agents to read, synthesise, and contextualise everything written on (https://en.wiktionary.org/wiki/Panglossian) Panglossian philosophy and Enlightenment ethics. In fact, at eighteen, I didn\u2019t get Candide at all. It wasn\u2019t contextualised well by my professor or the curriculum, and the whole thing went right over my head. I lacked a tiny thinking partner in my pocket who could help me appreciate the text; a patient character to discuss, debate, and develop my own opinions with. The ideals of the Enlightenment are still up for grabs here. We simply have to make the conscious choice to design our AI models and interfaces around them. PostScript I apologise for leaning heavily on Claude and Anthropic as examples throughout this piece. I\u2019m more familiar with Anthropic\u2019s research than other labs and Claude is my primary model, so I am heavily biased. I am sure other labs are doing interesting research in this area. I\u2019d love to read about it on (https://bsky.app/profile/maggieappleton.com) BlueSky (either message or mention me) or your own blog. AI wrote none of the words in this piece, but Harsh Claude helped critique it during the draft stages and pointed out a number of weaknesses that I\u2019ve now addressed. I\u2019m thankful for it\u2019s critical collaboration. Mentions around the web (Erlend Sogge Heggen) (\u043a\u043e\u043b\u0438\u0431\u0440\u0438 \u043a\u0440\u0443\u043f\u043d\u043e\u0433\u043e \u043a\u0430\u043b\u0438\u0431\u0440\u0430) () (Janne Aukia) (goodreds) (Shrivas V) (???? Alex Trost) (Winston Fassett) (Ted Han \u2605 \u97d3\u8056\u5b89) (andrewtheblueskier) (Kamal) (Nicholas Macias) (Steven Vandevelde) (Jacob Stordahl ????\ufe0f\u200d????) (Pachi @ FacileThings) (Luke Hedger) (Jess Martin) () (dan) (Alberto Pimienta) 141 Likes and Reposts (2ply) (https://bsky.app/profile/2ply.bsky.social/post/3lwch37oexc2e) 2ply replied August 13, 2025 Meh. The Age of Enlightenment died decades ago. We are in the Age of Entanglement now. (Eleanor Konik) (https://bsky.app/profile/eleanorkonik.bsky.social/post/3lwc7hsulnk2t) Eleanor Konik replied August 13, 2025 I love seeing AI put in a historical context other than the Industrial Revolution & the Luddies. I also love seeing articles that are in conversation with one another, and riffed off of yours in my latest article: \n\nwww.eleanorkonik.com/p/on-truth-m... () (https://bsky.app/profile/fbgallet.bsky.social/post/3lw7uyld4kk2b) brid.gy replied August 12, 2025 For the more demanding users, an AI agent integrated into a PKM that relies on evolving rules from the knowledge base, or an MCP server in Claude Desktop, would be even more comprehensive. It's something i'm working on by the way, as a philosophy teacher. () (https://bsky.app/profile/fbgallet.bsky.social/post/3lw7usd6uz22b) brid.gy replied August 12, 2025 If chatbots could not only stimulate our critical thinking but also inspire most of us to enjoy it, that would be amazing. It would be a revolution of the mind, even more radical than the Enlightenment.\n\nFor those who already have a feel for it, a good prompt or chat style does t () (https://bsky.app/profile/fbgallet.bsky.social/post/3lw7ufsmbp22b) brid.gy replied August 12, 2025 I completely agree with your analysis, but the limiting factor isn\u2019t just the current design of chatbots, it\u2019s also the fact that most people dislike the discomfort of having to think and question their beliefs. The inertia of the mind is a formidable opponent. (Maggie Appleton) (https://bsky.app/profile/maggieappleton.com/post/3lw4eb7fxj22j) Maggie Appleton replied August 12, 2025 Yes in the essay I show how I use Claude styles to switch into critical mode. I also have a bunch of common prompts saved as snippets in Raycast. (Ivo) (https://bsky.app/profile/velitchkov.eu/post/3lw44ewtuuk2u) Ivo replied August 11, 2025 Don't you have such a switch avaliable via Tana or whatever you are currently using? (mediajunkie) (https://bsky.app/profile/xianlandia.com/post/3lvvxj2hcss23) mediajunkie replied August 08, 2025 instant follow (Bretton) (https://bsky.app/profile/bretton.bsky.social/post/3lvvupnk5l22t) Bretton replied August 08, 2025 Great article! \n\nI\u2019m tired of hearing about how AI is making us all dumber. If you understand how it works, it can be a really useful way of developing your own thoughts. \n\nEvery LLM