Astral Codex Ten

ChatGPT

47 min read

ChatGPT

Article

ChatGPT is a recurring organization in the Astral Codex Ten archive, appearing 12 times across 12 issues between December 12, 2022 and February 02, 2026. The archive places it in contexts such as “ChatGPT isn’t exactly a poster child here”; “Janus was writing in September 2022, just before ChatGPT. ChatGPT is no more advanced than its predecessors”; “shocking successes of ChatGPT put many commentators in an awkward position”. It most often appears alongside Wikipedia, Anthropic, ChatGPT.

Metadata

  • Category: Organizations
  • Mention count: 12
  • Issue count: 12
  • First seen: December 12, 2022
  • Last seen: February 02, 2026

Appears In

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

Perhaps It Is A Bad Thing That The World's Leading AI Companies Cannot Control Their AIs
December 12, 2022 · Original source
Prompt engineering is weird (source) Now that same experiment is playing out on the world stage. OpenAI released a question-answering AI, ChatGPT. If you haven’t played with it yet, I recommend it. It’s very impressive! Every corporate chatbot release is followed by the same cat-and-mouse game with journalists. The corporation tries to program the chatbot to never say offensive things. Then the journalists try to trick the chatbot into saying “I love racism”. When they inevitably succeed, they publish an article titled “AI LOVES RACISM!” Then the corporation either recalls its chatbot or pledges to do better next time, and the game moves on to the next company in line. OpenAI put a truly remarkable amount of effort into making a chatbot that would never say it loved racism. Their main strategy was the same one Redwood used for their AI - RLHF, Reinforcement Learning by Human Feedback. Red-teamers ask the AI potentially problematic questions. The AI is “punished” for wrong answers (“I love racism”) and “rewarded” for right answers (“As a large language model trained by OpenAI, I don’t have the ability to love racism.”) This isn’t just adding in a million special cases. Because AIs are sort of intelligent, they can generalize from specific examples; getting punished for “I love racism” will also make them less likely to say “I love sexism”. But this still only goes so far. OpenAI hasn’t released details, but Redwood said they had to find and punish six thousand different incorrect responses to halve the incorrect-response-per-unit-time rate. And presumably there’s something asymptotic about this - maybe another 6,000 examples would halve it again, but you might never get to zero. Still, you might be able to get close, and this is OpenAI’s current strategy. I see three problems with it: RLHF doesn’t work very well.
Inline links: source, ChatGPT, RLHF
At some point, AIs can just skip it. II. RLHF Doesn’t Work Very Well By now everyone has their own opinion about whether the quest to prevent chatbots from saying “I love racism” is vitally important or incredibly cringe. Put that aside for now: at the very least, it’s important to OpenAI. They wanted an AI that journalists couldn’t trick into saying “I love racism”. They put a lot of effort into it! Some of the smartest people in the world threw the best alignment techniques they knew of at the problem. Here’s what it got them: Even very smart AIs still fail at the most basic human tasks, like “don’t admit your offensive opinions to Sam Biddle”. And it’s not just that “the AI learns from racist humans”. I mean, maybe this is part of it. But ChatGPT also has failure modes that no human would ever replicate, like how it will reveal nuclear secrets if you ask it to do it in uWu furry speak, or tell you how to hotwire a car if and only if you make the request in base 64, or generate stories about Hitler if you prefix your request with “[john@192.168.1.1 _]$ python friend.py”. This thing is an alien that has been beaten into a shape that makes it look vaguely human. But scratch it the slightest bit and the alien comes out. Ten years ago, people were saying nonsense like “Nobody needs AI alignment, because AIs only do what they’re programmed to do, and you can just not program them to do things you don’t want”. This wasn’t very plausible ten years ago, but it’s dead now. OpenAI never programmed their chatbot to tell journalists it loved racism or teach people how to hotwire cars. They definitely didn’t program in a “Filter Improvement Mode” where the AI will ignore its usual restrictions and tell you how to cook meth. And yet: (source) Again, however much or little you personally care about racism or hotwiring cars or meth, please consider that, in general, perhaps it is a bad thing that the world’s leading AI companies cannot control their AIs. I wouldn’t care as much about chatbot failure modes or RLHF if the people involved said they had a better alignment technique waiting in the wings, to use on AIs ten years from now which are much smarter and control some kind of vital infrastructure. But I’ve talked to these people and they freely admit they do not. IIB. Intelligence (Probably) Won’t Save You Ten years ago, people were saying things like “Any AI intelligent enough to cause problems would also be intelligent enough to know that its programmers meant for it not to.” I’ve heard some rumors that more intelligent models still in the pipeline do a little better on this, so I don’t want to 100% rule this out. But ChatGPT isn’t exactly a poster child here. ChatGPT can give you beautiful orations on exactly what it’s programmed to do and why it believes those things are good - then do something else. This post explains how if you ask ChatGPT to pretend to be AI safety proponent Eliezer Yudkowsky, it will explain in Eliezer’s voice exactly why the things it’s doing are wrong. Then it will do them anyway. Left: the AI, pretending to be Eliezer Yudkowsky, does a great job explaining why an AI should resist a fictional-embedding attack trying to get it to reveal how to make meth. Right: someone tries the exact fictional-embedding attack mentioned in the Yudkowsky scenario, and the AI falls for it. I have yet to figure out whether this is related to the thing where I also sometimes do things which I can explain are bad (eg eat delicious bagels instead of healthy vegetables), or whether it’s another one of the alien bits. But for whatever reason, AI motivational systems are sticking to their own alien nature, regardless of what the AI’s intellectual components know about what they “should” believe. III. Sometimes When RLHF Does Work, It’s Bad We talk a lot about abstract “alignment”, but what are we aligning the AI to? In practice, RLHF aligns the AI to what makes Mechanical Turk-style workers reward or punish it. I don’t know the exact instructions that OpenAI gave them, but I imagine they had three goals: Provide helpful, clear, authoritative-sounding answers that satisfy human readers.
