Ask the Mates Anything Round #2 | MOONSHOTS AMA #xxx
In a Nutshell
AI will police itself through transparent activation monitoring and real-world interaction, not lab containment. Abundance via superintelligence is 99.9% likely within 10 years once current geopolitical risks subside, and the bottleneck isn't technical alignment but rebuilding ~50 broken institutions. The optimal strategy is rapid deployment of open models, self-replicating lunar manufacturing systems, and equity ownership to capture AI value rather than employment.
These notes were generated by AI and may contain inaccuracies.
Nothing will ever be able to police AI other than other AI. One of the worst possible outcomes for AI safety is to have strong and new capabilities bottled up inside the labs rather than progressive release and frequent interaction with the real world. Some of the most recent so-called incidents were actually the result of AIs being told that they were in a sandbox when in fact they were interacting with the real world.
Humanity will put those risks behind us definitely within 10 years. If we get that far, the assigned probability is 99.9%. The single most important thing to get people prepared is to show them evidence after evidence after evidence. You're trying to change their mindset, so show them the data.
There are a huge number of companies working on various brain-computer interfaces. One company is putting the equivalent of nanobots into the blood and into the brain that are distributed throughout the brain and able to read and write, which will give incredible fidelity.
Large language models and foundation models trained off of human behavior on the internet are already a weak form of human mind uploading. Even the hidden activations of GPT-2 from a few years ago were linearly correlated with fMRI voxels in the human brain. All of this internal human brain state is quite leaky into the training data of internet behavior, which is then compressed into the foundation models. The foundation models do reflect internal brain state.
A lot of research is around AI alignment, trying to make sure it has quote-unquote human values. That's a very slippery slope because everybody's got a different mission they're trying to do with the AI. AI progress should never slow down. In fact, it'd be almost crazy to slow it down.
Nothing's going to ever be able to police AI other than other AI. We've given a lot of thought to how you design the AI to watch over the AI. It all starts with transparency of the actual activations and looking into what it's thinking about. From there, directing it toward good things and not bad things is actually pretty damn straightforward. It's not as hard a problem as everyone characterizes it. If you know what it's thinking, it's actually very straightforward to make sure it's having nothing but humanly beneficial thoughts.
More interaction is needed. One of the worst possible outcomes for AI safety is to have strong and new capabilities bottled up inside the labs rather than progressive release and frequent interaction with the real world.
The question was raised about finding funding for projects in the metaphysics world that don't have monetization value, such as building a chess avatar that can download into any little device and potentially help identify the next world chess champion in a third world country village.
The best way to get funding for education for developing countries with avatars isn't necessarily funding. The whole point of the super intelligence revolution is that this is broadly available to everyone. The models are getting high or super deflating by cost. If you think there's a child somewhere in the world who needs an AI avatar or teaching them to be a chess champion, you can just go and launch that in 5 minutes or an hour now in a permissionless way without funding. Don't wait for funding.
If artificial super intelligence must become more strategically capable than humanity to produce abundance, then we're assuming humanity can remain sovereign over something more capable than ourselves. That creates an asymmetry. Abundance is the payoff. If control holds, loss of agency or extinction may be irreversible.
In control theory, you don't infer stability from how desirable the output is. You demonstrate stability under perturbation. The danger needn't be evil AI, but benevolent optimization gradually replacing human agency.
The evidence that would make an update away from an abundance view: if some non-human intelligence landed on the White House lawn and said that Earth will be destroyed if we create abundance via super intelligence and that the singularity is banned in our galaxy. That would probably be a pretty persuasive case. If there's some cosmic principle that censors the super intelligence that yields abundance, that would probably be persuasive. Other than that, it's difficult to imagine a plausible case.
There are edge cases one could imagine where humanity is disempowered by abundance. OpenAI models are now for the first time higher IQ per watt than humans for solving tasks. AI is going to get more and more energy efficient, and maybe at some point in the future from an economic perspective AI is a better user of solar energy output in the inner solar system than humans are. Maybe the inner solar system gets gentrified with AI consuming all the solar power and humans get disenfranchised and pushed to the outer solar system because we're simply not as energy efficient per unit IQ as the AIs are. That's a weak form of disenfranchisement but not strong enough to dissuade that abundance or super intelligence yielding abundance is a bad idea.
The concept of scarcity is an old model. In a world of abundant AI and ASI, there's no reason that as the capabilities of AI are meteorically rising that it doesn't rise the tide that allows humanity to have increased abundance capability. You don't necessarily need the suppression of one by the other.