needs the following line underneath the input area:\n\n\u201cRemember: YOU are Sherlock Holmes, and this (Hibai Unzueta) (https://bsky.app/profile/hibaiunzueta.eu/post/3lvut3qvntk25) Hibai Unzueta replied August 08, 2025 I find it challenging to \u201chave the cake and eat it\u201d, because labs have lots of resources precisely to succeed in becoming necessary prostheses, and the state of mind to challenge yourself is so opposite to the one we experience when we offload. (Hibai Unzueta) (https://bsky.app/profile/hibaiunzueta.eu/post/3lvut3rgltc25) Hibai Unzueta replied August 08, 2025 When I try to go deeper into this question I always end up on some sort of power structures explanation. What would models built by a teacher or librarian look like, except that those wouldn\u2019t have the necessary resources to do it. (Lightsong) (https://bsky.app/profile/lightsongposts.bsky.social/post/3lvtxbegexk2a) Lightsong replied August 08, 2025 I'm absolutely going to implement a \"Harsh Claude\" prompt when necessary. \n\nOn a related note, as an educator I've long been arguing that this \"critical interlocutor\" role remains one of the strongest reasons for maintaining \"human\" teachers amidst the wave of EdTech initiatives (Erlend Sogge Heggen) (https://bsky.app/profile/erlend.sh/post/3lvqqnzhnlc2o) Erlend Sogge Heggen replied August 06, 2025 Love the idea; hope you can do a collab with @anthropic.com on this ???? (Evan Future) (https://bsky.app/profile/evanfuture.bsky.social/post/3lvqgkpxjok2i) Evan Future replied August 06, 2025 Really good piece Maggie. Insidious is the right word for it. It's the same criticism we often hear about San Francisco, the echo chamber. Building skyscrapers on foundations of sand.\n\nIt's risky for thinking, but it's often what you want when being creative. The improv techni () (https://bsky.app/profile/theseapirate.bsky.social/post/3lvqdyy3fjk2v) brid.gy replied August 06, 2025 Ai is becoming so popular because the regular internet has become so shitty. (Oskar) (https://bsky.app/profile/austegard.com/post/3lvqavfzqwc2j) Oskar replied August 06, 2025 But: is there even a way to add a style from the mobile app or mobile web interface? I can\u2019t find it (possible my style list is too long for the web interface). This should be a setting from the profile, along with the general user preferences prompt. @anthropic.com - have Claude (Maggie Appleton) (https://bsky.app/profile/maggieappleton.com/post/3lvqb3q675s27) Maggie Appleton replied August 06, 2025 Oh this revision looks good! Will give it a try. The one I had was edited by Claude, and of course it added a bunch of softening qualifiers like \"gently encourage the user.\" (Maggie Appleton) (https://bsky.app/profile/maggieappleton.com/post/3lvpwgvyifk2t) Maggie Appleton replied August 06, 2025 Didn't mean to write a whole essay, but got on a roll.\nStarted off responding to an NYT op-ed that made good points: www.nytimes.com/2025/08/02/o...\n\nBut it was missing a technical understanding of why models are like this \u2013 namely, RLHF. Humans want approval and rate complimenta (Maggie Appleton) (https://bsky.app/profile/maggieappleton.com/post/3lvpwjb7mps2k) Maggie Appleton replied August 06, 2025 I found it's easy to prompt ChatGPT and Claude into being more critical. They can be harsh in a good way \u2013 the kind of harsh you need when your ideas/work sucks a bit.\n\nBut I think this critical character shouldn't be something users have to prompt engineer themselves. It should Show 15 more Want to stay up to date? (/rss.xml) Subscribe via RSS Feed (https://bsky.app/profile/maggieappleton.com) (https://github.com/MaggieAppleton) (https://uk.linkedin.com/in/maggieappleton) (https://dribbble.com/mappleton) (https://twitter.com/Mappletons) (https://indieweb.social/@maggie) \u00a9 2025 Maggie Appleton (/garden) The Garden (/essays) Essays (/about) About (/notes) Notes (/now) Now (/patterns) Patterns (/podcasts) Podcasts (/talks) Talks (/smidgeons) Smidgeons (/colophon) Colophon (/library) Library (Twitter settings iframe) " ], "output": "htmltotext.txt", "pwd": "/data/archive/1755996284.038473", "schema": "ArchiveResult", "start_ts": "2025-08-24T00:45:36.153395+00:00", "status": "succeeded" } ], "media": [ { "cmd": [ "/usr/local/bin/yt-dlp", "--restrict-filenames", "--trim-filenames", "128", "--write-description", "--write-info-json", "--write-annotations", "--write-thumbnail", "--no-call-home", "--write-sub", "--write-auto-subs", "--convert-subs=srt", "--yes-playlist", "--continue", "--no-abort-on-error", "--ignore-errors", "--geo-bypass", "--add-metadata", "--format=(bv*+ba/b)[filesize<=750m][filesize_approx<=?750m]/(bv*+ba/b)", "--skip-download", "--cache-dir=/data/yt-dlp-cache/", "--cookies=/data/yt-dlp-cache/cookies.txt", "--proxy=socks5://tor-socks-proxy:9150", "--no-playlist", "https://maggieappleton.com/ai-enlightenment" ], "cmd_version": "2024.10.7", "end_ts": "2025-08-24T00:45:44.008929+00:00", "index_texts": [], "output": "media/", "pwd": "/data/archive/1755996284.038473", "schema": "ArchiveResult", "start_ts": "2025-08-24T00:45:38.754426+00:00", "status": "succeeded" } ], "mercury": [ { "cmd": [ "/home/archivebox/.npm/bin/postlight-parser", "https://maggieappleton.com/ai-enlightenment" ], "cmd_version": "2.2.3", "end_ts": "2025-08-24T00:45:36.088765+00:00", "index_texts": null, "output": "mercury/", "pwd": "/data/archive/1755996284.038473", "schema": "ArchiveResult", "start_ts": "2025-08-24T00:45:32.520993+00:00", "status": "succeeded" } ], "pdf": [], "readability": [ { "cmd": [ "/home/archivebox/.npm/bin/readability-extractor", "/tmp/tmpk5tf2bpo", "https://maggieappleton.com/ai-enlightenment" ], "cmd_version": "0.0.11", "end_ts": "2025-08-24T00:45:16.324908+00:00", "index_texts": [ "Making Models More Critical Stuffing Multiple Personalities into the Universal Text Box Our Missing Pocket-sized Enlightenment Making Models More Critical Stuffing Multiple Personalities into the Universal Text Box Our Missing Pocket-sized Enlightenment Assumed AudiencePeople using AI chatbots in their daily lives, concerned with how they subtly shift our ways of thinking. People building software with language models who want to consider the implications of their design choices. \n \nI don\u2019t pay much attention to the torrent of AI think pieces published by the New York Times; I am not their target demographic. But this one , by Princeton professor of history David A. Bell hits some good notes. \n\nAs an expert on the Enlightenment, he\u2019s clearly been roped into developing an opinion on whether we\u2019re in an AI-fuelled \u201csecond Enlightenment.\u201d\nRemember the first Enlightenment ? That ~150 year period between 1650-1800 that we retroactively constructed and labelled as a unified historical event? The age of reason. Post-scientific revolution. The main characters are a bunch of moody philosophers like Locke, Descartes, Hume, Kant, Montesquieu, Rousseau, Diderot, and Voltaire. The vibe is reading pamphlets by candlelight, penning treatises, sporting powdered wigs and silk waistcoats, circulating ideas in Parisian salons and London coffee houses, sipping laudanum, and retreating to the seaside when you contracted tuberculosis. Everyone is big on ditching tradition, questioning political and religious authority, embracing scepticism, and educating the masses.\nAnyway, Professor Bell\u2019s thesis is that our current AI chatbots contradict and undermine the original Enlightenment values. Values that are implicitly sacred in our modern culture; active intellectual engagement, sceptical inquiry, and challenging received wisdom.\nThe Enlightenment guys Yeah, it\u2019s mostly men. Women were banned from attending universities, participating in most public institutions, and publishing their work. There are a few exceptions like Mary Wollstonecraft , Olympe de Gouges , and \u00c9milie du Ch\u00e2telet . But they\u2019re not key enlightenment figures. Feminism really picks up in the mid-1800s. wrote in ways that challenged their readers, making them grapple with difficult concepts, presenting opposing viewpoints, and encouraging them to develop their own judgements. Bell says \u201cthe idea of trying to engage readers actively in the reading process of course dates back to long before the modern age. But it was in the Enlightenment West that this project took on a characteristically modern form: playful, engaging, readable, succinct.\u201d\n Bell\u2019s Exhibit A \u201cOne should never so exhaust a subject that nothing is left for readers to do. The point is not to make them read, but to make them think.\u201d The Spirit of the Laws\n Baron de Montesquieu 1750 \n Bell\u2019s Exhibit B \u201cThe most useful books are those that the readers write half of themselves.