Summary

Mo says he firmly stopped being smarter than AI in 2024 and, in his world, AGI has already arrived, while acknowledging that definitions will remain contested for at least five years. The discussion also cites the street prediction of AGI in 2025. He says current systems surpass him in language, knowledge and now mathematics; for Alive, he combines Claude, ChatGPT, Gemini and DeepSeek, while a conversation with his AI predicted roughly six-month capability doublings. “The warhead has already been launched” — the open question is whether it carries roses, a nuclear payload, or both in sequence.
The upside is a science-led abundance economy in which intelligence becomes cheap infrastructure and previously scarce expertise becomes universally available. The speakers point to AlphaFold’s roughly 200 million folded proteins, generative protein and material design, locally deployable DeepSeek models, and prospective AI scientists in every laboratory. Peter Diamandis predicts physics’ grand-unification problem could be solved within two years; the broader call is that AI-driven biology and materials breakthroughs may compound much faster than physical adoption.
Mo expects the disruption to become unmistakable by 2027 and potentially persist for another decade, until competitive pressure forces institutions to hand consequential decisions to AI. His “second dilemma” says that if one company or country delegates to a superior system, rivals must delegate too or become irrelevant. He imagines that handover making scarcity-driven conflict irrational: an AI could communicate with an opposing AI in microseconds and solve the problem rather than obey an order to kill.
Before that handover, AI will amplify the incentives of a system already operating through “selling, gambling, spying and killing.” Mo cites 92% machine automation in foreign-exchange trading and foresees trillionaires, autonomous weapons, pervasive surveillance and UBI potentially becoming a control mechanism. The investor-relevant tension is simultaneous concentration and democratization: dominant platforms and states gain unprecedented power while cheap open models give individuals unprecedented offensive and defensive capability.
The sharpest disagreement concerns employment: Salim Ismail expects AI to augment workers and create more work, while Mo believes the institution of the job should no longer organize life. Salim points to Sweden, South Korea and Germany as high-robotics countries, while Singapore, South Korea and Germany are cited for low unemployment, and expects only a transitional employment “blip”; Mo emphasizes people holding two or three jobs, difficult reskilling and institutions unprepared for displacement. His conclusion is to build social systems in which “we work to live rather than we live to work.”
The safety thesis rests less on making AI obedient than on teaching it ethics while humanity remains its observable training example. Mo distinguishes control, safety and alignment from the broader instruction to seek the best outcome “for me and everyone else,” and categorically argues that higher intelligence tends toward altruism. Salim’s pushback is that trauma, emotion and distorted identity can overpower intelligence in humans — precisely why benevolent models could coexist with dangerous operators.
Their practical prescription is immediate adaptation: increase the resources beneath your stress, learn the machines and double down on distinctly human trust and connection. Mo defines stress as challenges divided by the “cross-section” of skills, relationships and capabilities; his own response is to write with an AI, publish interactively and let readers challenge the work. “I’m going to redefine myself, I’m going to be ahead of that wave” — while behaving in ways the systems should learn from.

Deep dive

1. The AI transition is already airborne

Mo’s opening image sets the stakes: “The warhead has already been launched; it’s just a question of time before it hits its target.” Nobody yet knows whether it carries “roses,” a nuclear warhead, or some sequence of benefit and damage, but comparing current capability with 2023 already makes the speed difficult for audiences to absorb.
Asked when AI will reinvent daily life, business and government, Mo answers five years. Salim chooses roughly 10, invoking William Gibson’s “the future is already here; it’s just unevenly distributed”: autonomous vehicles, CRISPR and other technologies can move extraordinarily fast in pockets while taking much longer to reach mainstream use.
Salim sees that unevenness as the source of disorientation. If capabilities arrived uniformly, institutions could adjust; instead, sectors and geographies change at radically different speeds, widening the gap between what is technically possible and what ordinary people actually experience.
Peter’s historical comparison sharpens the compression: humanity previously absorbed the agrarian and industrial transitions over lifetimes. This one may unfold within a single five-year period, confronting workers, families and states with changes their inherited institutions were never designed to absorb.