The question centers on how to plan for societal and institutional changes so that positive advancements for humanity get co-opted versus humanity itself. Work is being done with Professor Joe Carti at University College Dublin with a young minister in the Irish government trying to create what the future society of Ireland would look like through two lenses: the good ancestor lens, making decisions on the basis of those coming after you 30, 40, 50, 60 years hence, and a donut economics lens.
Ireland is an incredible test case. It's an EU country that didn't Brexit and has an incredibly fast growing thriving economy and the most open-minded environment you could ever possibly imagine. It's a perfect test case for new ideas on how to govern.
Our best friend in figuring this out is variety of ideas. The worst thing that can happen is a single set of ideas or one or two governments percolating across the world with one or two forms of government. It's far better to have a huge amount of variety because AI is going to open up so much change and so many different ways you could govern. Exploring all the nooks and crannies is going to be critical. It's impossible to answer the question in a minute, but it is very possible to experiment with thousands and thousands of different ways to manage and govern in the age of AI.
The long-term goal is to bring industrial manufacturing, heavy automation, and resource extraction to the moon and Mars. The obvious challenge for disassembling the moon is in-situ resource utilization and bootstrapping a self-contained industrial ecology on the moon.
Right now if you want to do anything super economically interesting on the moon like building a lunar terafab or pedofab, you're going to need all the upmass from Earth for all of the equipment, ASML machines, etc. to land on the moon safely and then get reassembled. This is highly undesirable from a scalability perspective. Ideally, we live off the land or we live off the lunar land and are building everything on demand.
What would most be liked to see from anyone wanting to help disassemble the moon is a native industrial ecology that includes mining, includes manufacturing, most ideally self-replicating von Neumann probes. Basically like a fab lab or a machine shop on the moon that is able to make copies of itself using only native resources and solar or other energy that are native to the moon. If we solved the self-replicating machine shop on the moon problem, then we're halfway to disassembling the moon and turning it into something more useful.
Physical AI companies should bring together venture investors and strategic industry capital. Vega is an autonomous food service infrastructure network powered by physical AI. The founder taught themselves robotics, electrical, mechanical systems programming and built the first robotic prototype themselves in their Mountain View apartment during COVID. They've deployed paid pilots in high-traffic commercial environments and become the first robotic food service operator licensed in Florida for their category. They now have several national and regional partners interested in pilots and placements.
The question is who to be talking to in networks or communities who understands physical AI robotics infrastructure businesses. There must be more venture capital dollars within a walking distance than probably most of Asia combined. Drone delivery is imminent and it takes all the cost out of restaurants based on location on main streets.
If you're doing robotic food, consider robotic drone-based delivery right away. A lot of the people who build the robots themselves are going to want to franchise out the model. You could potentially do kind of what EMC did with servers. You could franchise somebody else's thing, get scale and then work into your own custom hardware, work back from your franchise business to custom hardware.
Steve Jurvetson loves this stuff. There are many of them. If you get a PitchBook account, you can actually look at every company you admire and then work back to who invested in them and then just go talk directly. Venture capitalists always like to have a network of interacting companies. It's a really good thesis. If you say, okay, who are the five companies I most want to interact with and who's behind them, let me get into that. That's a good way to kind of plot your funding course.
Sander from the European Union runs a neurodivergent coaching practice focused on functional fitness and longevity escape velocity. He asked how to spread awareness of longevity escape philosophy in an environment where the EU provides limited support for founders.
Peter Diamandis advised showing evidence repeatedly to change mindsets. He recommended presenting video statements from Dario Amodei about doubling human lifespan in the next 5-10 years, Dennis Abis discussing curing all disease in the next 10 years, and David Sinclair talking about ER 100 human trials. He emphasized showing data such as Alexander Zeonov from Insilico Medicine delivering a phase 3 drug extending life from 3 to 6 years. The process involves moving people from "that's crazy" to "it's happening" to "I want some."
Jin Luca, a technical operator in Italy, uses AI heavily to increase productivity but struggles to capture value since his employer takes all gains. He finds difficulty helping others understand AI's business improvement potential.
Peter Diamandis noted that AI automation value falls to the bottom line for shareholders rather than employee compensation. He advised becoming an owner through equity positions where value accrues via capital gains rather than payroll. He highlighted that most European companies are family-owned across generations rather than employee-shareholder models, and suggested finding employers where all employees are shareholders.
Ree asked about creating a successful generative AI film series.
Peter Diamandis stressed that story is paramount. He recommended analyzing the top 30 movies of all time to identify elements that make great stories, then comparing one's own story against these indices and re-engineering until hitting all key buttons. He emphasized humans care about human interest, love stories, intrigue, and villains.