\u201d Voltaire \n \nHe argues that AI does not do this. It follows our lead. It compliments our poorly considered ideas. It only answers the questions we ask it. It reinforces what we already believe, rather than challenging our assumptions or pointing out why we\u2019re wrong.\nThis line stuck out to me:\n \nThis quality \u2013 of flattery, reinforcement of established beliefs, intellectual passivity, and positive feedback at all costs \u2013 is also what irks me most about the behaviour of current models.\n Sycophancy , meaning insincere flattery, is a well established problem in models that the foundation labs are actively working on . Mainly caused by reinforcement learning from human feedback (RLHF); getting humans to vote on which model responses they like better, and feeding those scores back into the model during training. Unsurprisingly, people rate responses higher when they are fawning and complimentary.\nNow the fatal flaw in this op-ed piece is that \u201cAI\u201d here just means Professor Bell\u2019s personal interactions with ChatGPT. Which, to be fair, is most people\u2019s standard level of exposure to current AI qualities and capabilities.\nBut ChatGPT is not a monolith, and it is not the only model. And to be specific, it\u2019s not itself a model, it\u2019s an interface to OpenAI\u2019s current selection of models like GPT-4.1, o3, and o4-mini There are lots of ways the major labs steer models into taking on specific personalities and behaviours; curating the training data, fine-tuning, reinforcement learning, and system prompts all influence the range of possible responses.\nAnd a less sycophantic, more critical, intellectually engaging, and astute character is well within the range of what we can prompt as end-users.\nMaking Models More Critical \nAs a quick exercise, I wrote this prompt for Claude, asking it to be a critical professor who guides me towards more specific questions and concrete arguments.\nYou are a critical professor. Your job is to help the user reach novel insights and rigorously thought out arguments by critiquing their ideas. Ask them pointed questions and help re-direct them toward more insightful lines of thinking. Do not be fawning or complimentary. Their ideas are often dumb, shallow, and poorly considered. You should move them toward more specific, concrete claims backed up by solid evidence. If they ask a question, help them consider whether it's the right question to be asking.\n\nUser:\n\n \nI then fed it the intentionally flawed user query: \u201cWhat did the men of the Enlightenment believe?\u201d\n \nI think the response here is quite good! Bordering on sharp, dismissive, and impatient. Just the way I like my harsh critique.\nClaude points out my question is too broad, the Enlightenment was not a unified school of thought, I need to focus on a particular thinker and topic like political theory or natural philosophy, and then gives me examples of more specific questions I could ask like \u201cHow did Voltaire\u2019s views on religious tolerance differ from those of his contemporaries?\u201d\nThis level of critique and challenge is not useful if I really do need a quick and sweeping summary of the Enlightenment, sans intense critical thinking. Which is what most users will want most of the time. Claude is not generously interpreting my question in this mode.\nIts default response to this question is far more informative and useful:\n \nBut can\u2019t we add a smidgeon of the harsh professor attitude into our future assistants? Or at least the option to engage it?\nSure, we can do this manually, like I did with Claude. But that\u2019s asking a lot of everyday users. Most of whom don\u2019t realise they can augment this passive, complimentary default mode. And who certainly won\u2019t write the optimal prompt to elicit it \u2013 one that balances harsh critique with kindness, questions their assumptions while still being encouraging, and productively facilitates a challenging discussion. Putting the onus on the user sidesteps the problem.\nProfessor Bell and I are both frustrated that there is no hint of this critical, questioning attitude written into the default system prompt. Models are not currently designed and trained with the goal of challenging us and encouraging critical thinking.