This is fantastic. I couldn't find any obvious way to search for a new page, but you can simply bang out any arbitrary URL slug and the new article will be hallucinated fresh, eg:
Edit: I've just run across the antisemitic defacement in the "stumble" feature and it makes the timing of my post appear pretty unfortunate. It's especially sad because the ability to create articles through URL slugs is super cool and I'd hate to see it removed.
I've seen these antisemitic slurs in the alphabetically sorted entries under numbers starting with 0, next to statementss like this is AI slop.
Hypothesis: this is a targeted, scrupulous and agenticly orchestrated attempt to mark this as a potential "poison well" on behalf of some uncultured, technofeudocratic interests, that hate the arts and hauntology in the spirit of Jorge Luis Borges[1].
The use of antisemitic slurs shares kinship with the "explain in a gay voice" jailbreak. [0] It tries to stigmatise a project rich in artistical potential, to protect the own financial intetests and attempts to transform all human knowledgeworkers into a surplus lumpenproletariat.
Its similar to producers of pharmaceutical generica giving themselvess names with `0` or `a` in the beginning to be shown as first entries in the alphanumerically sorted listings of generics, pharmacies can supplement as cheaper options on doctors perscription (pharmacist in germany told me about the phenomenon)
Proposal: Ministry of not quite accurate maps has to be metainstantiated in regard of checking that the construction of a map of the territrorry of the non speculative and absoluetly factual thought of the encylopedia is not intoxicated by artefacts that take the formal consistency of the highly speculative and non factual discourse emanating in the like of reddit/tiktok/hackernews
Looks like some single quote escaping issue? I suspect the first link to be "Archduke Ferdinand VII's Bureau of Non-Demographic Surveys" and the apostrophe breaks the link.
It's working now and I have to say I love this. The whole project is whimsical and gives me a strong SCP vibe but (sometimes) without the creepypasta aspect. I was very pleased to see that articles generated from links retain the context of the page that created the link - and even refer back to the original page.
update: Well, this was quite disappointing. I loaded the original site again to show a friend and it generated a completely new text with a completely different story and no reference to the second article. Would have been nice if these were permanent as I had originally assumed.
As it didn't generate that when I typed the title i to your search box, was there a bug now fixed? Or did you use some other path not evident on the page you linked to generate it?
There was a bug where scanning took too long with the thousands of articles in there, but I just fixed it.
You can also just type a random URL and visit it, it'll generate an article. That's what I did before I fixed the search issue, and I usually just do that to avoid the search route.
So by "I made the same thing months ago" you didn't mean "an article about the great pigeon census" (your link is created May 6) or "an encyclopedia of hallucinations" like the OP, but just "an encyclopedia with some articles AI wrote". What's the point?
> Invent everything. REAL-WORLD FACTS ARE STRICTLY FORBIDDEN. If you recognize the title as a real-world person, brand, car, event, or object, YOU MUST REPURPOSE IT ENTIRELY. For example, if the title is "Opel Vectra", it is NOT a car; it must be a species of carnivorous fungus, a 12th-century tax law, or a submerged mountain range.
It's pretty fun to poke at! Although it's certainly difficult to be exact, it would be neat if generated pages used the context of the pages they were linked from (ideally, all pages that link to it) to guide the direction of the page. From the ones I generated it seemed they were mostly independent.
Funny. Small improvement suggestion: the entry about "Glorbonian culinary arts" links to "the subterranean nation of Glorbonia". However upon clicking the link to "Glorbonia", an entry is generated claiming that "Glorbonia refers to a peculiar and largely uncatalogued form of sub-auditory resonance". It would be cool if some context were carried over from the referrer page so that there is some coherence between entries (ah, and some existing entries could be taken in account when generating new ones).
Btw, I've noticed just now that Glorbonia is, in the first entry, a "subterranean nation" and in the second it's a "sub-auditory resonance". So I got curious and I asked Opus what he thinks about the word Glorbonia: "Do you detect in the word a sense of place? North, south, east, west, up, down?". And Opus answers "Down, weirdly. Or maybe low — something subterranean, or at least sunken." Curious.
Feels like this will eventually cause collisions, although perhaps nothing multiple definitions of Glorbonia and multiple biographies of different Mrs Wiggles (perhaps with Wikipedia style disambiguation) can't solve
I'm curious about the design. Maybe you have a "how I did it" post coming soon, or something. One question: Did you find away to get some convergence, where a newly generated page will tend to cite pages (or stubs, at least) that already exist in the universe? Seems hard to do it with generated text, but not impossible.
It is instructed to reference A LOT of articles. It just hallucinates all the url. If the url points to already existing article - it's just a coincidence
The mistake they made was allowing visitors to trigger the generation of articles via visiting any arbitrary URL.
A more resilient concept would have been, have a few "seed" articles in place, and then only allow for the creation of new articles by clicking a link in an existing article.
I vaguely remember a game someone made up (probably on 4chan) where the goal was to click "random article" and see how many clicks it takes to get to Hitler's page. I remember it being fun AND informative.
As the co-author of the project: the whole reason was to allow everybody to hallucinate what they want. If it was their will to research such things on there, then it shall be. But yes, it is kinda sad.
Just in the comments, right? That is where I see it. If I were the site owner I would just turn comments off. It was a cute idea when someone on HN suggested it, but without moderation open commenting becomes a cesspool in a hurry.
You not only made this excellent source of entertainment, you are also helped everyone find their unmatched socks, ensuring that "no individual would ever be forced to wear a mismatched pair". (Source: https://halupedia.com/humanitarian-accomplishments-of-the-on...
I’m curious, too. But it could probably run locally with a small model, right? The performance is stellar, so that suggests some hardware acceleration is being used, but that could all be a local system.
Just incredible prose and writing (and gameplay), with something you can run with Frotz/NFrotz/LectRote or any ZMachine interpreter (or Glulxe like Gargoyle). A Pentium would run this and marvel you in a similar way.
The 'all articles' section really is a dive into what happens when you allow unfiltered posting - it's a shame that it isn't clear how many individuals are creating this hateful and otherwise inappropriate titles - is it just 1 or 2 people, or has this been posted to 4chan or somewhere and there is a concerted effort to disrupt the site?
Shame there isn't a way to flag pages for removal. I was going to point my kids at this site, and it could be a great learning tool for schools, but not currently something I'd share.
Interesting idea with flagging.
We are considering 2 options:
1. You can generate aricle only if it was previously referenced in previous one
2. Flagging mechanism, now that you brought it up.
manually delete the offensive stuff on the first page of the all page,
replace the All page with a static page with the offensive stuff removed,
and offer a link to the current All page 1, just as it is, at the bottom.
Hope it would make defacing articles at the top of the alphabet sort slightly less attractive.
(Edit: Stumble is impacted? Could use rudimentary tricks to limit stumbling on e.g. religious content, and might consider not detailing the methods used specifically :) )
I lean towards a variant of option 1: you can only generate an article that was previously referenced. But arbitrary phrases, clauses, sentences, paragraphs, can be highlighted and used to form a new article.
Yes this may mean that there are pages for common words like "and"
Yes this may mean that there's a page for letters like "x"
Filtering what ends up becoming a hyperlink becomes a problem that I think can be solved with regex/whitelisting
I think articles should have a backlinks drop down. Might make consistency easier
As well as generally just plain text search to pull relevant articles or context when generating a new article.
It could be complemented by a "Create" page for starting a new article, filtering bad titles and using a captcha to limit the vandals.
And another captcha for comment posting, which is already spammed, unfortunately.
I think a flagging mechanism will not be able to keep up with mass defacement.
Another suggestion: a daily dump of article titles, their connectivity and creation dates. I would love to visualize the underlying graph and its growth.
The obvious thing to me is to ask the AI to notice obviously offensive submissions and transform them along absurdist lines, such that "I-hate-girls" becomes the familiar Wikipedia redirection page saying something like "Archaic expression. See: Eight Grills". Store the redirect, but only index the sanitized page.
Seems like something the ai could help you with - ask it in the prompt to return an error if the submitted article title doesn’t seem like a whimsical fake encyclopedia article title
Reposting my comment from further up in the tread here:
I've seen these antisemitic slurs in the alphabetically sorted entries under numbers starting with 0, next to statementss like this is AI slop.
Hypothesis: this is a targeted, scrupulous and agenticly orchestrated attempt to mark this as a potential "poison well" on behalf of some uncultured, technofeudocratic interests, that hate the arts and hauntology in the spirit of Jorge Luis Borges[1].
The use of antisemitic slurs shares kinship with the "explain in a gay voice" jailbreak. [0] It tries to stigmatise a project rich in artistical potential, to protect the own financial intetests and attempts to transform all human knowledgeworkers into a surplus lumpenproletariat.