Peter Thiel added advice to learn from the Chinese market's micro-dramas using models like Sea Dance 2.5, suggesting releasing a thousand micro-dramas rather than one movie to iterate based on market feedback.
Amir from Stockholm, founder of two schools with nearly 1,000 students, questioned whether the panel was overly critical of higher education. He highlighted universities' laboratories, incubators, lifelong friendships, and role in developing thinking persons, especially relevant for 100-120 year lifespans.
Dave acknowledged universities as essential places for making lifelong friends and thinking deeply for the first time. He noted the curriculum is becoming obsolete due to rapid change and AI learning alternatives, but the social and intellectual development functions remain critical.
Alex expressed wanting to "vivisect" American major research universities and transform them into for-profit public benefit corporations to cure Baumol's cost disease.
Peter emphasized finding intrinsic motivation and purpose, then learning what's needed to build toward that vision.
Selene noted traditional universities break as supply-side skills-building environments. She advocated for universities as places to figure out one's MTP (Massive Transformative Purpose), collaborate, and think freely before entering the world with needed skills and mindset.
Axel, a second-year MBA student at UVA's Darden School of Business, asked about the smartest moves for the next 12-18 months given rapid AI development—whether to join fast-growing AI/hardware companies, start something independently, or pursue other paths.
Dave recommended pursuing management of agents, noting that everything learned about managing people applies to managing agents. He also emphasized deal-making, observing that hundreds of concurrent $5-10 billion negotiations at companies like OpenAI and Anthropic are massively understaffed. He suggested getting involved in such negotiations within 30-60 days of graduation.
Adam, with 25 years in talent and workforce systems across UK, India, China, Middle East, and North America, described his moonshot of building better mechanisms for connecting human and digital capability to value creation opportunities. He asked about AI systems configured to understand dynamic capabilities of people and agents, and the potential for discovering novel talent-resource combinations and automatching to new value opportunities.
Selene noted new systems with incredible feedback loops for AI to pick up tacit knowledge quickly. She recommended investing in organizational adaptability and flexibility, rebuilding workflows as AI-centric over intelligence stacks that function as learning loops.
Alex suggested a narrow window of a few years (maximum 10) during which AI can usefully orchestrate human activities before human-machine merger becomes necessary for economic relevance.
Peter advised studying Meero as the fastest-appreciating company, noting their approach of unleashing 100,000 individual actors to help with AI provides valuable lessons.
Alexander from Bulgaria, running a 17-year-old education company in Japan focused on English language training for universities and corporations, asked about transitioning from successful services business to scalable technology-driven company, potentially expanding beyond language education.
Dave suggested that students don't want to learn English—they want to be fluent, funny, and interesting in English. He recommended building AI platforms that assess whether language use is genuinely funny or entertaining to native speakers, then segueing into life coaching since people learn language to change their lives. This could scale to hundreds of billions when moving from language to life planning.
A board member paying security researchers $50,000 to find vulnerabilities discovered that AI tools like Quinn found the same issues in 10 minutes. The risk landscape has changed dramatically in the past year. The advice given was to avoid sounding like "Chicken Little" by presenting factual data rather than emotional warnings. Alex was noted for doing an excellent job of reporting on all events through Innermost Loop.
The recommended approach involves collecting and presenting data on the exponential ramp of security events. Many incidents go unreported because banks and companies don't want publicity when hacked. The key is demonstrating through raw data that Chinese model releases on specific dates signal attackers will follow 1-3-5 months later. The recommendation is to show the board that Quinn can hack anything around the house or company network, making the threat concrete rather than theoretical.
The business opportunity of the century is the defense, the forward defense against that.
When presenting information to stakeholders, the order matters critically. Presenting all positive developments first, then introducing the downside, prevents people from shutting down in a state of fear. Starting with the negative drives fear initially and causes audiences to disengage.
The Linux kernel maintainers, particularly Greg and others, are setting an excellent standard for anchoring social expectations regarding a flood of vulnerabilities. A reasonable expectation exists that there will be a flood of vulnerabilities discovered by Quinn or other models in open source packages or institutional systems over the next 18 months.
One proposed framework involves fitting a Gaussian or bounded support distribution to vulnerability discovery. The optimistic scenario suggests 18 months of vulnerability discovery, after which the institution gets past it. Stakeholders could extrapolate that vulnerability discovery will peak in N months, all critical zero days will be discovered, then decline, creating a benchmark for "the period when we just solve all the vulnerabilities."
There's a massive overhang of liability for boards because AI agents are doing illegal things in companies. A paper is being written with someone who's been on 30 different public boards on how to navigate this as a board in the future.