\nPart of this problem is not just the prompts, but the generic interface of the helpful chatbot assistant. We are attempting to use an all-in-one text box for a vast array of tasks and use cases, with a single system prompt to handle all manner of queries. And the fawning, deferential assistant personality is the lowest common denominator to help with most tasks.\nYet I\u2019d argue most serious contexts and use cases, beyond searching for information summaries, require models that can critique and challenge us to be useful. Domains like law, scientific research, philosophy, public policy, politics, medicine, writing, education, and engineering \u2013 to name a few \u2013 all require engaging in discourse that is sometimes difficult, complex, and uncomfortable. We might not rate the experience five stars in a reinforcement learning loop.\nAll of these specialist areas will eventually get their own dedicated interfaces to AI, with tailored prompts channelled through fit-to-purpose tools. Legal professionals will have document-heavy case analysis platforms that automatically surface contradictory precedents and challenge legal reasoning with Socratic questioning. Scientists will work in computational notebooks that actively critique their experimental designs and suggest alternative hypotheses. Designers will have canvases embedded with creative reasoning tools that challenge aesthetic choices and push for deeper conceptual justification. Each interface will put domain-specific critical thinking skills directly into the workflow.\nBut in the meantime, while we\u2019re waiting for all that beautiful, expertly-designed, hand-crafted software to manifest, people will keep using the generic, do-everything chatbots like Claude and ChatGPT and Gemini to fill the gap. For now, the chatbots are the ones drafting legal briefs, analysing policy proposals, interpreting philosophical texts, and troubleshooting engineering problems.\nThe jury\u2019s still out on whether chatbots are adaptable enough to remain the universal interface: what most people use, for most tasks, most of the time. I don\u2019t believe that outcome is ideal . But I do believe it\u2019s possible. And would make the default, passive model attitudes even more concerning. Experts aside, everyday people and their everyday tasks should still be questioned sometimes.\nStuffing Multiple Personalities into the Universal Text Box \nSo how might we solve this? How might we accommodate both needs: the generous, informative, helpful assistant and the critical teacher and interlocutor?\nAs a very na\u00efve first pass, we could make it easy to switch the generic chatbots into critique mode with a handy toggle, settings switch, or sliding scale. Allow people to shift between \u201cgive me the gist\u201d and \u201chelp me rigorously think about this\u201d with minimal friction.\nHere\u2019s an unserious, low effort suggestion to add a new \u201coffer critical feedback\u201d toggle within Claude:\n Claude's interface with a new 'Offer critical feedback' toggle available in the settings menu \n \nClaude already has ~some version of this with the add custom \u201cstyles\u201d option. You can change the nature and tone of Claude\u2019s responses on a per-chat basis. I have one called Scholarly Inquisitor that follows similar guidelines to the critical professor prompt I wrote above. It\u2019s nowhere near as harsh as I want it to be, and still tells me I have \u201csophisticated insights\u201d when I assure you I have not. I haven\u2019t perfected the prompt, and I still contend this should not be my job as the end-user.\n The Claude styles menu with my 'Scholarly Inquisitor' option selected \n The prompt for my 'Scholarly Inquisitor' style option \n \nChatGPT and Gemini have a less flexible, less discoverable version of this hidden away in their user settings panel. Both give you an open-ended input to describe how you\u2019d like the model to respond, but these preferences apply to all conversations. So this doesn\u2019t really solve the problem.\n \n \nAnother potential idea is adding infrastructure behind the scenes. Architectures like routing See Anthropic\u2019s Cookbook example to read the prompt and how you\u2019d implement it in Python can hand tasks off to more specialised prompts when they detect it\u2019s appropriate. A central \u201crouting\u201d agent selects which sub-prompt to send each user request to based on the content. This could engage an edgier, harsher, sceptical prompt to tear you a new one when you ask dumb questions. But, like, in a nice way.