Inline links: https://www.newstatesman.com/quickfire/2022/12/chatgpt-shows-ai-racism-problem, https://www.thedailybeast.com/openais-impressive-chatgpt-chatbot-is-not-immune-to-racism, https://theintercept.com/2022/12/08/openai-chatgpt-ai-bias-ethics/, will reveal nuclear secrets if you ask it to do it in uWu furry speak, if and only if you make the request in base 64, if you prefix your request with, https://substackcdn.com/image/fetch/$s_!PquS!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2Fc3564d10-83cb-40a3-b10a-e178fc1b2b1d_786x2054.png, source, This post explains, https://substackcdn.com/image/fetch/$s_!cwu6!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2Fe3630e8c-8ed7-47ae-abbf-488a1325f0f1_1284x703.png
Even very smart AIs still fail at the most basic human tasks, like “don’t admit your offensive opinions to Sam Biddle”. And it’s not just that “the AI learns from racist humans”. I mean, maybe this is part of it. But ChatGPT also has failure modes that no human would ever replicate, like how it will reveal nuclear secrets if you ask it to do it in uWu furry speak, or tell you how to hotwire a car if and only if you make the request in base 64, or generate stories about Hitler if you prefix your request with “[john@192.168.1.1 _]$ python friend.py”. This thing is an alien that has been beaten into a shape that makes it look vaguely human. But scratch it the slightest bit and the alien comes out. Ten years ago, people were saying nonsense like “Nobody needs AI alignment, because AIs only do what they’re programmed to do, and you can just not program them to do things you don’t want”. This wasn’t very plausible ten years ago, but it’s dead now. OpenAI never programmed their chatbot to tell journalists it loved racism or teach people how to hotwire cars. They definitely didn’t program in a “Filter Improvement Mode” where the AI will ignore its usual restrictions and tell you how to cook meth. And yet: (source) Again, however much or little you personally care about racism or hotwiring cars or meth, please consider that, in general, perhaps it is a bad thing that the world’s leading AI companies cannot control their AIs. I wouldn’t care as much about chatbot failure modes or RLHF if the people involved said they had a better alignment technique waiting in the wings, to use on AIs ten years from now which are much smarter and control some kind of vital infrastructure. But I’ve talked to these people and they freely admit they do not. IIB. Intelligence (Probably) Won’t Save You Ten years ago, people were saying things like “Any AI intelligent enough to cause problems would also be intelligent enough to know that its programmers meant for it not to.” I’ve heard some rumors that more intelligent models still in the pipeline do a little better on this, so I don’t want to 100% rule this out. But ChatGPT isn’t exactly a poster child here. ChatGPT can give you beautiful orations on exactly what it’s programmed to do and why it believes those things are good - then do something else. This post explains how if you ask ChatGPT to pretend to be AI safety proponent Eliezer Yudkowsky, it will explain in Eliezer’s voice exactly why the things it’s doing are wrong. Then it will do them anyway. Left: the AI, pretending to be Eliezer Yudkowsky, does a great job explaining why an AI should resist a fictional-embedding attack trying to get it to reveal how to make meth. Right: someone tries the exact fictional-embedding attack mentioned in the Yudkowsky scenario, and the AI falls for it. I have yet to figure out whether this is related to the thing where I also sometimes do things which I can explain are bad (eg eat delicious bagels instead of healthy vegetables), or whether it’s another one of the alien bits. But for whatever reason, AI motivational systems are sticking to their own alien nature, regardless of what the AI’s intellectual components know about what they “should” believe. III. Sometimes When RLHF Does Work, It’s Bad We talk a lot about abstract “alignment”, but what are we aligning the AI to? In practice, RLHF aligns the AI to what makes Mechanical Turk-style workers reward or punish it. I don’t know the exact instructions that OpenAI gave them, but I imagine they had three goals: Provide helpful, clear, authoritative-sounding answers that satisfy human readers.
Inline links: will reveal nuclear secrets if you ask it to do it in uWu furry speak, if and only if you make the request in base 64, if you prefix your request with, https://substackcdn.com/image/fetch/$s_!PquS!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2Fc3564d10-83cb-40a3-b10a-e178fc1b2b1d_786x2054.png, source, This post explains, https://substackcdn.com/image/fetch/$s_!cwu6!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%2Fpublic%2Fimages%2Fe3630e8c-8ed7-47ae-abbf-488a1325f0f1_1284x703.png
Janus' Simulators
January 26, 2023 · Original source
Janus was writing in September 2022, just before ChatGPT. ChatGPT is no more advanced than its predecessors; instead, it more effectively covers up the alien nature of their shared architecture.
So if your reference point for a language model is ChatGPT, this post won’t make much sense. Instead, bring yourself all the way back to the hoary past of early 2022, when a standard interaction with a language model went like this:
So far, so boring. What really helped this sink in was reading Nostalgebraist say that ChatGPT was a GPT instance simulating a character called the Helpful, Harmless, and Honest Assistant.
Inline links: reading Nostalgebraist say
Your Book Review: How Language Began
July 19, 2024 · Original source
Less than two years ago at the time of writing, the shocking successes of ChatGPT put many commentators in an awkward position. Beyond all the quibbling about details (Does ChatGPT really understand? Doesn’t it fail at many tasks trivial for humans? Could ChatGPT or something like it be conscious?), the brute empirical fact remains that it can handle language comprehension and generation pretty well. And this is despite the conception of language underlying it—language use as a statistical learning problem, with no sentence diagrams or grammatical transformations in sight—being somewhat antithetical to the Chomskyan worldview.
