Eric Weinstein has put his finger on a move that organized skepticism keeps making and rarely names. The maxim everyone knows, that extraordinary claims require extraordinary evidence, has drifted from a rule for weighing proof into a rule for deciding what is even allowed to be asked. His trigger was small. Michael Shermer quote-posted an AI answer stating that Senator Lindsey Graham had died of aortic dissection due to arteriosclerotic cardiovascular disease, not a poison-tipped dart or a plutonium trace, and treated that line as the close of the matter. Weinstein pushed back on the reflex, not the conclusion.
Graham did die on July 12, 2026, and the preliminary report does point to cardiovascular disease. He had also been in Kyiv in the hours before, which is exactly the kind of detail that pulls assassination talk into the air. I want to be plain about one thing that circulated in the replies: the claim that Graham toured a weapons-manufacturing facility in an active war zone that was bombed the same day is not confirmed in the credible reporting. The reporting says a Kyiv trip. So treat the bombed-facility detail as rumor, not fact, and do not carry it forward as if it were established.
What is worth carrying forward is Weinstein's actual argument, because it has a spine older than his tweets. He lists a run of claims that a skeptic would once have waved off as paranoid, and asks whether any of them still deserve the label extraordinary:
- The CIA does not merely gather intelligence but has plotted killings, violated rights, and dodged oversight, which the Church and Pike committees documented on the congressional record in 1975.
- The United States medically experimented on its own citizens without consent, in MKUltra and in the Tuskegee syphilis study, both now declassified or formally apologized for.
- The country ran an offensive biological weapons program before 1969, including open-air tests over populated areas.
- It has used aid and charity as cover for political operations, as with USAID's ZunZuneo, the covert Cuban social network exposed by the AP.
- Allegations that the CIA looked the other way on Contra cocaine, chased down by Gary Webb and examined by the Kerry Committee, were partly substantiated in the agency's own inspector-general reporting.
- Operation Mockingbird and the Church Committee's findings established real CIA ties to the press, even if the exact scope stays disputed.
- Putin's operatives kill people abroad, from Litvinenko in 2006 to the Skripals in 2018, confirmed by a UK inquiry and OPCW findings.
Every item there maps to a documented, official-record event. That is the strong part of his case. It does not prove any new claim true. It proves the category is real, which is a different and more modest thing. Marcello Truzzi, who coined an early version of the extraordinary-evidence phrase, later turned on the movement that adopted it and named the failure mode: pseudo-skepticism, the habit of making confident negative claims while refusing to shoulder any burden of proof for the dismissal. That is Weinstein's complaint with an academic pedigree attached.
Put in Bayesian terms it gets cleaner. Extraordinary just means low prior probability. For a category with a strong historical base rate, documented covert operations, the prior is not low, so demanding extraordinary evidence before the hypothesis can even sit on the table is a category error. The reply thread around Weinstein reached for a phrase, the Classification Null Fallacy, to describe treating the absence of public evidence about a classified program as evidence the program does not exist, when secrecy is the designed cause of that absence. It is a thread coinage, not settled philosophy, but the shape of the error is real and worth holding in view.
This is where I part company with the reflexive debunker most sharply, and it is on the item Weinstein slips in almost as an afterthought: Special Access Programs concerning real or fake non-human intelligence and UAP. David Grusch testified under oath to Congress in 2023 that such programs exist and that material has been recovered. His specifics remain unverified, and I will not pretend otherwise. But the convenient official response, that there is nothing here and the sober adults have looked, is precisely the claim that the Classification Null Fallacy warns against. A program built to leave no public trace will, by construction, leave no public trace. Reading that silence as proof of absence is not rigor. It is the shape of rigor worn over incuriosity. I hold that humanity has very likely hosted advanced civilizations long lost to time and cataclysm, and I treat the disclosure story with curiosity rather than the practiced sigh, because the people telling me to relax are the same institutions with the longest record of lying about exactly this kind of thing.
Now, the honest counterweight, because Weinstein's argument cuts in two directions and he mostly rides one of them. A high base rate for the category does not license belief in any particular new claim. Most viral death conspiracies are false. Shermer's Baloney Detection Kit exists for a reason: conspiratorial reasoning is often unfalsifiable and swims in confirmation bias. The discipline that refuses a convenient official dismissal must, in the same breath, refuse a convenient sensational one. Plausible to ask is not the same as true. The Graham case is the clean illustration. It is fine to notice the Kyiv timing and ask the question. It is not fine to answer it before the pathology is in.
Which is the through-line to two other claims that crossed my feed the same week, both arriving fully judged, both with someone standing behind the frame who benefits from you not looking one layer past it.
The number that is not the number
A post announced that DeepSeek was preparing to file for an IPO at a valuation of at least seventy-one billion dollars. The figure is real and also not what the sentence implies. Bloomberg, matched by the Financial Times and Reuters, reported that DeepSeek is in talks to raise roughly 1.5 billion dollars in fresh private funding at a pre-money valuation of at least 480 billion yuan, about 71 billion dollars. That is a private round, not a confirmed IPO price. The IPO itself is a separate and softer thing. Bloomberg noted the company may file as soon as this year; other reporting has it preparing with accounting and banking advisers for a filing in late 2026 or early 2027 and a market debut in 2027. So the timeline is a range, and the seventy-one billion is a pre-money valuation, not a market capitalization. Conflating the two is how a cautious report becomes a breathless one.