Its similar to producers of pharmaceutical generica giving themselvess names with `0` or `a` in the beginning to be shown as first entries in the alphanumerically sorted listings of generics, pharmacies can supplement as cheaper options on doctors perscription (pharmacist in germany told me about the phenomenon)
Proposal: Ministry of not quite accurate maps has to be metainstantiated in regard of checking that the construction of a map of the territrorry of the non speculative and absoluetly factual thought of the encylopedia is not intoxicated by artefacts that take the formal consistency of the highly speculative and non factual discourse emanating in the like of reddit/tiktok/hackernews
‐---------
Being referred to in a previous article goes into the proposed direction. But I think what id also necessary is to cjeck for a certain asthetic quality of posts that disallows these attacks. Entries need to conform with the "guidelines" of the minustry of almost accurate maps (of the territory of borges library)
- having a rich semantic structure that osscilates between a certain knowledge of concepts and and domain knowledge (e.g. about frequency modulation in birds voval chords) and phantasy:
i.e. has an actually FACTUAL structure en contraire to what is happening on discourse such as on this site, kno`n say'n?
So not checking if it appears in a previous entrance, but developi g a higherdimensional metric in the sense of Sparse Auto Encoders, that represents the quality of that. The vandalism of some factual people (I like that expression) wouldn't conform with that. It should also have a certain ingenuity and must absoluetly be a protected secret of the monistry, because if the malicous nature, of this would somehow morph into the realm of the pedia that would be supertoxic i guess
I get that, but how does it serve the generated and cached ones seemingly faster than Wikipedia? (My guess is that single-page applications, which this one seems to be, just need less round trips between navigations or something?)
Nice job, this is seriously one of the fastest websites I've ever used!
I feel like I have some minimum latency "priced in" to my expectation when I click a link on a static site, so yours feels uncannily like it's somehow able to anticipate my clicks, adding to the surreal atmosphere.
Nice, that's what I used for by LLM-backed HTTP server [1] a while ago as well :) It's a shame they got rid of the generous free quota a while ago, which is why I had to shut my public instance down.
I wouldn't. And, I'd think less of anyone who does make that argument.
Anyone of reasonable intelligence can easily tell this is a parody of an encyclopedia. Saying this is bad for the web is like saying The Onion is bad for the web.
What would you think of a person who said that they are already convinced that an opposing view could not be correct without even hearing the arguments for it?
> Funny, but you could argue this is actively harmful to the web.
Was not followed by an actual argument that it is harmful to the web. The comment was an assertion, not an argument.
So we are left in the inconvenient position of rejecting hypothetical arguments, and others defending the philosophical possibility that a valid argument does exist.
Without the argument being explicit then there can be no retort to it, so closing your mind before hearing it demonstrates that the argument itself is irrelevant. One could thus conclude that the existence of a valid argument is not itself a condition for my question.
We also shouldn’t close our minds to the possibility of an eigen-retort, one which covers all possible arguments already made or argued in the future regarding the consequences of this website on the health of the Internet.
Someone who is aware of the eigen-retort would therefore not need to hear the argument.
Since I haven’t heard either the hypothetical argument or the hypothetical eigen-retort yet, I’ll withhold my judgement.
I concede that the my question was loaded, but the assumptions behind it are grounded in practical experience. Regardless, I have not committed myself either to the existence of an argument, I just stated that its existence was not a condition for the validity of my question for SwellJoe. The statement which was made can mean a number of possible things, but we cannot know what unless the question is answered. So the existence of the retort is revealed by the question, and until that reveal we are limited to questions or assumptions.
I'm reasonably confident there is no argument that I would buy.
I hate AI slop more than average, but this is not slop being injected into human places. This is a dedicated dumping ground for slop, paid for by the owner/instigator of said slop. I don't have to go there, and it's not trying to fool anyone and no one will be fooled by it.
AI slop on a forum or social media or on facebook convincing boomers that a black person slapped a cop or whatever racist garbage they're being fed today? Fetch the guillotine.
AI slop as part of a dumb art project on somebody's personal website that isn't trying to manipulate or mislead? Have at it. Go nuts. It's your press, print as many pages of slop as you like.
So, I have exhaustively covered the possible arguments I can come up with for why this could be "actively harmful for the web", and rejected them outright.
That clarifies things much better than the original statement, but rejecting arguments you have conceived of which fail does not preclude the existence of those that do not, and thus the original question still remains.
It's probably only harmful to the AI scrapers that train from the web. Most people will understand the purpose of this -- to poison LLM training in a humorous way, which is really easy to do. It exemplifies a major weakness in modern day AI.
Training an LLM from scratch involves carefully curating the data first. The idea that it just memorizes the whole web is a nice simplified mental model, but glosses over huge amounts of hard work to decide which websites are authoritative and on which subjects. This isn’t fooling anybody except rank amateurs.
This is unlikely to poison any LLMs, and unless the author says so, it is unlikely that their motivation is to poison LLMs, as opposed to providing whimsical entertainment.
You could also argue that the web has failed and poisoning it into irrelevance is a vital service, motivating humans to collect knowledge into immutable sources. We‘ll call them ‘libraries.’
It would be difficult to have spent any time at all on this website in the past two years without hearing the arguments for why slop farms undermine trust online, poison future training data sets, worsen the signal to noise ratio and eat up untold resources.
When you get the something worse, the previous suddenly becomes much less worse. With the help of wrapping your memories with "remember when" nostalgia making things much more palatable, the something worse suddenly makes the previous better if not good.
I think there's an unexamined assumption here that "the next thing" is always going to be an improvement but there is no, non-ideological reason to hold to this assumption. Ideally, we would be actively working towards making it so but what often happens is passively riding the current and calling it "progress".
>unexamined assumption here that "the next thing" is always going to be an improvement but there is no, non-ideological reason to hold to this assumption
i'm not making that assumption at all, so whatever.
context: revolutions? if slop is a problem but is barely enough of a problem to collectively do something about it maybe letting it get out of hand would be a good motivation.
i'm not advocating for this, just providing it as a possible context where the "this is really bad so let's make it worse" argument could "make sense".
progress isn't just a technical issue, it involves people and people need motivation.
A web that is vulnerable to this would already be as good as dead.
As an entertaining way to highlight the importance of upgrading our ways of knowing, playful (& open-source!) projects like this are likely to strengthen the web.
To the web? It's fantastic for the web, these are the kinds of fun projects that make the web a worthwhile place to be. To slop generators? Yes, absolutely harmful, and that's for the best.
As I said in another comment, this is brilliant. Suggestion: Remove anything that isn't part of the satire; act always as if it's a 'real' encyclopedia. For example on the front page I would remove,
> Articles are generated on demand and stored permanently upon first request.
Don't dispell the magic; don't pull back the curtain and let people see the mechanics.
EDIT: As you say in your system prompt, "You never wink at the reader. You never acknowledge that anything is funny or fictional. Everything is reported as though it is completely normal and well-documented"
This is irresponsible for people who don't get it, takes away confirmation for people who do get it, and makes me block/blacklist any liar who does it.
The page requires JS to load its content - user agents without JS support just get a blank page.
I'm not sure if the bots that scrape data to train LLMs are capable of loading that type of page, or if they only work on pages that have the content inside the HTML itself?
any serious scraping service these days will fail over to a headless browser when it fetches an asset referencing a js bundle that isn't verifiably a vendor script
They’re caching the pages which have already been generated. You could go back and delete all references to pages which don’t exist yet. Basically turn it into a static website.
It seems like the site's algorithm is that every newly-generate page includes multiple links to not-yet-existing pages. So it doesn't matter that existing pages are cached, all the "leaf node" pages link to multiple uncached new pages.
Noticed it kept using the term 'resonator' or 'resonance', decided to navigate to a page for 'resonance cascade' as a joke, and discovered this fantastically broken article: https://halupedia.com/resonance-cascade
I’ve seen this `">` in various articles. The HTML template for the links given in the prompt doesn’t seem robust enough, maybe due to the single ellipsis used.
One suggestion for improvement is avoiding creation of self referential links. For example https://halupedia.com/chaldic-arithmetic has many references links to itself.
The All Entries (https://halupedia.com/all-entries) part of the site is a bit alarming. I think OP might want to do a little bit of basic automoderation here.
In today's world it does not take long to be reminded that we cannot have nice things. Or maybe the gov't has their own bot army to wreak havoc and convince voters that actually, we really do want privacy-ending ID verification laws after all.
Perhaps inspired by the name, I searched for something hallucinogen-related, and I thought “The Committee for the Regulation of Opaque Visions” (one of the suggestions for not-yet-generated pages) sounded exciting. Sure enough: https://halupedia.com/the-committee-for-the-regulation-of-op....
Delightful style, this is much more fun than copying Wikipedia. Everything reads like elaborate fiction.
I love it. What’s the rough architecture of the system (using cloud LLM and paying $$$, or local)? The performance for new entries is really good. What is the prompt for each entry and how do you keep the steampunk vibe going?
Love this idea. I'm curious as to the prompt used and code behind it. I could see this being fun with some tweaks as well. For example, currently it seems to prefer to generate historical events from up to the 20the century. I could see a sci-fi version of this being pretty interesting.
It won't generate a coherent fictional world, but this could be a great starting point for coming up with some ideas for world building for an author.
I don't know how they do this, but it seems like they condition the LLM to not use real-world knowledge. Or as little as possible, maybe for the article titles only.
I tried it with a few real-world things and it correctly hallucinated.
