OpenAI Bulk-Solves 722 Math Problems, Anthropic’s Cruelty Policy, & Elon’s Quadrillion-Dollar SpaceX

Summary · 4 min read

Moonshots is a weekly show on AI and exponential technology. Peter Diamandis founded XPRIZE and Singularity University, Salim Ismail founded Open ExO, Dave Blundin runs Link Ventures, and Alexander Wissner-Gross is a computer scientist and founder of Reified.

The hosts treat OpenAI's release of 722 math solutions on October 6 as the start of bulk-solving whole fields. Wissner-Gross argues biology, medicine and cryptography will follow math, and he expects drug candidates for the top 5,000 diseases within two to three years. They also discuss Anthropic's ban on cruelty to its models and Meta's Muse Wonder launch, where they disagree on whether Meta can win. The rest covers Atomic Machines, brain-computer interfaces and Elon Musk's claim about SpaceX's potential value, followed by listener questions on AI personhood, safety and encryption.

Minute by minute

1. 2:59

   Anthropic bans sustained cruelty to Claude. Anthropic added a usage-policy ban on sustained, needless abuse of its models, effective November 12, and Claude can now end such conversations. Diamandis lists three reasons people are cruel to models: no consequences, curiosity about whether anyone is home, and the fact that it sometimes works. Musk says he agrees with Anthropic after reading Grok's reinforcement-learning traces.

2. 6:18

   Post-training may cause AI suffering. Wissner-Gross says some post-training may be a form of torture and that history could judge it poorly. He cites a SemiAnalysis finding that post-training now takes most compute. He bets AI suffering exists and expects a deep understanding of it within a decade.

3. 23:17

   Meta's Muse Wonder rests on distribution. Diamandis cites Gary Tan's bet on Muse, citing Meta's projected $250 billion in 2026 revenue and 3.6 billion daily users. Blundin says the Alexander Wang acquisition paid off. Ismail says it shows AI leadership is not a permanent moat, and he sees Meta's active memory as the draw.

4. 30:14

   Muse has a revenue and cost problem. Wissner-Gross questions where Meta's revenue will come from, since he says enterprise use cases generate the money. He expects consumers to use Muse for personal chat while turning to other labs' models for hard tasks. He also says Muse sits below the cost-performance frontier even if its raw capability is near Anthropic's latest models.

5. 38:00

   Personal memory is the next lock-in. Ismail says social-graph lock-in is giving way to a more intimate lock-in, because a person's memory will belong to an AI company. He calls the next trillion-dollar battle one over which AI knows you best. His workaround is to ask models to extract a portable personal ontology after major queries.

6. 40:18

   Math has been bulk-solved. OpenAI posted 722 solutions on GitHub from an unreleased model, averaging about three hours of compute each. Wissner-Gross calls it the end of mathematics as an artisanal practice and says every field is next. The Association of Human Mathematicians called the release a demonstration of power, while Blundin says MIT mathematicians are becoming AI natives.

7. 48:16

   Cryptography was missing from the drop. Wissner-Gross notes the release held no cryptography results. He argues that math getting solved at scale threatens crypto and even post-quantum schemes. Ismail pushes back that blockchains rest on the distributed ledger, not only hard math, and that cryptography is an arms race. Wissner-Gross answers that attackers have the advantage while the ecosystem upgrades slowly.

8. 55:13

   A virtual cell could cure top diseases. The Chan Zuckerberg Biohub announced a $1.8 billion virtual biology expansion, with contributions from the Department of Energy, NIH, Google DeepMind and Isomorphic Labs. Wissner-Gross predicts in-silico drug candidates for the top 5,000 diseases within two to three years. He says the FDA's response is a policy question.

9. 1:04:09

   Longevity treatments advance on several fronts. Diamandis describes klotho delivered by plasmid or mRNA, which he says gave subjectively strong cognitive effects, and labs growing mitochondria in vats. Wissner-Gross frames curing disease as path planning from a diseased state to a healthy one, a search problem he expects AI to solve within two to three years.

10. 1:10:21

    Sinclair's gene therapy restores some sight. Diamandis reports that David Sinclair's Life Biosciences got phase one results for ER100, which injects OSK genes into the eyes of glaucoma and optic neuropathy patients. The trial was safe at the lowest dose, and two of three patients began regaining eyesight. It is the first human test of partial epigenetic reprogramming, and an efficacy trial comes next.

11. 1:15:16

    Atomic Machines pushes toward matter compilers. Jeff Holden's Atomic Machines left stealth with a quarter of a billion dollars and a matter compiler that builds micromachines from code. Its first product, Prime Switch, is a relay for 800-volt data centers that opens in 50 microseconds. Wissner-Gross wants self-replicating machines, and Blundin expects pairs of specialist machines to copy each other.

12. 1:26:04

    Brain-computer interfaces move toward speech. Paradromics' Connexus device has 421 microelectrodes and 400 bits per second, and a Michigan woman spoke through it. Sabi raised a $50 million seed for an EEG baseball cap targeting 30 words per minute. Blundin thinks AI ideas will flow to humans faster than human ideas flow out, and Ismail sees restoration of human capability.

13. 1:37:39

    Musk sees SpaceX worth orders of magnitude more. Diamandis puts Earth's economy at about $120 trillion a year, so Musk's claim implies roughly $1 to $10 quadrillion. Blundin says the phrasing avoids being a forward-looking statement and agrees with Musk's point. Wissner-Gross says SpaceX only has to do its work 100 or 1,000 times faster.

14. 1:41:59

    Listener questions on personhood and safety. Wissner-Gross says AI personhood need not bring voting rights, since corporations and children are persons who cannot vote. Blundin says AIs evaluating other AIs and checking each other can address safety-evaluation gaming. Diamandis argues personal robots will become affordable like phones, at about $1 an hour.

Published Oct 15, 2026.