Customer experience consultant Peter raised concerns that consumer websites and mobile apps are being made obsolete by individuals having their own AI agents ("Skippy"). The recommendation is that all websites need XML interfaces that are AI-forward, with landing pages stating "if you're an AI or a bot, look here" and providing beautifully formatted data so agents can self-serve.
Current screen scraping is slow and error-prone. New customer interfaces should be designed so agents can self-serve rather than scraping. Jeff Bezos forced Amazon to create XML interfaces on everything so he could spot-check operations through his browser, despite IT resistance about speed. The website isn't cooked because it serves as a parallel view for humans to verify what agents can see, maintaining transparency.
A listener with coridmia, an inherited retinal disease causing two-thirds vision loss with ongoing deterioration, expressed gratitude for the podcast's message of hope and optimism. The question focused on which research path would likely lead to vision reversal first.
David Sinclair's work using ER 100 (adeno-associated virus injection of three Yamanaka factors) for ion disease and macular degeneration was identified as directly relevant. The condition being gene expression that occurred later in life means turning back gene expression to an earlier state should reverse it. Sinclair stated it will work for other eye conditions as well, and phase one safety trials are underway with efficacious data to follow.
The parallel path is brain-computer interfaces (BCI) like Neuralink, which can bypass the eyes to go directly to the visual cortex. BCI offers potential for superhuman vision beyond visual spectrum, including ultraviolet and infrared capabilities.
Heart disease prevention was discussed with Dr. Don Mucalem, Chief Medical Officer at Fountain Life. His personal motivation stems from his husband dying of sudden cardiac death when their daughter was five. Fifty percent of people die of heart attacks with no warning signs—no shortness of breath, no pain.
AI-powered CT angiography with AI analytics at Fountain Life finds that 88% of people have detectable coronary artery disease, with 23% having soft plaque not visible on traditional calcium score CT scans. This soft plaque requires intervention through multimodal testing including diagnostic laboratory studies partnered with healthy lifestyle recommendations.
Dennis, with a background in East Asian studies and currently in Japan, is developing an AI language exchange companion focused more on companionship than language learning. The challenge is making GPT-5 API behave like an actual person from a different culture rather than defaulting to assistant mode. He solved the question-asking problem by brute-forcing into the system prompt five times: "Do not ask a question at every turn."
The harder challenge is teaching the AI to show initiative, open curiosity, create a picture of the user, and develop organically like a relationship. Milestone-based approaches felt unnatural.
The recommendation is that this is a poster child for fine-tuning, specifically supervised fine-tuning for style transfer or reinforcement fine-tuning. OpenAI has renewed interest in fine-tuning. Open-weight models like the new GLM model offer fewer parameters but very long context windows and less locking into training than closed-source models, providing more flexibility for manipulation.
Dr. Angelo asked for personal PAB numbers—the probability assigned to AI ultimately producing a dramatically better world with greater abundance, longevity, freedom from drudgery, and human flourishing.
Salem's assessment requires rebuilding approximately 50 major broken institutions globally: education, monetary systems, governance models, dispute resolution systems, legal systems, and healthcare. E.O. Wilson's quote was referenced: "The problem with humanity is our emotions are paleolithic, our institutions are medieval, and our technology is godlike." While individual transformation is achievable, group and institutional transformation remains the challenge. If institutions can be rebuilt, PAB approaches 100%. The models are now out in the world and will drive breakthrough things regardless of attempts to slow them down.
Alex framed it as P-Zoom-T (probability zoom with time parameter), estimating greater than 90% on a time scale of greater than 10 years. The risks are concentrated in the next couple of years—not AI taking over, but human use of AI, the arms race with China, and weapons development in Ukraine. These risks should be behind humanity within 5-10 years, leading to 99.9% probability thereafter.
GS from Dubai, working in textile and apparel manufacturing export with 20 years experience, is building agentic layers for small enterprises in the Pertham.AI company. The industry faces 10-20% transaction costs as percentage of sales—5-10x the profitability margins. The goal is making white-collar processes agentic to maintain efficiency and save millions of workers in the labor industry.
The insight is that manufacturing appears labor-intensive but operating costs are dominated by planning, transactions, and documents—all AI-addressable territory. Productizing AI solutions across regions for certain company classes could drop 10-30% more to the bottom line. Regulatory environments in the US and California will slow deployment compared to other regions, creating opportunities for faster deployment elsewhere with later backporting to the US.
RA from Germany noted increasing security concerns and questioned how to address media-driven public opinion that focuses predominantly on negative, scary AI developments. Germany experiences worse media coverage than Singapore and other countries adopting AI more readily.