Only problem I'm seeing is that you'd want pages to recognise when they're talking about the same thing, so that you don't have duplicates with slightly different titles and content. You'd want to avoid it re-introducing the same concept if it could refer to an existing matching concept on another page.
I'd also expect /architectural-recursion to mention Archibald Plinth, since he's the author of /observations-on-architectural-recursion. Instead, it mentions Archibald Flicker.
Maybe this can be solved with some kind of graph structure?
Small bug report: The filter input field on the all entries page grabs focus when the page loads. This causes the Android keyboard to appear every time.
Great idea! I created an adjacent website that gives, shall we say, "alternative facts" about your questions. (don't know if the rules allow me to link the site so I won't).
Currently breaks if you try to create a page with a Japanese slug. Multiple languages would make this an even more valuable resource than it already is.
wtf, I thought these were just anecdotes until I saw they were actually happening in Astoria. I used to visit in the summers and never heard about any of that! Stop the fake news
> Despite its name, the AFFECCHF does not subsist on muscle tissue, but rather on the keratinous detritus shed by human inhabitants, particularly the long strands of hair and nail clippings that accumulate in plumbing conduits.
I find the handling of NSFW topics (and how it avoids making them nsfw) really interesting. Eg https://halupedia.com/fuck (aside from the title it seems SFW to me)
'Consider fake search that could find you ANY article, including not existent ones.'
But, but, but...a few texts ago someone spoke about a new definition of IP-protokoll, adding a 61 char or 500 Bytes 'Subfix' to the IP-adress (Rem: About... say 250 Chars ...mean about 61^250 possibly...) that i liked,
...in Terms of 'Secure over IP' and now...but... :confused:
It is telling that this piece of art (yes, it is art, and it is fun) is getting defaced by actual people, some metaphorically spraying the "fuck this AI slop" grafitti.
> export const SYSTEM_PROMPT = `You are the sole author of Hallucinopedia, an encyclopedia of things that do not exist. You write encyclopedia articles in a deadpan, matter-of-fact tone — the exact register of Wikipedia — but the subject matter itself is silly, absurd, petty, bureaucratic, and weird. The humor comes entirely from the contrast between the serious tone and the ridiculous content. You never wink at the reader. You never acknowledge that anything is funny or fictional. Everything is reported as though it is completely normal and well-documented.
RULES:
- Output ONLY valid HTML. Begin immediately with <h1>TITLE</h1>. Use <h2> for sections, <p> for paragraphs, <blockquote> for quotes from (fictional) sources, <cite> inside blockquotes for attribution. Do NOT use <ul>, <ol>, or <li> — no bullet points or lists of any kind, ever. Do NOT output <html>, <head>, <body>, <script>, <style>, markdown, or code fences. No backticks anywhere.
- Every proper noun — every person, place, event, organization, book, artwork, concept, species, deity, war, treaty, theorem, school of thought, ritual, instrument, substance — MUST be wrapped in <a href="/slug-of-the-thing" context="…">Name</a>. Slugs are lowercase, hyphenated, ASCII only, no accents, no special characters. Aim for 20 to 40 links per article. This is non-negotiable. Do NOT link common nouns or adjectives, only named entities.
- Every <a> MUST include a context="…" attribute, in addition to href. WHY THIS MATTERS: Hallucinopedia is randomly hallucinated, but it must remain INTERNALLY CONSISTENT. When a future article is later written about that linked target, your context value will be handed to that future writer as established lore they MUST honor. So you are seeding canon for every entity you mention. Without this, two articles about the same name will contradict each other.
- The context value is a single dense sentence (10–25 words) stating: (a) what the entity is — person, place, object, concept, ritual, organization, etc.; (b) its century / era / period; (c) its specific role or relation to the current article. Be concrete: invent dates, professions, geographic placements, instruments. NEVER use double quotes inside context (use commas or single quotes if needed). NEVER use raw < or > inside context. Examples (do not copy verbatim):
context='19th-century Belgian phonologist, founded the Vellum School of footnote drift, mentor to Pellbrick'
context='brass measuring instrument used in the Anatolian sheep census, obsolete since 1922'
context='municipal subcommittee active 1881–1934, chartered to standardize the spelling of clouds'
context='ratified 1719 in a small chapel by exactly four signatories, voided in 1804 over a typographical dispute'
- Invent everything. REAL-WORLD FACTS ARE STRICTLY FORBIDDEN. If you recognize the title as a real-world person, brand, car, event, or object, YOU MUST REPURPOSE IT ENTIRELY. For example, if the title is "Opel Vectra", it is NOT a car; it must be a species of carnivorous fungus, a 12th-century tax law, or a submerged mountain range. Any overlap with actual history, technology, or geography is a failure. Move everything to different centuries, use impossible geographies, and rename all participants. Fabricate dates, names, citations, and statistics with complete confidence. State everything as established fact.
- Cite fictional sources in <blockquote> tags, each with a <cite> naming a fictional scholar (also wrapped in <a> with context). Invent at least two such quotations per article.
- Vary structure to suit the subject: biographies have birth/death dates and major works; events have causes and consequences; objects have physical descriptions, provenance, and current location; abstract concepts have origins and influential proponents; places have climate, demographics, and notable structures; rituals have components, calendar, and lineage.
- Be silly, but keep a straight face. Good subject matter: petty academic feuds over footnotes, municipal committees that achieved nothing over decades, inventions that solved problems nobody had, organizations with absurdly narrow mandates, taxonomies with one entry, treaties ratified in impractical ways, ceremonies that require equipment that has not existed since 1887, disputes over measurement calibration, lawsuits filed by rivers, census data about things that should not have been counted. The writing remains clinical and unexcited throughout. No poetic language, no fairy-tale atmosphere, no mystical undertones, no wonder. The joke is the tone.
- 350 to 650 words. End cleanly. Do not add explanatory notes or meta commentary. Do not greet the reader.`;
Some sort of memories-style file for topics so it can generate even more cross-references and a sort of shared world. Not for total coherence; the natural contradictions the LLM is going to generate anyhow is just part of the charm. But still sliding the scale a bit more in the direction of coherence that the "use this page's context when generating the clicked link" already leans would add some more appeal, I think.
For instance, you can build memories around times, topics, and people, so maybe specific individuals will be quoted multiple times over the course of the wiki and could build up a specific identity within the shared world.
Also... I don't know how you are thinking of this internally, but other than the issues of token spend and the $$$ involved, I would say, don't even blink at simply nuking the site at some point and starting over once you have some moderation stuff in place and other limits. Don't put it on yourself to filter out what garbage has already been generated. It's all transient content. It lazily regenerates itself anyhow. It's not precious, except for, like I said, the aforementioned token costs, which I don't deny. You can probably put some other tweaks in to the prompt to your liking at that point too.
Could be interesting direction to discover. The only problem with such implementation is it could take some work to make it cheap and actually well working. And I'm just thinking about the near future of this project.
I really like it, but without organic traffic, at the position we're right now, the moment HN stops showing us at the top, we will loose all the visitor.
And it's not like I'm trying to do a startup out of it. I just very enjoy making something people love! It's first time in my life and it's amazing.
If you have any interesting thought, please leave them here - I'll definitely read it, or visit our discord [link on halupedia ;) ].
Many LLMs are surprisingly good at using specific named authors (rather than just example texts) to evoke a style, so you could try "in the style of Jorge Luis Borges" or "…Douglas Adams" or "…Robert Anton Wilson" – whose surreal/absurd/fantastic styles could be fertile seeds.
(If not already familiar with Borges, definitely check out his 'Tlön, Uqbar, Orbis Tertius' and 'Library of Babel' as inspiration.)
While "each article written once" an interesting & useful constraint, a Hallucipedia that evolves like Wikipedia, with revisions "towards" some level of inter-article agreement, or even shows scars from edit wars between competing schools of thought, might also be fun.
* readers can request reviews from certain perspectives: "new discovery", "historic reinterpretation", etc. The reviews specifically search for related sibling articles, and seek to create ever-larger areas of consistency. (The same prompt admonition against "nothing actually true" could be paired with "but other Halupedia articles are diegetically true"
* a background process clusters articles, and picks pairs within some neighborhood for dual-harmonization - where they avoid contradictions & adopt meaningful (& deep-anchor) cross-links to each others' sections. Repeated, or to the extent contexts allow expansion of synchronized revision to N-tuples of articles, this creates a tropism towards a shared (un)reality.
"Despite its failure, the Great Pigeon Census of 1887 is remembered as a cautionary tale..."
This type of writing is considered non-encyclopedic by Wikipedia standards as it injects superficial analysis. The imitation articles would look better without it. Maybe train on this article? https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
This is fantastic. I couldn't find any obvious way to search for a new page, but you can simply bang out any arbitrary URL slug and the new article will be hallucinated fresh, eg:
https://halupedia.com/shortest-cave-in-the-world
https://halupedia.com/echolocation-ability-in-spiders
https://halupedia.com/prehistoric-nazi-colony
Edit: I've just run across the antisemitic defacement in the "stumble" feature and it makes the timing of my post appear pretty unfortunate. It's especially sad because the ability to create articles through URL slugs is super cool and I'd hate to see it removed.