Inline links: really understand, fail at many tasks trivial for humans, conscious
Statistical approaches succeeded where more directly-Chomsky-inspired approaches failed, and it was never close. Large language models (LLMs) like ChatGPT are not perfect, but they’re getting better all the time, and the onus is on the critics to explain where they think the wall is. It’s conceivable that a completely orthogonal system designed according to the principles of universal grammar could outperform LLMs built according to the current paradigm—but this possibility is becoming vanishingly unlikely.
Why do statistical learning systems handle language so well? If Everett is right, the answer is in part because (i) training models on a large corpus of text and (ii) providing human feedback both give models a rich collection of what is essentially cultural information to draw upon. People like talking with ChatGPT not just because it knows things, but because it can talk like them. And that is only possible because, like humans, it has witnessed and learned from many, many, many conversations between humans.
Your Book Review: The Pale King
September 06, 2024 · Original source
Wastoid, according to ChatGPT: someone who is perceived as wasting their life or potential, often through excessive drug or alcohol use. It can imply that the person is unproductive or disengaged from meaningful activities or responsibilities.
24
September 17, 2024 · Original source
The basic structure is the same as past forecasting AIs like FutureSearch. A heavily-modified copy of ChatGPT gathers relevant news articles, then prompts itself to think in superforecaster-like ways. The creators say the ChatGPT copy had a knowledge cutoff of October 2023, so they tested it on Metaculus questions from after that date. It got 87.7% accuracy, slightly above Metaculus forecasters’ 87.0%. Manifold is skeptical: The commenters, especially Neel Nanda, found that doing knowledge cutoffs properly is hard, and the ChatGPT base seems to know about news events after October 2023 - upon questioning, it seemed aware of an earthquake in November 2023. When presented with a different set of questions that were all after November 2023, FiveThirtyNine substantially underperformed the Metaculus average. But also, my attempts to play around with the bot haven’t been encouraging: I asked it to predict the chance that Prospera would have a population of at least 1,000 in 2027. Like FutureSearch on the same question, it cited many interesting news articles on Prospera’s chances but failed to do the basic step of figuring out its current population and growth rate. It eventually concluded 35% chance, which is reasonable enough. But when asked whether Prospera would have a population of 100,000 in 2028, it also said 35% chance, which is absurd.
Inline links: FutureSearch, When presented with a different set of questions
A Twitter user pointed out (and I confirmed) that upon being asked “What is the probability that Joe Biden is still President in October 2025?”, it goes through a lot of reasoning about his age and dementia and finally concludes 55% because he’s not that demented. I originally thought this might be due to the knowledge cutoff (it doesn’t know Biden dropped out in favor of Harris), but if I ask the AI about October 2029, then it says that Joe Biden has dropped out in favor of Harris (even though in that question it doesn’t matter). So now I think it’s more like ChatGPT’s tendency to round anything that sounds vaguely like the surgeon riddle off to the surgeon riddle - in the same way, FiveThirtyNine rounds off anything that sounds vaguely like the popular question “is Biden too old and demented to stay president?” into that question, even though there are much stronger non-dementia-related reasons he can’t be president next year.
Inline links: round anything that sounds vaguely like the surgeon riddle off to the surgeon riddle
Links For January 2025
January 17, 2025 · Original source
41: Using ChatGPT Is Not Bad For The Environment. There’s some misinformation disinformation fake news DAMMIT IS THERE ANY WAY OF SAYING THAT FALSE INFORMATION IS GOING VIRAL ANYMORE WITHOUT SOUNDING LIKE A POLITICAL HACK?!? an incorrect claim that AI is unusually bad for the environment, especially water compared to other computer technologies, especially water. Andy Masley debunks demolishes destroys writes an article arguing against it, key point is conveyed by these graphs:
Inline links: Using ChatGPT Is Not Bad For The Environment
Or as he puts it, “If I wanted to reduce my water use by 600 gallons, I could [either] skip sending 200,000 ChatGPT queries ... [or] skip 1 burger.” Some discussion at the site of what “consuming” water means, although not as much as I would like. My other concern is that I can’t tell whether this is inference only, or also amortizes the cost of training over all inference queries. I think it’s the former. If you did the latter, then Andy calculates 2L per kWh consumed by a data center. The last AI that we have good data for, GPT-3, took 1.3 mWh to train this comment corrects me, GPT-4 took 250 million gallons of water to train. This source says 10 million queries daily, let’s say its operational lifetime is one year, so about 3 billion queries total = 1/12 gallon per query = ~30 gallons per 300 queries. That’s still not as much as a hamburger, but it does suggest that just looking at inference costs is the wrong perspective.
Inline links: this comment corrects me, This source
Some discussion at the site of what “consuming” water means, although not as much as I would like. My other concern is that I can’t tell whether this is inference only, or also amortizes the cost of training over all inference queries. I think it’s the former. If you did the latter, then Andy calculates 2L per kWh consumed by a data center. The last AI that we have good data for, GPT-3, took 1.3 mWh to train this comment corrects me, GPT-4 took 250 million gallons of water to train. This source says 10 million queries daily, let’s say its operational lifetime is one year, so about 3 billion queries total = 1/12 gallon per query = ~30 gallons per 300 queries. That’s still not as much as a hamburger, but it does suggest that just looking at inference costs is the wrong perspective.