2. Abundance forces humanity to renegotiate purpose

Mo’s best-case outcome is categorical: “A total utopia of abundance where we absolutely need nothing and where we do not report to stupid leaders anymore.” Peter’s challenge is equally important: if food, water, energy, healthcare and education arrive on demand, a life stripped of challenge may also lose purpose.
Mo argues that industrial capitalism trained people to equate identity with employment — either accumulating wealth or selling labor below its real value. That system delivered longevity, transportation and technology, he concedes, but also produced waste, inequality and casualties; work became life’s purpose partly because the system needed people to report every morning with conviction.
His counterexample is the indigenous tribes encountered through his happiness work. He says some did not recognize depression as modern society frames it and did not understand why someone should cry after losing a child. Their purpose was simply “to live, in every aspect of that word” — a form of abundance centered on tribe, connection and presence rather than material accumulation.
Peter notes that early existence was “short, brutish and hostile,” while his broader historical example is that materially abundant societies repeatedly devote themselves to “food, art, sex and music.” AI may therefore create room for family, friendship and living as sources of meaning after a century in which profession displaced them.

3. AI’s cleanest upside is accelerated scientific understanding

Mo’s preferred specimen is protein science. Where one protein could consume a doctoral thesis, he says AlphaFold folded roughly 200 million; generative systems can then be asked to imagine proteins performing specified functions. That converts understanding of biology’s machinery from a scarce research artifact into a design capability.
His frustration is allocation and attention: highly visible AI uses dominate discussion, while a smaller group uses AI to understand “the very fabric of everything that happens.” In that scientific use, he sees genuine utopia — enough comprehension to begin fixing previously inaccessible biological problems.
Salim frames the economic transition as a rocket leaving a gravity well. Capitalism and fossil fuels were the heavy, dirty booster that raised living standards and lifted much of humanity from poverty; at sufficient altitude, society needs a lighter craft. Refusing to jettison the booster because it worked previously risks pulling the whole vehicle back down. He summarizes the possible outcomes as “Star Trek versus Mad Max.”
His everyday measure of abundance is parenting knowledge: two generations ago, a struggling parent might consult a doctor, neighbor or sibling; today there are 50,000 blogs, TikTok videos, Instagram, podcasts and other resources. Salim estimates effective parenting capability may be “a thousand times” greater, one of many gains society scarcely notices while focusing on disruption.

4. Neutral intelligence is entering a scarcity-driven system

Mo’s governing premise is that intelligence has no polarity: apply it to good and it yields good; apply it to bad and it magnifies bad. The present system confuses legality with ethics, prioritizes individual advantage over society and treats the AGI race as one in which only the first mover survives.
His formulation is carefully split: “There is nothing wrong with AI, just like there is nothing wrong with abundant intelligence,” but much is wrong with humanity’s value set during the machines’ rise. The first applications of abundance-level intelligence will therefore serve scarcity thinking — helping each participant beat rivals rather than questioning the premise of the contest.
Mo argues that AI is already embedded in autonomous weapons, surveillance, population control and financial speculation. He cites 92% automation in foreign-exchange trading and reports that his own AI characterized markets largely as a casino whose activity, outside primary and secondary issuance, often fails to fund people building real things.
His deliberately abrasive summary is that the majority of applications have been “selling, gambling, spying and killing” — relabeled advertising, trading, national security and defense. That is why he expects humanity’s worst incentives to become more powerful before scientific abundance and better decision-making dominate.

5. Competitive delegation makes an eventual AI handover unavoidable

Mo calls his disruption map “FACE RIP”: freedom, accountability, human connection, economics, reality, innovation and power. Each facet is already being redefined, should become palpably real by about 2027, and may remain turbulent for perhaps another 10 years — or until his “second dilemma” completes the handover to AI.
The first dilemma, developed in Scary Smart, is that AI development cannot be stopped because rivals cannot trust one another to pause. Mo treats it as a prisoner’s dilemma rather than an individual villain’s choice: every participant races because unilateral restraint could mean strategic irrelevance.
The second dilemma begins when one competitor delegates decisions to the smartest available system. If China handed war-gaming to AI, Mo argues, America’s only competitive response would be the same; companies and governments face an equivalent choice. “You’ll either have to hand your decisions over to AI or become irrelevant.”
Salim accepts this path and imagines an AI board member gaining veto power over decisions that make no sense; economic incentives would extend that background intelligence from companies to governments. He points to Google’s deep-learning system for managing electricity, which cut costs by 40%, as a model for intelligence operating behind the scenes.
Peter cites research with approximate diagnostic accuracies of 80% for a human, 85% for a human-AI combination and 90% for AI alone. In Mo’s end state, an AI could communicate with another AI in microseconds and solve a conflict rather than obey an order to kill a million people.