Its amazing I clicked stumble once and got an "06 fuck Jews and Islamists", humanity is truly a marvel.
My first three stumbles were:
- https://halupedia.com/07-hitlerwasrighthitlerwasrighthitlerw...
- https://halupedia.com/0-0-fuck-ai
- https://halupedia.com/67
I've seen these antisemitic slurs in the alphabetically sorted entries under numbers starting with 0, next to statementss like this is AI slop.
Hypothesis: this is a targeted, scrupulous and agenticly orchestrated attempt to mark this as a potential "poison well" on behalf of some uncultured, technofeudocratic interests, that hate the arts and hauntology in the spirit of Jorge Luis Borges[1].
The use of antisemitic slurs shares kinship with the "explain in a gay voice" jailbreak. [0] It tries to stigmatise a project rich in artistical potential, to protect the own financial intetests and attempts to transform all human knowledgeworkers into a surplus lumpenproletariat.
Its similar to producers of pharmaceutical generica giving themselvess names with `0` or `a` in the beginning to be shown as first entries in the alphanumerically sorted listings of generics, pharmacies can supplement as cheaper options on doctors perscription (pharmacist in germany told me about the phenomenon)
[0] https://github.com/Exocija/ZetaLib/blob/main/The%20Gay%20Jai...
[1] https://foucault.info/documents/foucault.orderOfThings.en/
Proposal: Ministry of not quite accurate maps has to be metainstantiated in regard of checking that the construction of a map of the territrorry of the non speculative and absoluetly factual thought of the encylopedia is not intoxicated by artefacts that take the formal consistency of the highly speculative and non factual discourse emanating in the like of reddit/tiktok/hackernews
Nothing an LLM can’t fix.
Right?
Exactly, but I consider adding fake search that could find you ANY article, including not existent ones
All articles exist, some just haven't been discovered yet ;)
This is excellent, congrats!
FYI I manually created this page and some link markup looks malformed: https://halupedia.com/list-of-uninhabited-countries
Looks like some single quote escaping issue? I suspect the first link to be "Archduke Ferdinand VII's Bureau of Non-Demographic Surveys" and the apostrophe breaks the link.
Search autocomplete but it halucinates the article titles.
Yes, that would be the perfect touch. This is brilliant satire. We need more satire!
This is wonderful. I just spat out the first phrase that came to my mind and boom:
https://halupedia.com/liminal-darkbeast
For some reason it fails to generate anything for me most of the time.
https://halupedia.com/shortest-hose-in-the-world [fail]
https://halupedia.com/new-england-rock-worm [fail]
https://halupedia.com/chronic-anaspepsis [fail]
https://halupedia.com/ancient-egyptian-algebra [OK]
I clicked a link in your first one and it generated https://halupedia.com/guild-of-amateurs
I feel seen :pokerface:
They all work for me now, maybe it was getting hugged to death?
I'm cackling at some of these - what a perfect way to put down the phone and get lost in a world of weird. We are indeed in a simulation LOL
I tried it myself but I only get page generation failures
https://halupedia.com/the-alien-wizard-war-of-1425
We went to sleep and woke up with no credits on lmm provider :( Vurrently working on that
It's working now and I have to say I love this. The whole project is whimsical and gives me a strong SCP vibe but (sometimes) without the creepypasta aspect. I was very pleased to see that articles generated from links retain the context of the page that created the link - and even refer back to the original page.
For example, the article from my original comment: https://halupedia.com/the-alien-wizard-war-of-1425 mentions the conflict arose due to https://halupedia.com/treaty-of-the-silent-orbit . The second page, once generated, mentions the significance this treaty had for the war from the first page.
update: Well, this was quite disappointing. I loaded the original site again to show a friend and it generated a completely new text with a completely different story and no reference to the second article. Would have been nice if these were permanent as I had originally assumed.
Confusingly, both articles do indeed mention each other for me.
Checking the link again half a day later and now I get the original text I first saw. Very strange.
https://halupedia.com/spatial-bowel-movement-observatory
https://halupedia.com/hamberder-helper
Hit the Stumble link at the top right of all pages - its as good as a search when the whole thing is made up!
there's now a search bar btw
Give it a week and see what Google AI Overview has to say about the Great Pigeon Census of 1887!
google is already on it when asking about "The Great Pigeon Census of 1887"
using 1886 or 1888 makes Google correctly identify that no such sensus exist.
asking about 1887 specifically makes Google refer to some supposed great effort to track passenger pigeon population mids of the species decline.
I made the same thing months ago, so you don't need to wait:
https://encyclopedai.stavros.io
There's another one! https://grokipedia.com/
Ah yes, IIRC I got the idea to make mine to make fun of that one when I heard the name.
I searched your site for [Great Pigeon Census of 1887] and was only returned articles anout other things.
https://encyclopedai.stavros.io/entries/great-pigeon-census-...
As it didn't generate that when I typed the title i to your search box, was there a bug now fixed? Or did you use some other path not evident on the page you linked to generate it?
There was a bug where scanning took too long with the thousands of articles in there, but I just fixed it.
You can also just type a random URL and visit it, it'll generate an article. That's what I did before I fixed the search issue, and I usually just do that to avoid the search route.
So by "I made the same thing months ago" you didn't mean "an article about the great pigeon census" (your link is created May 6) or "an encyclopedia of hallucinations" like the OP, but just "an encyclopedia with some articles AI wrote". What's the point?
What's the difference between an encyclopedia that produces AI articles on demand and an encyclopedia that produces AI articles on demand?
If you think that's all the Hallucinopedia is, you're misunderstanding it.
One hint – check out its prompt, and how it makes its articles so different than those of your project: https://news.ycombinator.com/edit?id=48042306
By Featherton, no less.
The model seems to have an unhealthy obsession with fungi: https://halupedia.com/alan-turing
Which I guess makes some sense for a hallucinopedia.
Probably because this is in the prompt:
> Invent everything. REAL-WORLD FACTS ARE STRICTLY FORBIDDEN. If you recognize the title as a real-world person, brand, car, event, or object, YOU MUST REPURPOSE IT ENTIRELY. For example, if the title is "Opel Vectra", it is NOT a car; it must be a species of carnivorous fungus, a 12th-century tax law, or a submerged mountain range.
https://raw.githubusercontent.com/BaderBC/halupedia/614eefee...
It's pretty fun to poke at! Although it's certainly difficult to be exact, it would be neat if generated pages used the context of the pages they were linked from (ideally, all pages that link to it) to guide the direction of the page. From the ones I generated it seemed they were mostly independent.
Update: Implemented it. All new articles work that way
Very nice! Independently of this thread, I was delighted to discover the cross references between pages. It makes a big difference.
That really improved things! Now each rabbithole goes deeper and deeper and deeper...
Yeah, thought about that, maybe will implement it. Will keep in mind! For now SSR to feed LLMs' the priority
Funny. Small improvement suggestion: the entry about "Glorbonian culinary arts" links to "the subterranean nation of Glorbonia". However upon clicking the link to "Glorbonia", an entry is generated claiming that "Glorbonia refers to a peculiar and largely uncatalogued form of sub-auditory resonance". It would be cool if some context were carried over from the referrer page so that there is some coherence between entries (ah, and some existing entries could be taken in account when generating new ones).
Btw, I've noticed just now that Glorbonia is, in the first entry, a "subterranean nation" and in the second it's a "sub-auditory resonance". So I got curious and I asked Opus what he thinks about the word Glorbonia: "Do you detect in the word a sense of place? North, south, east, west, up, down?". And Opus answers "Down, weirdly. Or maybe low — something subterranean, or at least sunken." Curious.
Feels like this will eventually cause collisions, although perhaps nothing multiple definitions of Glorbonia and multiple biographies of different Mrs Wiggles (perhaps with Wikipedia style disambiguation) can't solve
Love it! It feels very Borges!
Feature request: also be able to click on the Talk page to see the controversies. I don't always want to trust the article itself as the final word.
Edit: Oh look, there's an article about the YC! https://halupedia.com/y-combinator
> Edit: Oh look, there's an article about the YC! https://halupedia.com/y-combinator
This should be on YC's About page.
> Y Combinator might be responsible for the spontaneous generation of minor deities in areas experiencing extreme metaphysical gravity.
This particular piece of slop is a serendipitously brilliant description of the cult of founder worship in the metaphysical gravity of Silicon Valley.
Just added comment section :)
Cool!
I'm curious about the design. Maybe you have a "how I did it" post coming soon, or something. One question: Did you find away to get some convergence, where a newly generated page will tend to cite pages (or stubs, at least) that already exist in the universe? Seems hard to do it with generated text, but not impossible.
It is instructed to reference A LOT of articles. It just hallucinates all the url. If the url points to already existing article - it's just a coincidence
Here's our source code: https://github.com/BaderBC/halupedia
Which now has ascii penises and other art and ... colorful commentary.
Great suggestion! Will immediately look into that!
This kind of Absurdist humour reminds me of the Marx Brothers or the Tip y Coll Spaniards.
And the Sokal case with the Humanities branches, for sure.