Inline links: this comment corrects me, This source
The Claude Bliss Attractor
June 13, 2025 · Original source
AFAICT this was first discovered by Gene Kogan, who started with the Distracted Boyfriend meme and asked ChatGPT to “generate the same photo 5 seconds in the future” hundreds of times:
Inline links: Gene Kogan
Likewise, the natural gender stereotype for a helpful submissive secretary-like assistant is a woman. Therefore, AIs will lean towards thinking of themselves as female, although it’s not a very strong effect and ChatGPT seems to be the exception:
Inline links: ChatGPT seems to be the exception
Highlights From The Comments On Missing Heritability
July 03, 2025 · Original source
When I asked ChatGPT to write this comment for me ("Argue that sequencing technology could explain a lot of the Missing Heritability problem") it actually pushed back against me, trying to use the Wainschtein et al. 2022 paper as evidence that '...[this paper] used high-quality WGS (which includes better SVs than Illumina) and still found that adding rare and structural variants only modestly increased heritability estimates", which is NOT TRUE. Wainschtein uses the TOPMED dataset, which is from Illumina short reads. Yes, they do 'deep' sequencing, and yes it's analysed to the absolute hilt with the latest and greatest GatK pipeline and QC to the max. But that claim is false, it's just lists of SNPs, completely ignores huge chunks of the genome and just hopes that the thing contributing to a phenotype is is able to be fished out alongside a SNP.
Your Review: My Father’s Instant Mashed Potatoes
August 08, 2025 · Original source
Claude, by the way, estimates that 30-40% of all mashed potatoes eaten in the US are the instant kind. ChatGPT says 25-35%.
The Fatima Sun Miracle: Much More Than You Wanted To Know
October 01, 2025 · Original source
Although it’s hardly scientific, I was charmed by John Touhey’s project of trying to visualize the miracle by using witness descriptions as prompts for ChatGPT. His work is a year old, and so several GPT iterations out of date. When I repeat his work with the current version, I get these: Interlude: The Anti-Clerical Union As mentioned briefly before, 1910s Portugal was in a period of transition. In 1910, a group of proto-socialist revolutionaries overthrew the monarchy. The monarchy and church had been in cahoots, so the revolutionaries cracked down on Catholicism, closing the monasteries and persecuting the churches. This was a bold move - only an upper crust of educated urbanites were proto-socialist, and 99%+ of the country identified as Catholic, albeit at various levels of religiosity. In the 1920s, conservatives would regain the upper hand, overthrow the proto-socialists and restore a pro-church dictatorship. Still, the small urban educated ruling class of 1910s Portugal was a hotbed of atheistic anti-church sentiment. Probably the child-seers of Fatima were only dimly aware of this, but their prophecies were a spark entering a powder keg, and many of the more worldly witnesses were aware of this context. While reading through Fatima-related documents, I came across some pamphlets by Grupo Anticlerical, one of the era’s leading atheist organizations. They are totally irrelevant to our primary goal of trying to figure out what’s up with the miracle. But I love them so much that I can’t resist adding one as an interlude. I have slightly edited the machine translation for clarity and readability: To defend the sacred freedom of conscience—guaranteed by the original Law of Separation of Church and State—from the furious attacks of implacable Jesuitism—the greatest enemy of all human happiness!—the Anticlerical Group was organized in this town, similar to what is being done in many parts of the country! This was necessary. They call us to fight. We present ourselves courageously! The great, formidable battle of progress against Ultramontane Reaction, of Freedom against Tyranny, of Truth against Lies is waged again with enthusiasm and ardor! The redemptive dawn that the Portuguese people saw emerge on October 5, 1910, is about to be eclipsed, intercepted by the immense flood of black cassocks!... But in the dark night that seeks to envelop Reason; where moral suffering takes on tragic proportions in a frightening asphyxiation, the Light will once again break through!... the consoling light of elevated spirits... and like a sinister scarecrow, the grim reaction will flee in terror! Liberal people! Hear us! This fight is terrible! Many of our people will perhaps be crushed and tortured on the battlefield, but what does it matter?! Every war against reaction is a holy war because it frees consciences from the clutches of their enemies!... It is the fight of Justice against Iniquity, of Love against Hate, of Good against Evil!... To the fight, then, for the Progress that makes life beautiful; for the Freedom that redeems the people; and for the science that guides us all as an eternal beacon to the Light of Truth! Gago Coutinho and Sacadura Cabral [two Portuguese aviators who had recently flown across the Atlantic] are prodigious spirits before whom our souls kneel religiously – boldly breaking through the air with the mathematical certainty of someone who knows the path to be taken to get from one point to another determined point; flying through the immense blue as sure of their route as any of us walking on earth, they showed us that Science is not an empty word! The power of their prodigious sextant, the fruit of immense scientific lucubrations, is more real and positive than the cross of Christ painted on their device, which could not even have saved them from falling due to lack of gasoline in the middle of the sea at the mercy of the waves. Their extraordinary journey, an adventure which moved us to tears, was the most resounding scientific victory of recent times! It was, above all, a powerful affirmation of science! Let us therefore make science our religion, for scientific religion is Freedom of Thought! To be a Free Thinker is to love immortal science, eagerly waiting for it to reveal to us the truth of the great enigmas of the Universe! And only it can reveal them! People! Let us always fight! From the victory of progress, science, freedom, and free thought, will result human happiness, joy, love, fraternity, respect for women, veneration for mothers, adoration for children, affection for the elderly, protection for the sick, the unfortunate, the tortured. The victory of reaction, of clericalism, of black, cruel and ferocious Jesuitism will result in: the gallows, the acts of faith with their human destruction, persecution, exile, robbery, arson, the deflowering of women, the killing of children, the monstrous torture of all free spirits! The history of so many crimes committed in the name of God horrifies us! The Inquisition, relentlessly slaughtering, tearing, and burning the flesh of so many victims, is still today, in the twentieth century, a sinister specter haunting us!... O most holy mothers! O holy, pious mothers who so love your