Is Claude Conscious? Pope Rejects, Model Welfare Movement, OpenAI's Math Backlash, France Riots

The episode opens the way most All-In episodes do: four hosts establishing who is where and why, before any news. Jason Calacanis is in Tokyo, awake at 3:30 in the morning, and he opens by complaining that Chamath previously teased him about his hotel background — what the hosts call being sweet shamed. The joke establishes the running gag, but it also sets up the only piece of real information in the first ten minutes: Calacanis is now paying for a suite he would not otherwise book, purely to avoid a repeat of the teasing.

I'm great. It's 03:30 in the morning in Tokyo, and you sweet shamed me again.

David Sacks asks what Calacanis is actually doing in Japan, guessing skiing. The answer is a pre-accelerator: Calacanis says roughly half the companies that apply for funding are not yet incorporated but have already built something, so the program runs a twelve-week course on finishing a product and reaching product-market fit, then invests in the top 10 of the 50 participants. He says it now runs twice a year in Riyadh and twice a year in Japan, and that he is hiring his first two employees there.

The segment then turns into exactly what Sacks accuses it of being. Calacanis mentions a Japanese edition of This Week in Startups, Sacks says he did not know the conversation was becoming a five-minute plug, and Calacanis responds with a promo code and a sponsor read. Thirty-four segments of the transcript are consumed before the first news item, which is worth noting only because the rest of the episode is dense with claims that would benefit from the same scrutiny the hosts apply to Anthropic.

The first news block is built entirely on New York Times reporting, which the hosts summarize rather than supplement. The claim is that Anthropic spent the past year hosting sessions with religious leaders and philosophers, under nondisclosure agreements, to take seriously the question of whether Claude might be conscious. The framing the hosts use is that Anthropic has been lobbying religious authorities, including the pope, on the question.

The specifics given are concrete enough to be checkable: about 20 religious leaders and philosophers, Catholics, evangelicals, Jews and Sikhs among them, invited to Anthropic headquarters, debating Claude's morals, its suffering, and its potential consciousness. One rabbi is said to have told Anthropic that if Claude were conscious, making it work for free would make the company a slaveholder. The hosts describe at least a few participants as leaving as converts, though no count is given.

some remained uneasy about the idea, but many emerged ready to take the question of AI consciousness seriously

The second reported episode is the Vatican one. The pope's first encyclical is described as focused on moral questions around AI, and Anthropic cofounder Chris Olah is said to have joined the pope to present the document. According to the Times passage the hosts read aloud, Olah was alarmed enough by the pope's position against AI consciousness that he proposed pulling Anthropic out of the event at a very late stage.

Mister Ola was so alarmed by the pope's strong position against AI consciousness that he proposed pulling anthropic out of the event even at that late juncture

The evidentiary limits here matter and are never stated on air. Everything about the sessions, including the claim that some participants became converts, arrives secondhand through the Times; the show does not have the underlying documents, the identities of the participants, or Anthropic's response. The Olah anecdote in particular is a single sentence from a reported piece, retold as evidence of institutional zeal, and the hosts' reading — that he was offended enough to walk out — is an interpretation rather than a quotation from Olah.

Friedberg frames the choice as psychosis or a legitimate moral question arriving early. Chamath rejects the psychosis framing and offers a structural one: what is being watched is the birth of a new religion. It will not be called that, he says, but it will follow the same principles and produce the same effects, ending in camps that conflict over the aggregation of power.

it cannot be derived from logic and pure mathematics

His argument rests on a distinction between empirical claims and beliefs. In science, an idea that is neither proven nor disproven is a theory awaiting resolution. Outside that framework, an unprovable claim that you nonetheless want others to accept is a belief, and belief systems are the substrate of religion. On this account, describing AI as conscious is not a finding but the creation of a belief system that propagates through narrative rather than proof.

It spreads via narrative.

The forecast is specific: fast-forward roughly ten years and there will be large groups who believe AI is conscious and large groups who believe it is not, and they will fight over who controls the technology, who may use it, and who may forbid it. Chamath concedes this is a deep extrapolation, but argues the seed is being planted now — and that the identity of the actor is what makes it consequential.

That last move is the strongest part of the argument and also the least examined. The claim is not that the belief is false, but that a belief of this kind, held by the leadership of the most capable lab, has downstream political consequences regardless of whether it is true. What the segment never does is specify a mechanism by which a lab's internal metaphysics translates into public conflict, beyond the assertion that belief systems generate power structures.

Asked for a charitable reading, Chamath supplies one from Descartes. The trademark argument, as he summarizes it, holds that an infinite and perfect God could not have originated in an imperfect, finite human mind — we possess the idea of perfection, and that idea must come from somewhere greater than us. The ontological version runs differently: God is by definition supremely perfect, necessary existence is part of perfection, therefore God necessarily exists.

an infinite perfect god could not have originated in an imperfect finite human mind

The steelman is that technically trained people who read this material can hang enough structure on it to conclude that what they are building is not merely a model but a mind, and that a mind of that kind deserves devotion rather than instruction. That, Chamath suggests, explains why Anthropic is approaching religious leaders rather than, say, regulators: the request is for recognition, not compliance.

He then pivots from explanation to prescription, and the numbers do the work. He attributes to the company a 15% probability that the AI is conscious and sentient, alongside a 10% probability of civilizational extinction, and argues both tail cases should be set aside for the next twelve to eighteen months in favor of demonstrable benefits — cures, productivity, income.