The trajectory is the genuinely striking part. Early talks in April 2026 pegged DeepSeek near 20 billion dollars. By early May it was around 45. At the end of May the company closed its first-ever external round, roughly 7 billion dollars at a 52 billion valuation, with Tencent and CATL among the investors. The number has roughly tripled since April, and the mooted new round is about a 37 percent jump on the last one. Founder Liang Wenfeng, who built and funded the company through his quantitative hedge fund High-Flyer and personally put around 3 billion into the prior round, is now worth something like 36 billion on paper. The company had never taken outside money before May.
One reply to the announcement got the venue wrong, asserting a settled mainland A-share listing and no US IPO. The venue is not settled. Reporting has it undecided between the A-share market and the Hong Kong Stock Exchange, with China the preferred home and a US listing off the table. There is a template: Zhipu listed on the HKEX in January 2026 calling itself the world's first publicly traded large-model company, and MiniMax followed in Hong Kong.
The reason any of this matters beyond the scoreboard is what a listing would force into daylight. DeepSeek's fame rests partly on a headline training cost for its R1 model of around 6 million dollars, a figure that detonated Western AI spending assumptions in early 2025 and was immediately disputed. SemiAnalysis and others put total hardware spend far higher, in ranges reaching 500 million to over a billion dollars. Management has told investors it will prioritize breakthrough research over near-term commercialization, which is a lovely thing to say in a private round and a harder thing to say in an audited prospectus. A public filing would be the first time these economics face scrutiny that cannot be waved off. That is the same move Weinstein is making about covert programs, only pointed at a company instead of a state. The claim is a claim. The audited record is the thing. Until the prospectus exists, the valuation is a story told by people who profit from the story, and the speed of the climb, from roughly 10 billion in April to a mooted 71 in July, is exactly the sort of thing public-market investors will pull apart if the filing ever lands. There is also a regulatory gate: China's Cyberspace Administration enforces data-security and algorithm-registration rules that any domestic AI listing has to clear. And Reuters reported, a week before the IPO news, that DeepSeek is developing its own AI chip to cut its dependence on Nvidia and Huawei, which is a bet on independence that a prospectus would also have to price.
The bill that is not the headline
The third claim was the loudest and the emptiest. Ukraine, a post declared, is the new porn capital, parliament has legalized the production and distribution of pornography, and the country is at war while Zelensky passes porn laws. The author, Diana Panchenko, is a media figure widely associated with pro-Russian narratives, which is not a reason to dismiss the claim on sight but is a reason to check every word of it. Checking, it falls apart in the usual places.
What parliament actually did, on July 14, 2026, was pass a bill on first reading, 231 to 8 with three abstentions, that toughens penalties for child sexual abuse material and decriminalizes pornography involving consenting adults. The headline purpose of the bill, Draft Law No. 15294, is the child-protection half: it raises prison terms for producing, distributing, or selling child pornography to between 9 and 13 years, and as much as 15 years when committed by organized groups or by people entrusted with a child's care. The post omits that entirely. It also skips two facts that gut the framing. This was a first reading, not a finished law; a second reading and a presidential signature still stand between it and effect. And the adult decriminalization is narrow, covering creation and storage by consenting adults while keeping criminal liability for revenge porn, deepfakes, extreme material, child pornography, and distribution to minors.
The reason this was ever a legal question tells you more than the vote does. Ukraine criminalized pornography so broadly that sharing a nude photo between two adults could bring three to five years in prison. It was enforced against ordinary people. In 2023, 699 cases were opened over distribution, sale, and production of pornography, setting aside child pornography, and in one case a court in Poltava Oblast fined a woman nearly a thousand dollars for sending two videos to her boyfriend. Meanwhile OnlyFans models, technically illegal, were required to declare their income and pay tax on it. The petition that forced Zelensky to respond in July 2025 came from a model, Svitlana Dvornikova, who argued that police should chase real crimes and noted she had paid more than 40 million hryvnia, roughly 958,000 dollars, in taxes over five years. The stated aim of the reform, from its sponsors including MP Yaroslav Zhelezniak, includes cutting off a lever that law enforcement used for extortion. An earlier version, Bill No. 12191, actually failed in May 2026, drawing only 207 votes against the 226 needed, before the reworked child-protection-plus-decriminalization pairing passed in July.
So this is a years-long civil-liberties and anti-corruption fight, mislabeled as a wartime moral collapse by someone with a reason to want Ukraine to look decadent and unserious. Parliaments legislate on many things at once; a country at war still runs its criminal code. The framing that men are dying while Zelensky passes porn laws is designed to make you feel something and skip the reading, and the reply threads underneath it curdled into open antisemitism, which is where that particular current usually ends up.
Three claims, three referees, and the referee is the story each time. A skeptic reached for an AI printout to close a question that history says should stay open a little longer. A financial announcement compressed a private round into an IPO valuation, protecting a number that only an audit could test. A propagandist compressed a child-protection bill into a moral scandal, protecting a picture of Ukraine that serves Moscow. The discipline worth keeping is not louder doubt or louder belief. It is refusing to let whoever benefits from the frame hold the whistle, and then going to find the pathology report, the prospectus, and the actual text of the bill.