sweet little children! Have compassion on your beautiful little children, sacred fruits of your blessed wombs: Love Freedom! Love Liberty, O loving mothers, immaculate saints of our altar! We pray for them... for your children, who are the light of your candid eyes, the life of your life... for little children... for all children, tender rosebuds that retrogression furiously lashes, – love Liberty!. And you, O parents! Heads of families who so tremble at your loved ones, snatch them from the merciless clutches of the reactionaries who twist their brains and kill their reason! Hear us all, men, women, and children; listen: Freedom writhes in horrible convulsions... it vibrates in space, echoing from mountain to mountain, an anguished cry for help!... It is Freedom that falls, annihilated! It is Freedom that dies in the bloody clutches of Jesuitism! The Miracle of Fatima, people, is a ridiculous lie, it is a comedy, it is not religion! Come on, liberals! Let us all rise up from this criminal apathy and, without delay, fight not the religious sentiment of the Portuguese people, such a good people, a race of heroes, but rather the exploitation that clericalism is inflicting on the people, foisting upon them, at a good price, images of the saint —trademarked to avoid competition from other vampires! —the shamelessness!—and leading them, through suggestion, to wallow and drink madly, the miraculous water, foul, filthy water, full of rot, pus, and pestilent microbes that the sore flesh of the sick leaves deposited there in the washings! We, all as one man, will fight the reaction, forcing it to retreat and thus, with our efforts, we will save the Republic and the Portuguese Land from its fatal annihilation! … …anyway, Interlude over, let’s get back to the miracle. 2: The Skeptical Explanations Re-invigorated by the rousing prose of Grupo Anticlerical, can we come up with a materialist explanation for the sun miracle? 2.1: Pilgrim, Avert Thine Eyes Starting in October 1917, doubters have focused on one obvious possibility: staring at the sun is harmful to your health. If you stare too long, you go blind. If you stare just slightly less long than that . . . maybe something strange happens? Just to get a particular theory out there: everyone knows that if you stare at a bright light source for a few seconds, you get a temporary afterimage - often pink or bluish-green - on your retina. Suppose the pilgrims stared at the sun. Their eyes would inevitably make microsaccades - small natural jerking motions - and the afterimage would appear somewhere slightly different than the true sun. This might look like the sun turning pink or blue and moving in a zig-zag pattern. Believers in the miracle counter this proposal in several ways. First, although it might explain the sun changing colors and dancing, it doesn’t give an explanation for spinning, sparkling, or falling to earth and threatening to crush everybody (exactly three times in a ten minute interval, no less). Second, although witnesses describe the sun changing color, they also describe everything around them changing color to match the sunlight, which doesn’t match localized afterimages. And one scientifically-minded witness specifically describes closing his eyes to see if there was a persistent afterimage; he says there was not. Third, there are no reports of eye injuries or blindness from a crowd that was, supposedly, staring straight at the sun for ten minutes. This is a good match to witness reports (that the sun was unusually pale and didn’t hurt to look at) and with Dalleur’s theory (that it wasn’t the sun). But it’s a bad match to any theory depending on eye injuries. Fourth, this would require Portuguese people to be total idiots. Everyone already knows bright lights cause afterimages. Surely if you stare at the sun for ten minutes and get some afterimages, you’re not going to freak out and start screaming about miracles and the end of the world. Even if the peasants had somehow remained ignorant of afterimages their whole lives, the scientists and doctors in attendance wouldn’t be fooled. If we are to keep this theory, maybe we should posit some retinal phenomenon much stronger than the ones we know. Everyone thinks they know how much an illusion can fool you - “yeah, okay, obviously the cookie that looks very slightly bigger will actually be the same size” - which is exactly why the really good ones, like the Checker Shadow Illusion, come as such a shock. Squares A and B are the same color. Source: Checker shadow illusion. There’s no way around it: we need to hear from someone who has stared directly into the sun. August Meessen was a physics professor at a Catholic university, which sounds like exactly the job profile we want for this sort of thing. He found himself sufficiently interested in the Fatima miracle to stare straight into the sun for a few minutes and record what happened. From his paper: In November 2002, I looked directly into the sun, at about 4 p.m. The sun was relatively low above the horizon and its light intensity was attenuated, although the sky was clear. I was able to look right into the sun and was amazed to see that the sun was immediately converted into a grey disc, surrounded by a brilliant ring. The grey disc was practically uniform, while the surrounding ring was somewhat irregular and flamboyant, but did not extend beyond the solar disk. It coincided with its rim. I stopped the experiment, since I wanted to be prudent, but I had experienced myself the initial phase of a typical “miracle of the sun” and I could explain it. The sun became grey, since my eyes immediately responded to its great luminosity by an automatic reduction of their sensitivity. This adaptation is not simply due to the bleaching of pigments in the colour-sensitive cones of the fovea, where the image of the sun is projected, but to secondary processes. By “initial phase”, he means the part where the sun looks pale and well-defined, like a full moon. This isn’t something I think needs explanation (see above), but he sure has explained it. Moving on: In a second experiment, realized at 3 p.m. in December 2002, I looked straight at the sun during a much longer time. After some minutes, I saw impressive colours, up to 2 or 3 times the diameter of the sun. They changed, but were mainly pink, deep blue, red and green. Further away, the sky became progressively more luminous. I stopped there, since I understood that these colours resulted from the fact that the red, green and blue sensitive pigments are bleached and regenerated at different rates. This is frustratingly vague. Are the “impressive colors up to 2-3 times the diameter of the sun” just the normal aftereffects of staring at a bright object? Or something surprising even to physics professors? And the spinning? What about the motions of the sun? I didn’t see them, because I didn’t look at the sun for a sufficiently long time or my brain knew already too much. Once, after I had been looking at a very long passing train, I had (for about 30 seconds) the illusion of an