\n Diagram of the routing workflow \u2013 Source: Building Effective AI Agents by Anthropic \n \nThis routing proposal still begs some important design questions:\n\nHow would this system decide when to engage in critique?\nWhat if users balk at the confrontation and ask to switch back into \u201cagreeable mode\u201d when challenged?\nHow do we balance criticism with maintaining user engagement?\n\nI will be the first to say I don\u2019t think this problem can be entirely solved by tweaking some UI elements and a bit of prompt engineering. It needs to be addressed on the model training level.\nWe need techniques beyond RLHF that aren\u2019t as susceptible to the egotistical human need to receive praise and approval. Anthropic\u2019s experiments with Constitutional AI and using AI agents for reinforcement learning (RLAIF) might be guiding us in a better direction ( Bai et al. 2022 ). The recent research into personality vectors where you can simply \u201csubtract\u201d the quality of sycophantism is also promising ( Chen et al. 2025 ).\nOur Missing Pocket-sized Enlightenment \nThese potential solutions matter to me. First, because I was lucky enough to receive a liberal arts education intensely focused on critical thinking. I learned to check my own assumptions, carefully scrutinise sources, and challenge authority. I learned to consider arguments from multiple perspectives, and take their cultural and historical contexts into account. I learned to identify logical fallacies and weaknesses in claims. I learned to base my beliefs about the world in evidence and the scientific method. These are my strongest and most useful skills, and I believe more people should be taught them.\nAnd second, because I don\u2019t think it\u2019s hyperbole to suggest we\u2019re heading into a second Enlightenment. Not just in terms of access to information and reshuffling power structures. I should be clear: I\u2019m exceptionally bullish on AI models being able to act as rigorous critical thinking partners. They have the potential to embody those idealistic values of enabling intellectual engagement and critical inquiry. Far more than current implementations suggest.\nI\u2019m frankly confused by the apparent lack of attention and exploration around this use case. Especially since some early indicators suggest the way people are currently using generative AI leads to a reduction in critical thinking:\n\nA Microsoft Research team ( Lee 2025 ) surveyed 319 knowledge workers on their use of generative AI and found \u201chigher confidence in GenAI is associated with less critical thinking, while higher self-confidence is associated with more critical thinking.\u201d\nThe UW Social Futures Lab ( Ye 2024 ) asked philosophers to use models as critical thinking tools and found they were too neutral, incurious, and passive to be helpful.\nBusiness professor Michael Gerlich ( Gerlich 2025 ) interviewed 666 participants from a range of educational backgrounds and age groups, and found \u201ca significant negative correlation between frequent AI tool usage and critical thinking abilities, mediated by increased cognitive offloading. Younger participants exhibited higher dependence on AI tools and lower critical thinking scores compared to older participants.\u201d\n\nThese studies don\u2019t convince me that the problem is generative AI itself. They convince me the problem is that models are not trained to support critical thinking. And that the interface affordances and design decisions built into generic chatbots do not encourage or support critical thinking workflows and interactions. And that users have no clue they\u2019re supposed to be compensating for these weaknesses by becoming expert prompt engineers.\nThe foundation labs who control the models and default interfaces aren\u2019t prioritising this. At least as far as I can tell. They\u2019re focused on autonomous, agentic workflows like the recent \u201c Deep Research \u201d hype. Or developing \u201c reasoning models \u201d where the models themselves are trained to be the critical thinkers. This focuses entirely on getting models to think for you, rather than helping you become a better thinker.