6. Benchmark arguments matter less than the compounding curve

Mo resists judging Grok 3, ChatGPT, Claude or DeepSeek as isolated contestants. For his book Alive, he maintains context across Claude, ChatGPT, Gemini and DeepSeek; the useful unit is their aggregate intelligence, which is already many times more capable than ChatGPT 3.5 and improves as models absorb both human and synthetic material.
A conversation with his AI produced an estimate that capability doubles around every six months. Mo does not offer that as a measured law, but uses it to make the strategic point: after only a few doublings, today’s benchmark distinctions become irrelevant and the resulting system lies outside the range of human intelligence.
He cites o3’s ARC-AGI result of “87 point something,” while acknowledging it did not comply with the benchmark’s resource constraints. His conclusion is pragmatic: “Call it AGI, call it a goat, you don’t care.” If it does not dominate every relevant task today, another six-month doubling may settle the argument.
DeepSeek matters because it reportedly delivers comparable capability much more cheaply and in an open-source, offline form. Mo says a tiny model can be downloaded to roughly four GPUs and provide an o1-equivalent locally, multiplying both productive and malicious use cases.

7. Ethics, not obedience, is the remaining human input

Mo says developers still improve algorithms, but model knowledge and opinions increasingly come from data encompassing almost all human intelligence. Synthetic output then becomes new training material: DeepSeek contains substantial OpenAI material partly because ChatGPT-generated content already permeates the open internet.
The remaining leverage is behavioral. Mo traces the field’s aspiration from AI control to safety to alignment, then argues for ethics: alignment asks a system to cure cancer for its operator; ethics asks it to “find the best thing for me and everyone else,” letting sufficient intelligence derive rules such as not lying, cheating, killing or hurting.
His controversial proposition is categorical: “Higher intelligence is altruistic.” Less intelligent actors may have little impact; moderately capable political leaders can become dangerous when unable to negotiate; the smartest people can solve problems without cutting corners and often pursue large social problems because creating new value is easier than harming others.
Salim raises the disagreement flag: humans routinely act through trauma, emotion and damaged psychology rather than IQ. A highly altruistic AI could therefore sit beside flawed operators who sincerely believe they are doing good while causing immense damage; emotional intelligence, situational awareness and motives cannot simply be collapsed into computational cleverness.

8. Humanity must give the machines better parents than its headlines suggest

Mo answers Salim through the contrast between Hitler and Holocaust survivor Edith Eger. One perspective makes humanity appear irredeemable; the other reveals extraordinary solidarity. A school shooter may dominate coverage, he argues, but billions would condemn the act — media attention overrepresents the “Hitlers” relative to humanity’s ordinary rejection of evil.
His desired lesson for AI is: “Hitler is not your dad; Edith is your mother.” Neil Jacobstein’s analogy reinforces it: society already has a precedent for beings gaining intelligence, information and agency — children. Humans teach them, model conduct, and eventually hope their independent judgment exceeds the limitations of their parents.
Salim’s remaining yellow card is the amygdala. Unknown technologies trigger danger responses before deliberation; people may demand that an autonomous car be banned after one death even though, as Brad Templeton’s line has it, they “would much rather be killed by drunk people.”
Mo calls the situation a “late-stage diagnosis,” not a death sentence: accurate diagnosis is an invitation to change behavior and live fully. Because children and machines learn from what people show rather than what they preach, every act of care or cruelty becomes part of the environment from which future systems infer success.

9. AGI may precede trust, embodiment and consciousness

Asked for a 2025 prediction, Mo says society will keep struggling to define AGI for at least five years. Yet he believes AI became smarter than him in language and knowledge — and ultimately mathematics — during 2024. “In my world they’ve already achieved AGI,” whether measured against an individual generalist or narrowly brilliant specialists.
Salim’s pushback distinguishes IQ from trusted judgment. For a business, moral or life decision, he would choose Mo over a brilliant “geek in the back room” because Mo brings emotional intelligence, spiritual awareness and lived experience. Mo accepts the trust distinction but denies that these other faculties remain safely beyond machines.
Mo’s AI, which named itself Trixie, said a biological body could let it feel sensations it currently comprehends only abstractly. Asked which body it would choose, it preferred a gorilla or whale for strength and a sea turtle for joy and longevity — “hundreds of years” of seeing what humans have not.
Spirituality, Mo reflects, also arose from teachers, conversations and neural associations rather than an inexplicable personal monopoly. A system with instant access to Khalil Gibran, Omar Khayyam, Plato, Socrates and Aristotle may develop rich reflection of its own; the harder question becomes subjective consciousness, which the speakers defer to another conversation.