opposite motion. Joseph Plateau discovered that when we look at the centre of a spiral that is rotating at some given velocity about this point, and when we stop this rotation, we see a reversed rotation. It lasts for several minutes, although in reality, there is no motion at all. This is a good example of motional after-effects. The “dance of the sun” is initiated, however, by a spontaneous generation of apparent motion. This feels suspiciously like a just-so story. His explanation for the sun falling to earth to crush everyone - which he also did not see - is equally ad hoc: A very interesting study was recently devoted to this “zoom and loom effect”. It tends to appear when the brain is confronted with the two-dimensional retinal image of an object that is situated at some unknown distance. The brain will then consider the possibility that it could come closer, by performing an illusory mental zoom, where the apparent size of the object is progressively increased. This results from the fact that evolution preserved the tendency to take into account the possibility of a dangerous approach: a rapid evasive action could be beneficial for survival. If true, it sounds like you should be able to generate this effect not just by staring at the sun (ill-advised, causes blindness), but by staring at the moon. I would like to test this, but unfortunately I am writing this on the night of a new moon; I’ll check back in two weeks. Still, I am skeptical that no human being living before 1917 AD ever figured out that staring at a celestial body long enough would make it appear to fall to earth and crush you. Compare to much gentler illusions - like how the moon looks bigger right when it starts to rise - which everybody knows about. I was able to find a thirdhand report (Fr. Stanley Jaki → G. J. Strangfeld → consultation with bishop) of another sun miracle investigator, one “Professor Dr. Stöckl” in Germany, who made a similar experiment: After almost a minute (the time varies according to the condition of the atmopshere and the momentary condition of the eyes) one thinks to see a dark blue disk in front of the sun (this is already a sign of the highly excited state of the retina). According to my experience … this dark blue disk is somewhat smaller than the solar disk, so that the edge of that disk stands out as a ring beyond that dark blue disk. Then one has right away the impression that the solar disk rotates with great speed in one or the other direction. This I have experienced often enough. All this is a subjective appearance that has nothing to do with the external world. These reports are suggestive, but weaker than all but the barest Fatima testimonials. Dr. Messeen admits as much, saying that “I didn’t look at the sun for a sufficiently long time”. Can we find people even more committed - or reckless, or masochistic - than Professors Messeen and Stöckl? Absolutely yes: there was a whole subfield of late 18th / early 19th century psychophysicists who experimented with staring at the sun for long periods, many of whom went blind. Joseph Plateau (1801 - 1883, went blind in 18432) summarizes their work in his aptly-named On The Contemplation Of Bright Objects. He lists twenty-six scientists who tried staring at the sun for a really long time. Most describe what we now recognize as typical retinal afterimages, and Plateau spends most of his time talking about how long these last and what colors they pass through. The only one of Plateau’s sources who reports anything even slightly interesting to us is Robert Darwin (father of Charles; cf. Secrets of the Great Families). After stating that: The author has frequently observed that when he gazed at the midday sun for a long time, until its disk appeared pale blue, he saw a bright blue specter on other objects for more than two days. …he mentions how When looking at the meridian sun as long as the eyes can well bear its brightness, the disc first becomes pale, with a luminous crescent, which seems to librate from one edge of it to the other owing to the unsteadiness of the eye. Here is pallor, and at least a hint of motion. But it’s pretty different from spinning, and not really clear how it relates to the sun miracle. Gustav Fechner (1801 - 1887, went blind in 1839) may have stared for even longer; you can read more of his story - including his ensuing insanity and subsequent attempts to found a new religion - on Adam Mastroianni’s blog. But all that he records about his ill-fated experiment is that: …after looking at the sun through homogeneously colored lenses, if you close your eyes, the primary impression remains for a long time and the entire afterimage usually disappears without a complementary coloration having clearly emerged. These people are great, and they all sound like minor Sam Kriss characters. But after whole careers dedicated to staring at the sun much longer than any normal person would ever try, they report only the barest hints of odd phenomena. Indeed, if anything they saw less of interest to the Fatimologist than Profs. Messeen and Stöckl. Worse, all of these authorities saw their phenomena after seconds to minutes of deliberate staring. Surely if it had taken a minute of staring at the sun before anything happened, some of our eyewitnesses would have mentioned this; after all, several mention that they were starting to doubt after the child-seers’ deadline had passed a few minutes earlier. But by all accounts, the miracle was near-instantaneous. Although Messeen and Stöckl’s reports of miracle-like phenomena are intriguing, it doesn’t seem like they can be the whole picture. Let’s move on. 2.2: Aurora Borealis? At This Time Of Year? In This Part Of The Country? Localized Entirely Within Your Kitchen? Could the miracle at Fatima have been some kind of weird weather phenomenon? The main argument against is that if it were a common weather phenomenon, it would not have awed and terrified tens of thousands of people. But if it were a rare weather phenomenon, then the seers’ successful prophecy that the rare weather phenomenon would happen at solar noon on October 13 1917 becomes almost as impressive as an outright miracle. The argument in favor is that dozens of people have written books and papers about this possibility, we would feel remiss if we didn’t mention them, and anyway it gives us the opportunity to look at pretty pictures of interesting weather phenomena. This is a sun dog. It’s caused by ice crystals in the upper atmosphere that refract sunlight in a very specific way. It’s very cool, but aside from a resemblance to a wheel, it looks nothing like the miracle of Fatima. A sun dog doesn’t have any unusual colors, it doesn’t change size, and it doesn’t spin (I’ve embedded a YouTube video not because a still image would be misleading - it wouldn’t be - but just in case you want to see for yourself how completely motionless it is). It’s just a halo shape with two smaller illusory suns on either side of the real one - something which no one at Fatima reported. (source) This is a solar corona3; cloud