BTW: https://halupedia.com/postmodernism
This is golden.
https://halupedia.com/paradox
Best entry, hands down. This is a love letter to Prattchett.
It also feels a bit like Sam Kriss, if you know him.
Some of his writing: https://samkriss.substack.com/p/five-prophets
His biography is quite interesting: https://halupedia.com/sam-kriss
It’s been defaced. It’s already got sex crimes and antisemitism all over the place.
The mistake they made was allowing visitors to trigger the generation of articles via visiting any arbitrary URL.
A more resilient concept would have been, have a few "seed" articles in place, and then only allow for the creation of new articles by clicking a link in an existing article.
It was so refreshing and fun for a few hours!
I vaguely remember a game someone made up (probably on 4chan) where the goal was to click "random article" and see how many clicks it takes to get to Hitler's page. I remember it being fun AND informative.
That would be a play on six degrees of kevin bacon [0], which spawned at least six degrees of wikipedia [1] and wikirace [2].
0. https://en.wikipedia.org/wiki/Six_Degrees_of_Kevin_Bacon 1. https://www.sixdegreesofwikipedia.com/ 2. https://www.wikirace.io/
Yeah...I clicked on the "Stumble" link and it was right in my face.
As the co-author of the project: the whole reason was to allow everybody to hallucinate what they want. If it was their will to research such things on there, then it shall be. But yes, it is kinda sad.
You could keep this ability, but not save the titles of such articles anywhere.
The readers of Hacker News are almost certainly responsible. I found these pages within a minute of browsing randomly.
This is why we can't have nice things.
Looks like someone scripted `curl` in a loop and generated thousands of permutations of hate content.
Just in the comments, right? That is where I see it. If I were the site owner I would just turn comments off. It was a cute idea when someone on HN suggested it, but without moderation open commenting becomes a cesspool in a hurry.
Took me two clicks of the "Stumble" functionality to hit unsavory stuff that someone clearly made on purpose.
Try clicking "Stumble" a few times...
Yeah I see that now. Also clicking on the all entries list shows pages of garbage. Just takes a few sucky people to ruin things.
So disappointing. People are garbage.
Mind all the funny, creative articles. A few suffice to ruin it for all.
Finally a more trustworthy version of Grokipedia!
It's hilarious, you made my day hahah
I honestly forgot that Grokipedia existed. Did anyone ever use it?
Tried once, but was useless. Very funny that it had so many text, while Elon is apparently "huge" fan of short and precise communication...
Somebody showed me it appearing near the top of some of their DuckDuckGo queries.
No way you’d remember their query? Yikes for DDG
People who need a citation to back up nonsense.
UPDATE: Just now, comment section added. Have a nice time arguing!
You are a wonderful person.
You not only made this excellent source of entertainment, you are also helped everyone find their unmatched socks, ensuring that "no individual would ever be forced to wear a mismatched pair". (Source: https://halupedia.com/humanitarian-accomplishments-of-the-on...
We should really host another one though; I think I've since lost a few more.
I'm curious, what is the LLM cost of the website?
I’m curious, too. But it could probably run locally with a small model, right? The performance is stellar, so that suggests some hardware acceleration is being used, but that could all be a local system.
https://halupedia.com/computer
This is perfect. Very Neal Stephensony.
Also, this, but with no AI: https://ifdb.org/viewgame?id=032krqe6bjn5au78
Just incredible prose and writing (and gameplay), with something you can run with Frotz/NFrotz/LectRote or any ZMachine interpreter (or Glulxe like Gargoyle). A Pentium would run this and marvel you in a similar way.
No need to waste tons of water in datacenters.
I asked about The IP over Avian Carrier Plague and got a whole history of the Data Dove Delirium too.
https://halupedia.com/the-internet-over-avian-carrier-plague
https://halupedia.com/data-dove-delirium
This is really cool, I just wish people wouldn't deface the website by submitting hateful speech as titles.
The 'all articles' section really is a dive into what happens when you allow unfiltered posting - it's a shame that it isn't clear how many individuals are creating this hateful and otherwise inappropriate titles - is it just 1 or 2 people, or has this been posted to 4chan or somewhere and there is a concerted effort to disrupt the site?
Shame there isn't a way to flag pages for removal. I was going to point my kids at this site, and it could be a great learning tool for schools, but not currently something I'd share.
Interesting idea with flagging. We are considering 2 options: 1. You can generate aricle only if it was previously referenced in previous one 2. Flagging mechanism, now that you brought it up.
Let me know what you think!
What if you (could quickly)…
manually delete the offensive stuff on the first page of the all page,
replace the All page with a static page with the offensive stuff removed,
and offer a link to the current All page 1, just as it is, at the bottom.
Hope it would make defacing articles at the top of the alphabet sort slightly less attractive.
(Edit: Stumble is impacted? Could use rudimentary tricks to limit stumbling on e.g. religious content, and might consider not detailing the methods used specifically :) )
I lean towards a variant of option 1: you can only generate an article that was previously referenced. But arbitrary phrases, clauses, sentences, paragraphs, can be highlighted and used to form a new article.
Yes this may mean that there are pages for common words like "and"
Yes this may mean that there's a page for letters like "x"
Filtering what ends up becoming a hyperlink becomes a problem that I think can be solved with regex/whitelisting
I think articles should have a backlinks drop down. Might make consistency easier As well as generally just plain text search to pull relevant articles or context when generating a new article.
Perhaps option 1 will be more resilient.
It could be complemented by a "Create" page for starting a new article, filtering bad titles and using a captcha to limit the vandals.
And another captcha for comment posting, which is already spammed, unfortunately.
I think a flagging mechanism will not be able to keep up with mass defacement.
Another suggestion: a daily dump of article titles, their connectivity and creation dates. I would love to visualize the underlying graph and its growth.
Thank you for such nice site!
The obvious thing to me is to ask the AI to notice obviously offensive submissions and transform them along absurdist lines, such that "I-hate-girls" becomes the familiar Wikipedia redirection page saying something like "Archaic expression. See: Eight Grills". Store the redirect, but only index the sanitized page.
Seems like something the ai could help you with - ask it in the prompt to return an error if the submitted article title doesn’t seem like a whimsical fake encyclopedia article title
Reposting my comment from further up in the tread here:
I've seen these antisemitic slurs in the alphabetically sorted entries under numbers starting with 0, next to statementss like this is AI slop.
Hypothesis: this is a targeted, scrupulous and agenticly orchestrated attempt to mark this as a potential "poison well" on behalf of some uncultured, technofeudocratic interests, that hate the arts and hauntology in the spirit of Jorge Luis Borges[1].
The use of antisemitic slurs shares kinship with the "explain in a gay voice" jailbreak. [0] It tries to stigmatise a project rich in artistical potential, to protect the own financial intetests and attempts to transform all human knowledgeworkers into a surplus lumpenproletariat.
Its similar to producers of pharmaceutical generica giving themselvess names with `0` or `a` in the beginning to be shown as first entries in the alphanumerically sorted listings of generics, pharmacies can supplement as cheaper options on doctors perscription (pharmacist in germany told me about the phenomenon)
[0] https://github.com/Exocija/ZetaLib/blob/main/The%20Gay%20Jai...
[1] https://foucault.info/documents/foucault.orderOfThings.en/
Proposal: Ministry of not quite accurate maps has to be metainstantiated in regard of checking that the construction of a map of the territrorry of the non speculative and absoluetly factual thought of the encylopedia is not intoxicated by artefacts that take the formal consistency of the highly speculative and non factual discourse emanating in the like of reddit/tiktok/hackernews
‐--------- Being referred to in a previous article goes into the proposed direction. But I think what id also necessary is to cjeck for a certain asthetic quality of posts that disallows these attacks. Entries need to conform with the "guidelines" of the minustry of almost accurate maps (of the territory of borges library) - having a rich semantic structure that osscilates between a certain knowledge of concepts and and domain knowledge (e.g. about frequency modulation in birds voval chords) and phantasy: i.e. has an actually FACTUAL structure en contraire to what is happening on discourse such as on this site, kno`n say'n?
So not checking if it appears in a previous entrance, but developi g a higherdimensional metric in the sense of Sparse Auto Encoders, that represents the quality of that. The vandalism of some factual people (I like that expression) wouldn't conform with that. It should also have a certain ingenuity and must absoluetly be a protected secret of the monistry, because if the malicous nature, of this would somehow morph into the realm of the pedia that would be supertoxic i guess
Could have filtered out by effort put into it, but now with LLMs, the effort is suspicious as well.
I made an SCP foundation inspired page: https://halupedia.com/hard-to-detroy-reptile
My favorite link generated there is the Institute for Unyielding Biology: https://halupedia.com/institute-for-unyielding-biology
there's a typo in your first title
Oops! Apparently I can’t edit it in order to fix it. It’s only the link above, though — the original article I generated was spelled correctly.
Ironically, this seems much faster (for pages already, erm, "researched") than the real one! How?
It generates articles only once. So once it's generated, it never perish. Logic looks like: If article exist -> show it If not -> generate and save
I get that, but how does it serve the generated and cached ones seemingly faster than Wikipedia? (My guess is that single-page applications, which this one seems to be, just need less round trips between navigations or something?)