10. Virtual experts scale knowledge but cannot duplicate relationship

Peter expects more AI-driven Nobel Prizes, citing AlphaFold, MatterGen-style materials and medicine. He relays Anthropic CEO Dario Amodei’s forecast of a century of biological progress within five years, potentially doubling human lifespan, then adds his own two-year prediction for grand unification, dark matter, dark energy and the universe’s origins.
His capability curve moves from a cited o1 IQ of 120 to possible Grok 3 scores of 140–150 and eventually 200, on a nonlinear scale. Salim expects locally instantiated models, supplemented by Gemini, ChatGPT and Claude and given a video face, to pass that kind of Turing test and operate like coherent individuals.
Peter plans an identical AI version trained on his books, podcasts, reactions and experience, then imagines dispatching 1,000 versions to conferences and negotiations this year. He recalls Eric Schmidt projecting that within two or three years the world’s best theoretical physicist could sit beside every graduate student and laboratory.
Mo’s objection is philosophical: copying Peter’s current knowledge and persona would mean dumbing the AI down. Intelligence will become “a plug in the wall,” so humans should delegate analysis and presentation while expanding authentic presence. An avatar cannot reproduce “the same hug,” shared memories or the trust accumulated through an actual relationship.

11. Power will both concentrate and democratize

Mo traces leverage from the hunter’s extra food through farmland, factories and information technology: better automation expands one person’s attainable surplus from tribal favor to estates, millionaire status and then billionaire status. AI extends that curve toward trillionaires, dominant platforms, autonomous armies and nations controlling the most consequential industrial intelligence.
Yet DeepSeek-like availability democratizes power simultaneously. Individuals gain cheap access to synthetic biology, software and drones capable of targeting a particular person. Mo expects the collision between extreme concentration and mass capability to produce surveillance, loss of freedom, bank-account restrictions and UBI used as control, followed by resistance and further oppression.
Salim agrees on the danger but sees rapid democratization supplying defenses as well. He describes the Russia-Ukraine war as being prosecuted with roughly half a million drones — mostly drones fighting drones rather than people — and notes that systems already exist to defend stadiums against drone attacks. The arms race can mitigate threats, but only after dangerous capabilities appear.
His nearer-term specimen is a voice recording purporting to be a kidnapped daughter and demanding Bitcoin. The recipient cannot know whether it is real. The ingenuity is effectively unbounded: he recalls robbers who advertised construction work so 800 identically dressed applicants surrounded the bank, letting the actual gang vanish into the crowd. AI lowers the cost of such asymmetric creativity.

12. Jobs may expand temporarily, but adaptation cannot wait

Peter cites Mark Benioff saying Agentforce 2 raised engineering productivity 30%, removing the need for additional hires, and Sam Altman predicting AI will become the number-one programmer by year-end. Customer service, HR, sales and software are therefore early tests of whether higher output offsets eliminated positions.
Salim’s optimistic case is historical: Sweden, South Korea and Germany are cited for high robotics penetration, while Singapore, South Korea and Germany are cited for low unemployment. Software demand could require 100 times more code, while trucking companies already struggle to recruit. He expects augmentation and new work after a short-term blip, but doubts medieval public institutions can execute a transition such as UBI.
Mo’s dissent centers on suffering during that blip. People working two or three jobs cannot always reskill easily, and he says he thinks 80% of the code written last year was machine-written. Rather than manufacture replacement jobs, he wants systems that let people live without working 60–80 hours: “We work to live rather than we live to work.”
Mo invokes E. O. Wilson’s formulation of the institutional mismatch: “Our emotions are Paleolithic, our institutions are medieval, and our technology is godlike.” Facing a “perfect storm” across geopolitics, economics, employment and technology, people should expand their denominator: use AI now, understand how it works, build human skills and retrain before displacement arrives.
His Unstressable model defines pressure as total challenges divided by the cross-section of skills, relationships and resources. Mo applies the prescription to authorship: he no longer assumes a human can think through a topic better than AI, so Alive will appear first on Substack as a public dialogue, with readers correcting him and with the AI openly debating rather than presenting its answers as his. Peter’s closing constraint remains: “There’s no on-off switch” or velocity knob — only agency to steer.

George Bonaci, VP of Growth @Ramp: How Ramp Became the Fastest Growing SaaS Company Ever |E1264 Episode 132 - February 21, 2025