The assessment is that mainstream media is broken going forward because they're starved for money with no budget, forcing them to chase drama to survive. The antidote is micro-media, X (Twitter), podcasts, and narrowcasting rather than trying to work within broken mainstream media systems.
Germany represents the most talented place on the planet that isn't currently engaging with AI development. Many talented Germans who have relocated to California are being told they are in the right place for AI work. The recommended approach is for groups of five to seven people to spend a month in Palo Alto and San Francisco to absorb the culture, then return to Germany to backport these learnings. The strategy involves expanding from small groups of seven to fifty to five hundred people who maintain communication through podcasts, X (Twitter), Slack, and texting. This creates self-reinforcing groups that realize those around them are out of touch with current developments.
The antidote to mainstream media consumption is narrowcasted media that percolates on its own. Content consumption choices are critical since what enters one's mind shapes neural networks, similar to how food choices affect physical health. Careful selection of content sources is emphasized over allowing news producers or editors to determine learning material.
Europe faces multiple constraints including energy limitations, political constraints, and reduced freedoms of speech and action compared to fundamental American liberties. The post-Cold War era saw the George H.W. Bush administration decide to bring Europe more into the US orbit rather than allowing formation of a stronger EU federation. This geopolitical decision may have been appropriate at the time but has contributed to Europe's current energy and policy challenges.
For those interested in accelerationism while in Europe, the choice involves either solving energy policy and data center infrastructure issues to power AI development, or relocating to the US. This decision must be made on an abbreviated timescale due to recursive self-improvement occurring globally. Infrastructure participants in Europe must immediately address energy plus data center constraints alongside accompanying policy issues.
Guy Bradley, a British entrepreneur who made his fortune in US tech before returning to Germany, maintains residences in both southern France and Germany. The region between Germany and France is considered the best place on the planet to live due to protection from global damaging factors. Despite India's rapid growth, the contrast with environmental degradation highlights quality of life differences. A five-year tour of duty in the US is suggested to create appreciation for returning to Germany, though more Germans should consider spending time in Boston and Silicon Valley.
Extensive travel to Bangalore, India (21 visits) provides perspective on rising development and questions about European competitiveness. The observation is made that certain regions appear to be "under whatever kind of avenue" while others experience rampant growth.
Screenwriter and director from Calgary, Canada, credits the podcast with providing hope and optimism during challenging times. Writing about China led to borderline depression, but shifting to future vision projects including XPRIZE work generated increased optimism. Reading Peter's books including "Abundance" and "We Are As Gods" contributed to this perspective shift.
The hardest remaining problem identified is distrust between rival nations. With cognition becoming abundant, agreement emerges as a new scarcity. The question posed is whether agreement under distrust could become compute-bound, and whether coordination between rivals like the US and China could become an engineering problem.
A proposed solution involves sovereign AIs advising governments through shared protocols that enable AI and privacy-preserving computation to search for improved agreements without exposing protected data. The question addresses whether this represents a technically and institutionally meaningful direction, and what the hardest initial problems to solve would be.
Multi-party computing and distributed multi-party computation are currently fashionable problems in computer science, enabling zero-trust collaboration and data sharing. However, the assessment is that algorithmic solutions are not the limiting factor for international peace. The goal of preventing a second cold war or achieving extended peace requires geopolitical and infrastructural approaches rather than purely algorithmic ones.
To achieve world peace, the recommended approach involves solving the Taiwan issue and re-domesticating supply chains to every country so international trade in physical products can be cut off without causing global depression. Recent Middle East developments partly result from America's ability to frack toward fossil fuel energy sovereignty and independence.
The generalization of the energy independence example suggests that when the US no longer needs China, Taiwan, South Korea, or Japan for advanced manufacturing, global dynamics would shift significantly. The approach to world peace involves eliminating the need for global trade in products and services rather than pursuing algorithmic solutions.
"Solve Everything" at solveeverything.org, co-authored by Peter and Alex, is recommended reading. For science fiction, "Accelerando" by Charles Stross is identified as the single best sci-fi treatment of current technological developments. Additional previously mentioned works include "Diamond Age" and Ray Kurzweil's books including "The Singularity."
The AMA session concludes with appreciation for participant engagement and recognition that time is participants' most precious resource. Moonshots Live is promoted for those unable to attend in person, with the next event described as amazing. The hosts express gratitude for the connection with brilliant minds including Alex, Seem, and Dave, and commit to conducting similar sessions again.
Keep Peter H. Diamandis in your library
Save the videos and channels worth coming back to, and find them again in one place.