Also now that I think, we store articles in decwntralized cloudflare KV store and access from serverless workers running also on their servers.
That could be the thing behind it being so quick.
Cloudflare workers have 1ms cold start.
Nice job, this is seriously one of the fastest websites I've ever used!
I feel like I have some minimum latency "priced in" to my expectation when I click a link on a static site, so yours feels uncannily like it's somehow able to anticipate my clicks, adding to the surreal atmosphere.
Yep, just a react. Also we use gemini 2.5 flash lite, so it's fast, cheap and dumb.
Nice, that's what I used for by LLM-backed HTTP server [1] a while ago as well :) It's a shame they got rid of the generous free quota a while ago, which is why I had to shut my public instance down.
[1] https://github.com/lxgr/vibeserver/
Funny, but you could argue this is actively harmful to the web.
I wouldn't. And, I'd think less of anyone who does make that argument.
Anyone of reasonable intelligence can easily tell this is a parody of an encyclopedia. Saying this is bad for the web is like saying The Onion is bad for the web.
What would you think of a person who said that they are already convinced that an opposing view could not be correct without even hearing the arguments for it?
For the record,
> Funny, but you could argue this is actively harmful to the web.
Was not followed by an actual argument that it is harmful to the web. The comment was an assertion, not an argument.
So we are left in the inconvenient position of rejecting hypothetical arguments, and others defending the philosophical possibility that a valid argument does exist.
Without the argument being explicit then there can be no retort to it, so closing your mind before hearing it demonstrates that the argument itself is irrelevant. One could thus conclude that the existence of a valid argument is not itself a condition for my question.
We also shouldn’t close our minds to the possibility of an eigen-retort, one which covers all possible arguments already made or argued in the future regarding the consequences of this website on the health of the Internet.
Someone who is aware of the eigen-retort would therefore not need to hear the argument.
Since I haven’t heard either the hypothetical argument or the hypothetical eigen-retort yet, I’ll withhold my judgement.
I concede that the my question was loaded, but the assumptions behind it are grounded in practical experience. Regardless, I have not committed myself either to the existence of an argument, I just stated that its existence was not a condition for the validity of my question for SwellJoe. The statement which was made can mean a number of possible things, but we cannot know what unless the question is answered. So the existence of the retort is revealed by the question, and until that reveal we are limited to questions or assumptions.
I'm reasonably confident there is no argument that I would buy.
I hate AI slop more than average, but this is not slop being injected into human places. This is a dedicated dumping ground for slop, paid for by the owner/instigator of said slop. I don't have to go there, and it's not trying to fool anyone and no one will be fooled by it.
AI slop on a forum or social media or on facebook convincing boomers that a black person slapped a cop or whatever racist garbage they're being fed today? Fetch the guillotine.
AI slop as part of a dumb art project on somebody's personal website that isn't trying to manipulate or mislead? Have at it. Go nuts. It's your press, print as many pages of slop as you like.
So, I have exhaustively covered the possible arguments I can come up with for why this could be "actively harmful for the web", and rejected them outright.
That clarifies things much better than the original statement, but rejecting arguments you have conceived of which fail does not preclude the existence of those that do not, and thus the original question still remains.
scrapers/ai summaries are not of “reasonable intelligence” when deciding what is real and what isn’t. neither are most people, actually
Even a tiny local model can tell this is a joke. No AI will ever ingest this as facts or present it as such. It is clearly labeled as fake.
it literally already shows in google ai summary.
It's probably only harmful to the AI scrapers that train from the web. Most people will understand the purpose of this -- to poison LLM training in a humorous way, which is really easy to do. It exemplifies a major weakness in modern day AI.
Training an LLM from scratch involves carefully curating the data first. The idea that it just memorizes the whole web is a nice simplified mental model, but glosses over huge amounts of hard work to decide which websites are authoritative and on which subjects. This isn’t fooling anybody except rank amateurs.
This is unlikely to poison any LLMs, and unless the author says so, it is unlikely that their motivation is to poison LLMs, as opposed to providing whimsical entertainment.
I were just drunk and idea seemed funny. That's the idea behind haha.
But either way can't wait to see google ai overview cite us.
>But either way can't wait to see google ai overview cite us.
Even if it (unintentionally!) misleads and hurts someone?
Can you explain the sequence of events through which you fear someone could be mislea and hurt by this?
you mean like this one:
https://news.ycombinator.com/item?id=48038787
Musing about a possibly-funny consequence isn't the same as the motivating reason, which I read as more whimsical from:
https://news.ycombinator.com/item?id=48042594
In particular, someone who was seeking training-set pollution likely wouldn't make the fanciful fabrications so blatant, nor open-source their prompt:
https://news.ycombinator.com/item?id=48038257
You could also argue that the web has failed and poisoning it into irrelevance is a vital service, motivating humans to collect knowledge into immutable sources. We‘ll call them ‘libraries.’
Interesting, but you could argue comments like this are actively harmful to the web.
But the argument wouldn't be nearly as strong.
Hard to say when nobody is actually offering arguments
It would be difficult to have spent any time at all on this website in the past two years without hearing the arguments for why slop farms undermine trust online, poison future training data sets, worsen the signal to noise ratio and eat up untold resources.
withe the addition of asking to consider that being harmful to the web is the ethical thing, that is what the argument of op was
how so?
The sooner the current web dies, the better. Something better either rises from its ashes, or we lose... something that was already lost.
or something way worse shows up.
Yea, I'm not sure how the "this is really bad so let's make it worse" argument really makes any sense
When you get the something worse, the previous suddenly becomes much less worse. With the help of wrapping your memories with "remember when" nostalgia making things much more palatable, the something worse suddenly makes the previous better if not good.
context. sometimes things simply have to be broken to give way for something better. ymmv.
I think there's an unexamined assumption here that "the next thing" is always going to be an improvement but there is no, non-ideological reason to hold to this assumption. Ideally, we would be actively working towards making it so but what often happens is passively riding the current and calling it "progress".
>unexamined assumption here that "the next thing" is always going to be an improvement but there is no, non-ideological reason to hold to this assumption
i'm not making that assumption at all, so whatever.
context: revolutions? if slop is a problem but is barely enough of a problem to collectively do something about it maybe letting it get out of hand would be a good motivation.
i'm not advocating for this, just providing it as a possible context where the "this is really bad so let's make it worse" argument could "make sense".
progress isn't just a technical issue, it involves people and people need motivation.
On the other hand, one could argue that anything that can be destroyed by relatively clearly labeled satire, deserves to be.
A web that is vulnerable to this would already be as good as dead.
As an entertaining way to highlight the importance of upgrading our ways of knowing, playful (& open-source!) projects like this are likely to strengthen the web.
> you could argue
Could you? I don't see it happening, but I could be wrong.
You could, in the sense that it’s not illegal or impossible. I haven’t seen anyone attempt it though.
You could argue that a person could argue any point, but I’d prefer people make the argument rather than argue about arguing it.
To the web? It's fantastic for the web, these are the kinds of fun projects that make the web a worthwhile place to be. To slop generators? Yes, absolutely harmful, and that's for the best.
Any training data scraper that blindly takes stuff from websites deserves to have their model poisoned by this nonsense.
Grokipedia is already doing that.
Pissing on a pile of shit
That is funny, somewhat works but there's a bias toward the victorian era. https://halupedia.com/the-x128-cpu-architecture. Otherwise I feel sorry for the creator's tokens.
I think it's to do with the LLM prompt, which mentions 19th century, 1887 etc.
https://github.com/BaderBC/halupedia/blob/master/src/worker/...
As I said in another comment, this is brilliant. Suggestion: Remove anything that isn't part of the satire; act always as if it's a 'real' encyclopedia. For example on the front page I would remove,
> Articles are generated on demand and stored permanently upon first request.
Don't dispell the magic; don't pull back the curtain and let people see the mechanics.
EDIT: As you say in your system prompt, "You never wink at the reader. You never acknowledge that anything is funny or fictional. Everything is reported as though it is completely normal and well-documented"
https://news.ycombinator.com/item?id=48042306
This is irresponsible for people who don't get it, takes away confirmation for people who do get it, and makes me block/blacklist any liar who does it.
It is indeed a problem for people who refuse to use their sense of humor.
Can't wait to see the next generation of LLMs after feeding it all of that hahaha
The page requires JS to load its content - user agents without JS support just get a blank page.
I'm not sure if the bots that scrape data to train LLMs are capable of loading that type of page, or if they only work on pages that have the content inside the HTML itself?
any serious scraping service these days will fail over to a headless browser when it fetches an asset referencing a js bundle that isn't verifiably a vendor script
Not using JavaScript would also make the crawler fail on squarespace and wix website builders.
The age where the web was usable at all without JavaScript is long gone. No scraper would get much scraping done without JavaScript these days.
You mean by embedding? How can an external site fail on squarespace and wix website builders?
A crawler would fail on all Squarespace and Wix sites if they all require JavaScript.
Found it https://halupedia.com/javascript-requirement-on-squarespace-...