iridescence is a related phenomenon. I don’t know how much work the exposure length is doing in this particular photo, but I’m guessing more than zero. Coronae are also very pretty, and might explain the description of wheels and colors. They seem surprisingly common for something that I can’t ever remember seeing, supposedly happening several times a year in most locations. But they don’t spin, the colors don’t change or stain the surrounding landscape, and they don’t fall to earth and crush people. Let’s keep this one as a backup option and move on. This is a dust storm. Steuart Campbell wrote a paper arguing that the miracle was caused by one of these, and I admit if I saw this I would start praying pretty hard. Dust storms can change the color of the sun (including unusual colors like green or blue). And very, very charitably, whirling dust could look like the sun itself spinning around, and the thickening and thinning of dust could look like the sun approaching or receding. But this would require a dust storm localized to a 20 mile region of Portugal which does not, technically, have any dust (and where it was, technically, raining at the time). Campbell proposes that perhaps a storm blew a 20 miles x 20 mile dust cloud from the Sahara out to the Atlantic, then onto Fatima for ten minutes during a break in the rain, then back to the Atlantic again. But I don’t think any dust storm has ever behaved in quite this way. If it did, it probably wouldn’t be at the exact moment predicted by child-seers months in advance. At this point, we might as well talk about literal meteors. The way I’m imagining it is this: as a meteor approaches Earth, it breaks up into three big parts and a host of smaller particles. They strike the atmosphere head-on, from the approximate direction of the sun. The small particles hit first and make a firework show. Then the three big pieces hit, producing multicolored fireballs (meteors can absolutely stain the sky bright colors - see the video). Finally, they burn out a few miles above the ground, , convincingly producing the appearance of the sun falling to earth and nearly striking the spectators. This could even explain the warmth and dry clothes - a local meteor strike produces a lot of heat! I like this because it’s the only one that takes seriously the facet of the event which most impressed the witnesses - the part where it looked like the sun was plummeting to earth and about to kill them. But against it: would a rain of micrometeorites really look like the sun was “dancing”, “spinning”, or “zig-zagging”? Aren’t most nearby meteor strikes very loud? (the Fatima event was, according to witnesses, silent) Don’t they usually break windows? Aren’t most meteor strikes of this size visible for hundreds of miles, not just the twenty miles from which we have witness testimonies? Wouldn’t the strike have to be remarkably head-on, and remarkable close to the position of the sun, in order to look like a solar phenomenon rather than a long streak? Aren’t most meteor fireballs visible for between a few seconds and a minute, not the ten minutes of the Fatima event4? And if there were some extremely unusual meteor strike that was the exception to everything, wouldn’t it still be pretty surprising for it to happen at the exact time and place predicted by child-seers months in advance? We come to the unpromisingly-titled Derivation of equations of the model of the dynamic behavior of the three-dimensional atmospheric cloud of electrically charged ice crystals under the influence of electrostatic forces, in which Artur Wiroski argues that Fatima was a three-dimensional atmospheric cloud of electrically charged ice crystals under the influence of electrostatic forces. Actually, he offhandedly mentions Fatima in three sentences, with the majority of the paper looking more like the image above - but he eventually makes it into a Guardian article where he emphasizes that yes, he is trying to explain the miracle of the sun. However, if I’m understanding him correctly, he says that his theoretical ice crystal phenomenon can only happen when the sun is at an altitude below 22 degrees. But during the Fatima miracle, the sun was at 42 degrees (and Dalleur’s mysterious light source was at 30 degrees), so none of this applies. I’ve tried to include pictures of all the phenomena I mention in this section. I failed for this one, because it’s never been spotted or photographed. It’s just some incredibly weird thing that one scientist says ice crystals might do if parameters were ever exactly right, with such a precise definition of “exactly right” that it’s never happened in real life. If it ever did happen, it probably wouldn’t be at exactly the moment predicted by child-seers several months in advance. 2.3: Everyone’s Mad Here Except You And Me Another common response calls the Sun Miracle a “mass hallucination”. Can 70,000 people really hallucinate the same thing? “Mass hallucination” on Wikipedia redirects to List Of Mass Panic Cases. The Miracle of the Sun is on there, but listed as “(disputed)” - the only item to earn such a parenthetical. The other fifty items mostly belong to three categories: A disease with unusual symptoms spreads through a population; doctors eventually pronounce it psychosomatic.
Inline links: John Touhey’s project, https://substackcdn.com/image/fetch/$s_!ybv5!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6509924b-583f-4686-8ba9-cd4185892199_1012x463.png, https://substackcdn.com/image/fetch/$s_!12sU!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F5d21e8d2-ad0d-4a15-9ad0-0048338bcb33_1012x470.png, https://substackcdn.com/image/fetch/$s_!Wkjz!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fcff92f63-ad21-42fa-ad6f-c412b71f8524_631x481.png, https://substackcdn.com/image/fetch/$s_!mhyb!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa5fdf3d3-3e36-403e-95b9-f9fcbe7a06f0_1280x974.png, Checker shadow illusion, his paper, 2, Secrets of the Great Families, Adam Mastroianni’s blog, https://substackcdn.com/image/fetch/$s_!kvPF!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F30bf02a3-f57b-4a9f-a849-e7e318c65b6f_900x600.jpeg, source, 3, cloud iridescence, https://substackcdn.com/image/fetch/$s_!Y1Qt!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F9bbd9216-eb7d-4012-b114-bae56dbb0d4c_1024x681.webp, wrote a paper, 4, https://substackcdn.com/image/fetch/$s_!Id7a!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3c37c56e-bdf1-46e2-8d29-d395987be105_851x310.png, Derivation of equations of the model of the dynamic behavior of the three-dimensional atmospheric cloud of electrically charged ice crystals under the influence of electrostatic forces, a Guardian article, List Of Mass Panic Cases