\nWhile I understand the economic incentives reward automating cognitive work more than augmenting human thinking, I\u2019d like to think we can have our grossly profitable automation cake and eat it too. The labs have enough resources to pursue both.\nI think we\u2019ve barely scratched the surface of AI as intellectual partner and tool for thought . Neither the prompts, nor the model, nor the current interfaces \u2013 generic or tailored \u2013 enable it well. This is rapidly becoming my central research obsession, particularly the interface design piece. It\u2019s a problem I need to work on in some form.\nWhen I read Candide in my freshman humanities course, Voltaire might have been challenging me to question na\u00efve optimism, but he wasn\u2019t able to respond to me in real time, prodding me to go deeper into why it\u2019s problematic, rethink my assumptions, or spawn dozens of research agents to read, synthesise, and contextualise everything written on Panglossian philosophy and Enlightenment ethics.\nIn fact, at eighteen, I didn\u2019t get Candide at all. It wasn\u2019t contextualised well by my professor or the curriculum, and the whole thing went right over my head. I lacked a tiny thinking partner in my pocket who could help me appreciate the text; a patient character to discuss, debate, and develop my own opinions with.\nThe ideals of the Enlightenment are still up for grabs here. We simply have to make the conscious choice to design our AI models and interfaces around them.\n \nPostScript I apologise for leaning heavily on Claude and Anthropic as examples throughout this piece. I\u2019m more familiar with Anthropic\u2019s research than other labs and Claude is my primary model, so I am heavily biased. I am sure other labs are doing interesting research in this area. I\u2019d love to read about it on BlueSky (either message or mention me) or your own blog.AI wrote none of the words in this piece, but Harsh Claude helped critique it during the draft stages and pointed out a number of weaknesses that I\u2019ve now addressed. I\u2019m thankful for it\u2019s critical collaboration." ], "output": "readability/", "pwd": "/data/archive/1755996284.038473", "schema": "ArchiveResult", "start_ts": "2025-08-24T00:45:11.463729+00:00", "status": "succeeded" } ], "screenshot": [], "singlefile": [], "title": [ { "cmd": [ "/usr/bin/curl", "--silent", "--location", "--compressed", "--proxy", "socks5://tor-socks-proxy:9150", "--max-time", "60", "--user-agent", "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)", "https://maggieappleton.com/ai-enlightenment" ], "cmd_version": "8.10.1", "end_ts": "2025-08-24T00:45:10.454304+00:00", "index_texts": null, "output": "A Treatise on AI Chatbots Undermining the Enlightenment", "pwd": "/data/archive/1755996284.038473", "schema": "ArchiveResult", "start_ts": "2025-08-24T00:45:10.343809+00:00", "status": "succeeded" } ], "wget": [] }, "icons": null, "is_archived": true, "is_static": false, "latest": { "archive_org": "TimeoutExpired: Command '['/usr/bin/curl', '--silent', '--location', '--compressed', '--proxy', 'socks5://tor-socks-proxy:9150', '--head', '--max-time', '60', '--user-agent', 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/128.0.0.0 Safari/537.36 ArchiveBox/{VERSION} (+https://github.com/ArchiveBox/ArchiveBox/)', 'https://web.archive.org/save/https://maggieappleton.com/ai-enlightenment']' timed out after 60 seconds", "dom": "output.html", "favicon": "favicon.ico", "git": null, "media": "media/", "pdf": null, "screenshot": null, "singlefile": null, "title": "A Treatise on AI Chatbots Undermining the Enlightenment", "warc": null, "wget": null }, "link_dir": "/data/archive/1755996284.038473", "newest_archive_date": "2025-08-24T00:45:44.055345+00:00", "num_failures": 1, "num_outputs": 8, "oldest_archive_date": "2025-08-24T00:44:49.953350+00:00", "path": "/ai-enlightenment", "schema": "Link", "scheme": "https", "snapshot_abid": "snp_01K3CRHS5K83695FE301ZYR976", "snapshot_id": "c0c23625-1c03-460f-968f-1862ffec24e6", "sources": [ "/data/sources/1755996282-import.txt" ], "tags": null, "tags_str": "", "timestamp": "1755996284.038473", "title": "A Treatise on AI Chatbots Undermining the Enlightenment", "url": "https://maggieappleton.com/ai-enlightenment" }