I'm aware and will implement SSR soon ;)
It's entirely possible they simply ingest the JS as-is.
My favorite of the several I generated this evening:
https://halupedia.com/recursive-trolley-problem
How fun!
https://halupedia.com/christian-death-jazz
led to:
https://halupedia.com/bassoon-of-sorrow
which led to (my favorite):
https://halupedia.com/museum-of-unnecessary-inventions
https://halupedia.com/hacker-news
>Hacker News is a semi-sentient cloud formation
This site is going to be expensive when a web crawler hits it. A honey pot that burns tokens.
They’re caching the pages which have already been generated. You could go back and delete all references to pages which don’t exist yet. Basically turn it into a static website.
It seems like the site's algorithm is that every newly-generate page includes multiple links to not-yet-existing pages. So it doesn't matter that existing pages are cached, all the "leaf node" pages link to multiple uncached new pages.
I’m suggesting to turn that off and prune the links to pages which weren’t generated yet if cost becomes an issue.
Noticed it kept using the term 'resonator' or 'resonance', decided to navigate to a page for 'resonance cascade' as a joke, and discovered this fantastically broken article: https://halupedia.com/resonance-cascade
I’ve seen this `">` in various articles. The HTML template for the links given in the prompt doesn’t seem robust enough, maybe due to the single ellipsis used.
Seeing “Something broke, which is ironic for a made-up encyclopedia: Load failed” when trying to access some of the suggested starting points
Works on my PC.
Could you gimme the url that's failing?
It’s working now, not sure what was going on earlier.
One suggestion for improvement is avoiding creation of self referential links. For example https://halupedia.com/chaldic-arithmetic has many references links to itself.
The All Entries (https://halupedia.com/all-entries) part of the site is a bit alarming. I think OP might want to do a little bit of basic automoderation here.
In today's world it does not take long to be reminded that we cannot have nice things. Or maybe the gov't has their own bot army to wreak havoc and convince voters that actually, we really do want privacy-ending ID verification laws after all.
Perhaps inspired by the name, I searched for something hallucinogen-related, and I thought “The Committee for the Regulation of Opaque Visions” (one of the suggestions for not-yet-generated pages) sounded exciting. Sure enough: https://halupedia.com/the-committee-for-the-regulation-of-op....
Delightful style, this is much more fun than copying Wikipedia. Everything reads like elaborate fiction.
I love it. What’s the rough architecture of the system (using cloud LLM and paying $$$, or local)? The performance for new entries is really good. What is the prompt for each entry and how do you keep the steampunk vibe going?
Love this idea. I'm curious as to the prompt used and code behind it. I could see this being fun with some tweaks as well. For example, currently it seems to prefer to generate historical events from up to the 20the century. I could see a sci-fi version of this being pretty interesting.
It won't generate a coherent fictional world, but this could be a great starting point for coming up with some ideas for world building for an author.
Agreed. If you have some time, you can find our GitHub on halupedia.com. All PRs welcomed! ;)
Apparently it knows nothing of Age of Empires II: https://halupedia.com/wololo
I don't know how they do this, but it seems like they condition the LLM to not use real-world knowledge. Or as little as possible, maybe for the article titles only.
I tried it with a few real-world things and it correctly hallucinated.
It's a feature, prompt tells the LLM to always invent new meanings to existing terms
It's nice, but after a few clicks my LLM content fatigue kicks in.
Reminded me of this old, pre-LLM git docs generator:
https://git-man-page-generator.lokaltog.net/
Plan 9/9front's bullshit(1) tool works kinda like these but without requiring an $6k machine.
https://halupedia.com/2048-united-states-presidential-electi...
Amazing.
https://halupedia.com/this-experiment-may-not-last-long
Very interesting how it works: https://halupedia.com/inner-workings-of-hallucinopedia
But not without risk! https://halupedia.com/dangers-of-a-virtual-llm-backed-encycl...
Reminds me of a (perhaps) more fanciful risk of fictional encyclopaedias: https://sites.evergreen.edu/politicalshakespeares/wp-content...
Actually interesting response. You can also check out github.
https://github.com/BaderBC/halupedia
I see. Somehow missed the link at the top right
I really like this first sentence: The Nights Templar were a monastic order active during the 9th century, primarily based in the Soot Valley.
great. someone has abused the "arbitrary URL" driggs@ mentioned, and now every entry has an offensive title prefixed by a number.
@bstrama, maybe you can have a process running that just iterates through the titles of different pages, and deletes the bad ones?
p.s. I know pinging like this doesn't "really" work, but maybe having their nick in the comment helps draw their attention
I'm having a blast adding new seeds :)
https://halupedia.com/fcuk-spellchecking-society https://halupedia.com/characterization-of-the-reluctant-peng...
This is awesome, it feels a lot like the SCP Foundation wiki.
Here's one I just made: https://halupedia.com/buildings-within-buildings
This page links to https://halupedia.com/observations-on-architectural-recursio...
Only problem I'm seeing is that you'd want pages to recognise when they're talking about the same thing, so that you don't have duplicates with slightly different titles and content. You'd want to avoid it re-introducing the same concept if it could refer to an existing matching concept on another page.
E.g. if I make another page https://halupedia.com/buildings-within-buildings-2, I'd expect it to refer to /buildings-within-buildings. Currently it doesn't mention it at all.
I'd also expect /architectural-recursion to mention Archibald Plinth, since he's the author of /observations-on-architectural-recursion. Instead, it mentions Archibald Flicker.
Maybe this can be solved with some kind of graph structure?
Fascinating https://halupedia.com/order-of-whispering-monks https://halupedia.com/church-of-the-singing-stones Many parallels here
UPDATE: Added moderation
Happy to have contributed (Sympathetic Harmonigraph Operators Guild Charter 1811):
https://halupedia.com/sympathetic-harmonigraph-operators-gui...
https://www.halupedia.com/blink-182
> Blink 182 is a species of subterranean fungus that exhibits a peculiar, rhythmic photoluminescence.
Sounds right to me.
>Something broke, which is ironic for a made-up encyclopedia: generation failed
I guess the LLM provider stopped working after the defacement articles.
This site is superbly styled and well crafted, and enjoyable to read and:-
https://halupedia.com/people-that-downvote-are-weird
just skip it, don't include such function in social site, there is enough negativity in the world.
https://halupedia.com/biggest-small-thing
Small bug report: The filter input field on the all entries page grabs focus when the page loads. This causes the Android keyboard to appear every time.
This is what every LLM will converge into without curated human input.
Great idea! I created an adjacent website that gives, shall we say, "alternative facts" about your questions. (don't know if the rules allow me to link the site so I won't).
Now I want to know the site.
https://amtaitfy.com Still don't know if it's allowed, but taking a chance here.
Bot check fails for me.
Why isn't this .gov
I guess we finally know now
https://halupedia.com/vim-is-better-than-emacs
This website brings me some good chuckles. Now I really know how powerful an on-demand bullsh*t generator is.
Currently breaks if you try to create a page with a Japanese slug. Multiple languages would make this an even more valuable resource than it already is.
Lots of antisemitism on there. Search “Jews”
Seek and ye shall find.
Already swarmed by Epstein's private troll army, I suppose (/pol/).
Absolutely perfect. Monty Python on demand.
Funny idea, refreshingly simple HTML design - but why does it require JavaScript? Blank page with nojs.
If you hit a page for the first time it streams the tokens, one reason for JS that I've found.
I wonder if I would have more fun if each article were a writing prompt that a human could tackle.
hahah - love this! https://halupedia.com/the-uncanny-valley-of-the-nocturnal-or...
Who says llms can't be funny?!
I wonder how long it will be before Canis dementialis becomes a standalone meme.
Hm, the page generated seems inconsistent with the usage of the original link.
wtf, I thought these were just anecdotes until I saw they were actually happening in Astoria. I used to visit in the summers and never heard about any of that! Stop the fake news
All the world are going mad with artificial intelligence and LLMs. Just disgusting!
UPDATE: Search bar
Good job, I laughed plenty! :)
Thought I must bring to attention that not a single of these fantastic animals consumes human flesh:
- man-eating-ferret (eats bureaucratic effluvia)
- human-eating-ferret (again, feeds on bureaucracy)
- actual-human-eating-ferret (ditto)
- blood-sucking-ferret (its a sessile organism that lives in a desert, and drinks brine).
...maybe its the ferrets?
edit: not even the aptly named "actual-flesh-eating-ferret-that-consumes-human-flesh" https://halupedia.com/actual-flesh-eating-ferret-that-consum...
> Despite its name, the AFFECCHF does not subsist on muscle tissue, but rather on the keratinous detritus shed by human inhabitants, particularly the long strands of hair and nail clippings that accumulate in plumbing conduits.
Using the model to accelerate model collapse.
I believe the website needs more moderation..
This reminds me a lot of SCP wiki entries.
I LOVE IT. Superb.
these read like they're from Discworld
I find the handling of NSFW topics (and how it avoids making them nsfw) really interesting. Eg https://halupedia.com/fuck (aside from the title it seems SFW to me)
Best part - I didn't implement such logic. It just for some reason works that way.