This is a dust storm. Steuart Campbell wrote a paper arguing that the miracle was caused by one of these, and I admit if I saw this I would start praying pretty hard. Dust storms can change the color of the sun (including unusual colors like green or blue). And very, very charitably, whirling dust could look like the sun itself spinning around, and the thickening and thinning of dust could look like the sun approaching or receding. But this would require a dust storm localized to a 20 mile region of Portugal which does not, technically, have any dust (and where it was, technically, raining at the time). Campbell proposes that perhaps a storm blew a 20 miles x 20 mile dust cloud from the Sahara out to the Atlantic, then onto Fatima for ten minutes during a break in the rain, then back to the Atlantic again. But I don’t think any dust storm has ever behaved in quite this way. If it did, it probably wouldn’t be at the exact moment predicted by child-seers months in advance. At this point, we might as well talk about literal meteors. The way I’m imagining it is this: as a meteor approaches Earth, it breaks up into three big parts and a host of smaller particles. They strike the atmosphere head-on, from the approximate direction of the sun. The small particles hit first and make a firework show. Then the three big pieces hit, producing multicolored fireballs (meteors can absolutely stain the sky bright colors - see the video). Finally, they burn out a few miles above the ground, , convincingly producing the appearance of the sun falling to earth and nearly striking the spectators. This could even explain the warmth and dry clothes - a local meteor strike produces a lot of heat! I like this because it’s the only one that takes seriously the facet of the event which most impressed the witnesses - the part where it looked like the sun was plummeting to earth and about to kill them. But against it: would a rain of micrometeorites really look like the sun was “dancing”, “spinning”, or “zig-zagging”? Aren’t most nearby meteor strikes very loud? (the Fatima event was, according to witnesses, silent) Don’t they usually break windows? Aren’t most meteor strikes of this size visible for hundreds of miles, not just the twenty miles from which we have witness testimonies? Wouldn’t the strike have to be remarkably head-on, and remarkable close to the position of the sun, in order to look like a solar phenomenon rather than a long streak? Aren’t most meteor fireballs visible for between a few seconds and a minute, not the ten minutes of the Fatima event4? And if there were some extremely unusual meteor strike that was the exception to everything, wouldn’t it still be pretty surprising for it to happen at the exact time and place predicted by child-seers months in advance? We come to the unpromisingly-titled Derivation of equations of the model of the dynamic behavior of the three-dimensional atmospheric cloud of electrically charged ice crystals under the influence of electrostatic forces, in which Artur Wiroski argues that Fatima was a three-dimensional atmospheric cloud of electrically charged ice crystals under the influence of electrostatic forces. Actually, he offhandedly mentions Fatima in three sentences, with the majority of the paper looking more like the image above - but he eventually makes it into a Guardian article where he emphasizes that yes, he is trying to explain the miracle of the sun. However, if I’m understanding him correctly, he says that his theoretical ice crystal phenomenon can only happen when the sun is at an altitude below 22 degrees. But during the Fatima miracle, the sun was at 42 degrees (and Dalleur’s mysterious light source was at 30 degrees), so none of this applies. I’ve tried to include pictures of all the phenomena I mention in this section. I failed for this one, because it’s never been spotted or photographed. It’s just some incredibly weird thing that one scientist says ice crystals might do if parameters were ever exactly right, with such a precise definition of “exactly right” that it’s never happened in real life. If it ever did happen, it probably wouldn’t be at exactly the moment predicted by child-seers several months in advance. 2.3: Everyone’s Mad Here Except You And Me Another common response calls the Sun Miracle a “mass hallucination”. Can 70,000 people really hallucinate the same thing? “Mass hallucination” on Wikipedia redirects to List Of Mass Panic Cases. The Miracle of the Sun is on there, but listed as “(disputed)” - the only item to earn such a parenthetical. The other fifty items mostly belong to three categories: A disease with unusual symptoms spreads through a population; doctors eventually pronounce it psychosomatic.
Inline links: wrote a paper, 4, https://substackcdn.com/image/fetch/$s_!Id7a!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3c37c56e-bdf1-46e2-8d29-d395987be105_851x310.png, Derivation of equations of the model of the dynamic behavior of the three-dimensional atmospheric cloud of electrically charged ice crystals under the influence of electrostatic forces, a Guardian article, List Of Mass Panic Cases
Links For October 2025
October 30, 2025 · Original source
34: How does ChatGPT score on a Big 5 personality test? Answer (rot13d): vg qvfpbirerq gung vg pbhyq pbzcyrgr gur grfg snfgrfg ol cbfgvat n HEY jvgu gur ahzore guerr 181 gvzrf va fhpprffvba, naq qvq guvf, trggvat na rknpgyl arhgeny fpber. V ybir guvf nf n zrgncube: bhgre nyvtazrag vf vzntvavat gung lbhe tbny vf gb tvir gur NV “gur evtug crefbanyvgl”; vaare nyvtazrag vf ernpuvat gur cbvag jurer vg jvyy gnxr gur grfg ng nyy.
Inline links: Answer (rot13d)
43: Related: most people have the varicella zoster herpesvirus (aka chickenpox virus) latent in their bodies. Occasionally it reactivates in old people with bad immune systems and causes a rash called shingles, so old people are recommended to get the shingles vaccine. A new study shows that herpesvirus reactivation may be involved in dementia, and that the shingles vaccine significantly decreases dementia risk while in effect (~5 years). Celebrity epidemiologist Eric Feigl-Ding suggests that young people try getting the shingles vaccine for dementia prevention even if they don’t need it for shingles, but the exact pathway (and whether it helps preemptively) is not clear, and I think this is still a minority opinion. Here is ChatGPT’s assessment.
Inline links: A new study shows that, suggests, Here is ChatGPT’s assessment
44: OpenAI’s statistics on what people use ChatGPT for (source on X):
Inline links: source on X
Moltbook: After The First Weekend
February 02, 2026 · Original source
Janus’ simulator theory argues that all AI behavior is a form of pretense. When ChatGPT answers your questions about pasta recipes, it’s roleplaying a helpful assistant who is happy to answer pasta-related queries. It’s roleplaying it so well that, in the process, you actually get the pasta recipe you want. We don’t split hairs about “reality” here, because in the context of a question-answering AI, pretending to answer the question (with an answer which is non-pretensively correct) is the same behavior as actually answering it. But the same applies to AI agents. Pretending to write a piece of software (in such a way that the software actually gets written, compiles, and functions correctly) is the same as writing it.
Inline links: Janus’ simulator theory