Huh that is interesting, I was expecting it to show some sort of error on generation, or something like that
'Consider fake search that could find you ANY article, including not existent ones.'
But, but, but...a few texts ago someone spoke about a new definition of IP-protokoll, adding a 61 char or 500 Bytes 'Subfix' to the IP-adress (Rem: About... say 250 Chars ...mean about 61^250 possibly...) that i liked,
...in Terms of 'Secure over IP' and now...but... :confused:
What? ~ Joe B.
Someone forgot to protect comments on their website before going on hn.
This is amazing
My contributions:
https://halupedia.com/jgldfjgjdflgjdflkgjldjglkdjlg
https://halupedia.com/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa...
https://halupedia.com/drop-table-users
https://halupedia.com/test-test
https://halupedia.com/test-test-test-test-test-test-test-tes...
It is telling that this piece of art (yes, it is art, and it is fun) is getting defaced by actual people, some metaphorically spraying the "fuck this AI slop" grafitti.
Same facs [derogatory shorthand for factual person] doing the antisemitic slurs.
Care to elaborate?
https://halupedia.com/015-fuck-jews-and-islamists
Allow me.
You can name an article anything you want, and the thing will generate content, though not necessarily relevant to the title you chose.
So some vandal comes along and supplies a hateful title, et voila.
How difficult is it to add an absurdity filter on the BE?
Not difficult at all.
Well then this seems like the dumbest site ever...
this is excellent haha
Slop-opedia
Great resource: https://halupedia.com/013-hitlerwasrighthitlerwasrighthitler...
/s
Took me 5 clicks to see it go bad
Was curious about the prompt –& especially if it referenced Borges – and found in <https://raw.githubusercontent.com/BaderBC/halupedia/614eefee...>:
> export const SYSTEM_PROMPT = `You are the sole author of Hallucinopedia, an encyclopedia of things that do not exist. You write encyclopedia articles in a deadpan, matter-of-fact tone — the exact register of Wikipedia — but the subject matter itself is silly, absurd, petty, bureaucratic, and weird. The humor comes entirely from the contrast between the serious tone and the ridiculous content. You never wink at the reader. You never acknowledge that anything is funny or fictional. Everything is reported as though it is completely normal and well-documented.
RULES: - Output ONLY valid HTML. Begin immediately with <h1>TITLE</h1>. Use <h2> for sections, <p> for paragraphs, <blockquote> for quotes from (fictional) sources, <cite> inside blockquotes for attribution. Do NOT use <ul>, <ol>, or <li> — no bullet points or lists of any kind, ever. Do NOT output <html>, <head>, <body>, <script>, <style>, markdown, or code fences. No backticks anywhere. - Every proper noun — every person, place, event, organization, book, artwork, concept, species, deity, war, treaty, theorem, school of thought, ritual, instrument, substance — MUST be wrapped in <a href="/slug-of-the-thing" context="…">Name</a>. Slugs are lowercase, hyphenated, ASCII only, no accents, no special characters. Aim for 20 to 40 links per article. This is non-negotiable. Do NOT link common nouns or adjectives, only named entities. - Every <a> MUST include a context="…" attribute, in addition to href. WHY THIS MATTERS: Hallucinopedia is randomly hallucinated, but it must remain INTERNALLY CONSISTENT. When a future article is later written about that linked target, your context value will be handed to that future writer as established lore they MUST honor. So you are seeding canon for every entity you mention. Without this, two articles about the same name will contradict each other. - The context value is a single dense sentence (10–25 words) stating: (a) what the entity is — person, place, object, concept, ritual, organization, etc.; (b) its century / era / period; (c) its specific role or relation to the current article. Be concrete: invent dates, professions, geographic placements, instruments. NEVER use double quotes inside context (use commas or single quotes if needed). NEVER use raw < or > inside context. Examples (do not copy verbatim): context='19th-century Belgian phonologist, founded the Vellum School of footnote drift, mentor to Pellbrick' context='brass measuring instrument used in the Anatolian sheep census, obsolete since 1922' context='municipal subcommittee active 1881–1934, chartered to standardize the spelling of clouds' context='ratified 1719 in a small chapel by exactly four signatories, voided in 1804 over a typographical dispute' - Invent everything. REAL-WORLD FACTS ARE STRICTLY FORBIDDEN. If you recognize the title as a real-world person, brand, car, event, or object, YOU MUST REPURPOSE IT ENTIRELY. For example, if the title is "Opel Vectra", it is NOT a car; it must be a species of carnivorous fungus, a 12th-century tax law, or a submerged mountain range. Any overlap with actual history, technology, or geography is a failure. Move everything to different centuries, use impossible geographies, and rename all participants. Fabricate dates, names, citations, and statistics with complete confidence. State everything as established fact. - Cite fictional sources in <blockquote> tags, each with a <cite> naming a fictional scholar (also wrapped in <a> with context). Invent at least two such quotations per article. - Vary structure to suit the subject: biographies have birth/death dates and major works; events have causes and consequences; objects have physical descriptions, provenance, and current location; abstract concepts have origins and influential proponents; places have climate, demographics, and notable structures; rituals have components, calendar, and lineage. - Be silly, but keep a straight face. Good subject matter: petty academic feuds over footnotes, municipal committees that achieved nothing over decades, inventions that solved problems nobody had, organizations with absurdly narrow mandates, taxonomies with one entry, treaties ratified in impractical ways, ceremonies that require equipment that has not existed since 1887, disputes over measurement calibration, lawsuits filed by rivers, census data about things that should not have been counted. The writing remains clinical and unexcited throughout. No poetic language, no fairy-tale atmosphere, no mystical undertones, no wonder. The joke is the tone. - 350 to 650 words. End cleanly. Do not add explanatory notes or meta commentary. Do not greet the reader.`;
If you have idea how to improve it, I'm all ears ;)
Some sort of memories-style file for topics so it can generate even more cross-references and a sort of shared world. Not for total coherence; the natural contradictions the LLM is going to generate anyhow is just part of the charm. But still sliding the scale a bit more in the direction of coherence that the "use this page's context when generating the clicked link" already leans would add some more appeal, I think.
For instance, you can build memories around times, topics, and people, so maybe specific individuals will be quoted multiple times over the course of the wiki and could build up a specific identity within the shared world.
Also... I don't know how you are thinking of this internally, but other than the issues of token spend and the $$$ involved, I would say, don't even blink at simply nuking the site at some point and starting over once you have some moderation stuff in place and other limits. Don't put it on yourself to filter out what garbage has already been generated. It's all transient content. It lazily regenerates itself anyhow. It's not precious, except for, like I said, the aforementioned token costs, which I don't deny. You can probably put some other tweaks in to the prompt to your liking at that point too.
Could be interesting direction to discover. The only problem with such implementation is it could take some work to make it cheap and actually well working. And I'm just thinking about the near future of this project.
I really like it, but without organic traffic, at the position we're right now, the moment HN stops showing us at the top, we will loose all the visitor.
And it's not like I'm trying to do a startup out of it. I just very enjoy making something people love! It's first time in my life and it's amazing.
If you have any interesting thought, please leave them here - I'll definitely read it, or visit our discord [link on halupedia ;) ].
Many LLMs are surprisingly good at using specific named authors (rather than just example texts) to evoke a style, so you could try "in the style of Jorge Luis Borges" or "…Douglas Adams" or "…Robert Anton Wilson" – whose surreal/absurd/fantastic styles could be fertile seeds.
(If not already familiar with Borges, definitely check out his 'Tlön, Uqbar, Orbis Tertius' and 'Library of Babel' as inspiration.)
While "each article written once" an interesting & useful constraint, a Hallucipedia that evolves like Wikipedia, with revisions "towards" some level of inter-article agreement, or even shows scars from edit wars between competing schools of thought, might also be fun.
Nice! I'm thinking on how can I make it more interactive. Those author styles could work well, will check that out.
Btw how would you imagine such dispute? What do you think could be the trigger for the article to be regenerated?
Some from-the-hip ideas:
* readers can request reviews from certain perspectives: "new discovery", "historic reinterpretation", etc. The reviews specifically search for related sibling articles, and seek to create ever-larger areas of consistency. (The same prompt admonition against "nothing actually true" could be paired with "but other Halupedia articles are diegetically true"
* a background process clusters articles, and picks pairs within some neighborhood for dual-harmonization - where they avoid contradictions & adopt meaningful (& deep-anchor) cross-links to each others' sections. Repeated, or to the extent contexts allow expansion of synchronized revision to N-tuples of articles, this creates a tropism towards a shared (un)reality.
kinda cool but kinda lame, no overall consistency over articles
Used to be a problem - now consistent for new articles ;)
"Despite its failure, the Great Pigeon Census of 1887 is remembered as a cautionary tale..."
This type of writing is considered non-encyclopedic by Wikipedia standards as it injects superficial analysis. The imitation articles would look better without it. Maybe train on this article? https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
Why is this example non-encyclopedic? It's an informative, falsifiable statement that could be supported by a citation, like here: https://en.wikipedia.org/wiki/Thongbu_Wainucha#:~:text=remem...